The first family lives in a neighborhood where the zoned public school has mediocre ratings and a reputation that keeps most of their neighbors up at night. They send their two kids to a private school at roughly $7,800 per child. The mother works a second job on weekends to cover it. Ask her what she would cut if money got tight and she doesn’t even blink. Tuition is the last line she would touch. Private school, for her, is absolutely essential.

The second family earns the same income, lives in the same metro area, and has the same number of kids. Three years ago, they moved specifically because the suburb they targeted has one of the top-rated public-school districts in the state. Their kids walk to a public school that sends graduates to top universities every year. Ask this mother whether private school is essential and the question doesn’t even register. Private school, for her, is irrelevant.

Same line item. Same income. Same city. One household treats it as non-negotiable. The other doesn’t even think about it.

The Tomato-Tomahto of the Household Budget

This is the heart of what new PYMNTS Intelligence data shows.

What makes something feel non-negotiable is almost never about what it costs in absolute dollars. Essential isn’t a characteristic of the expense. It’s the characteristic of the person spending the money on it.

A PYMNTS Intelligence survey of more than 3,400 consumers put the same 22 line items in front of everyone and asked them to classify each one as absolutely essential, necessary but a choice or purely discretionary. The results don’t line up by income. They line up by life stage, family structure and the commitments each household has locked in over the years.

Take private school. It’s rated absolutely essential by 17%, necessary but a choice by 23%, and purely discretionary by 60%. Family financial support splits 26/34/40.

Then there’s grocery delivery. Half of consumers earning less than $50,000 say it’s essential or necessary, versus 42% of those earning $150,000 or more. For a household earning $45,000, getting groceries delivered might be a logistics requirement rather than a premium convenience if the consumer doesn’t own a car or works irregular hours or multiple jobs and can’t easily get to a store.

Compare all of that to insurance, rated essential by 65% of consumers with only 11% calling it discretionary.

Nobody really disagrees about insurance. The items where consumers are closer together than apart are universal, structural necessities. The items where consumers split are the ones where essential means whatever the specific household at the specific life stage decides it means.

Read More: The Three Blind Spots in How Consumer Sentiment Is Measured

 A Better Way to Think About Essential

Source: PYMNTS Intelligence framework, New Reality Check: The Paycheck-to-Paycheck Report, January 2026.

The traditional way finance professionals sort spending is into fixed, variable and discretionary buckets. The flaw with that classification is that it isn’t always correlated with how consumers manage their money.

A better frame has two dimensions:

• Why the expense feels essential
• How locked in it is

Structurally-Locked expenses are forced by circumstances that feel immovable. Childcare for a working parent. A mortgage. A car in a sprawling metro area. Student debt from a decision made a decade ago. PYMNTS Intelligence data finds that more than half (54%) of the full sample rates mortgage/housing absolutely essential, half rate car ownership essential, and 37% of married parents rate childcare essential.

Values-Locked expenses are conscious priorities that have hardened into a financial commitment. Private school tuition. A gym for pain management. Monthly support to an aging parent. The households rating these as essential aren’t wealthier than the ones that don’t. They’re different because of what they’ve decided matters, and is therefore essential.

Circumstantial Recurring expenses are related to needs shaped by a personal situation rather than universal necessity. They can be changed or eliminated, but not easily and not without other potential financial consequences. Accordingly, 65% of consumers rate insurance as essential and 58% of consumers rate healthcare as essential; these expenses dominate this quadrant and the overall rankings.

Lifestyle Choices are the only expenses that map cleanly to the conventional discretionary bucket. Subscriptions (14% of consumers say they’re essential), travel (13%), entertainment (10%) and expedited shipping (10%) are nice-to-haves but could be dialed back or eliminated, depending.

When expenses are put into this framework, the same items land in different buckets. A gym membership lives in Lifestyle Choices for one household and in Values-Locked for another. Childcare is Structurally-Locked for a 34-year-old working parent and irrelevant to a 68-year-old retiree. Essential becomes a subjective call, not an objective one.

How Life Stage Rewrites the Priority List

The clearest way to see the many sides of “essential” at work is to walk through the U.S. consumer base segment by segment. Each one tells a different story about what gets pulled into the non-negotiable tier and why.

Let’s start with millennials, the generation living through the years when earlier commitments show up as non-negotiable monthly line items. Nearly seven in ten, 69%, say they live paycheck to paycheck. Half point to long-term life decisions as the reason, a rate 19 points higher than boomers and the steepest of any generation.

Read More: 38% of Millennials Pay Out of Pocket for Healthcare

Their non-negotiable tier reads like a structural inventory. Insurance, healthcare, the car, the mortgage, outside family support. What sets millennials apart is the tier just below.

More than a third (37%) cite childcare as essential.  Grocery delivery at 31%, because when both parents work and a toddler is in the car seat in the back, the friction associated with a trip to the grocery store gets real. Student loans at 27%. Where they live at 23%. None of those come across as lifestyle preferences. They’re commitments made years ago that can’t easily be unwound. That explains why only 54% of millennials feel in control of their financial situation even as they stay stuck inside it.

Married parents are the most obvious case of how Structural Lock-in operates. Sixty-seven percent live paycheck to paycheck. Sixty percent say they could change their situation with effort. Their budget says otherwise.

Married parents cite long-term life decisions as the reason for their financial situation at 59%, the highest rate of any household segment in the survey. Their top essentials look like everyone else’s: insurance, healthcare, mortgage. The difference shows up in the next tier.

Childcare and clothing at 37%. Grocery delivery at 30%. Private school at 29%, nearly double the 17% full-sample rate. Where they live at 25%. Every one of those categories runs at least eight points above the full sample, and none of them are stated preferences. They’re what it takes to run a household with kids. A second car to cover two drop-off routes. A neighborhood chosen for its middle school. Groceries delivered because nobody has time to shop on the way home. That’s why the married-parent budget is the most locked-in profile in this study.

Single parents show what happens when the same financial scaffolding has to stand on one income. More than eight in ten (82%) live paycheck to paycheck, the highest rate of any segment profiled. Only three in ten report any flexibility to cut. Clothing jumps to 40% essential, eleven points above the full sample, because a single income is feeding and clothing kids who outgrow their shoes every four months. Grocery delivery hits 34%, twelve points above the full sample. Family financial support, 39%. Private school, 28%. Childcare, 30%.

The most telling insight isn’t any single line item. It’s the pattern across the three reasons people give for their financial situation. Day-to-day spending, long-term decisions and unexpected events all rank within five points of each other. For most households, one of those three clearly dominates. For single parents, all three hit at once. There is no single lever to pull when things get tight because the budget is pressured from every direction.

The paycheck-to-paycheck struggling household closes the segment story in a way that looks paradoxical at first. Nearly half (48%) report little or no perceived control over their finances. Forty-three percent say spending changes alone can’t fix it. What drives their budget isn’t a pattern of everyday choices but shocks they didn’t plan for. Sixty-one percent point to unexpected events as the reason for their financial circumstances, compared with 31% of non-paycheck-to-paycheck consumers.

But here’s the counterintuitive part. Their essential ratings are lower across the board than other segments. Insurance at 55%. Healthcare at 50%. Mortgage at 46%. Car at 44%. This isn’t because they care less. It’s because they’ve already cut everything that could be cut. What remains is non-negotiable, and there’s nothing more left to trim.

Read More: How 30 Million Workers Borrow from Tomorrow to Pay for Today

Generation layers on top of all this. The sharpest divide between age groups isn’t on insurance or housing, where consensus about essentials is broad. It’s on the small daily line items that older generations treat as obviously discretionary.

Gen Z and millennials are up to ten times more likely than boomers to classify coffee, lunch out, food delivery, subscriptions and gym memberships as essential. For a 24-year-old with a two-hour commute, a food delivery subscription is about logistics and convenience. For a 72-year-old retiree with a fully stocked pantry, it’s an obvious waste. Neither is wrong. They’re describing different lives with different priorities.

Read More: Healthcare on Hold: Why 1 in 4 Gen Z Consumers Skip the Doctor

Parenthood is the single variable that reshapes the priority list more than any other. Married and single parents rate childcare, private school, family support, grocery delivery, food delivery, clothing and meal kits dramatically higher than adults without children. Childcare is obvious. The less obvious ones tell the real story. Meal kits, food delivery, grocery delivery. For parents, these aren’t indulgences. They’re how the household runs.

Read More: How Time Became the Next Great Asset Class

Grocery delivery sits at 22% essential in the full sample, 30% for married parents, 34% for single parents. Food delivery shows the same logic from a different angle. Thirty-four percent of households earning under $50,000 rate it essential or necessary, compared with 31% of households earning over $150,000. A worker juggling two hourly jobs treats the $12 delivery fee as the cost of eating the meal that has to happen between shifts. For a higher earner with a predictable schedule, the same line item is convenience.

The clearest signal in the entire dataset is clothing for single parents. Forty percent call it essential and another 46% call it necessary, putting 86% at or above necessary. Only 14% call clothing purely discretionary. For most consumers, clothing is lifestyle. For a single parent with growing children, it is regarded more like the utility bill.

Financial lifestyle adds a final twist to the picture. Nearly a quarter of paycheck-to-paycheck consumers (23%) earn $100,000 or more. Not because they spend carelessly, but because of commitments already locked in. The mortgage on a house in a good school district. Childcare for two kids. Student loans still running a decade after graduation. A six-figure salary doesn’t unwind any of that.

Read More: Who Is the Paycheck-to-Paycheck Consumer in America?

Consumers who aren’t paycheck to paycheck actually rate structural items like insurance, healthcare and mortgage as more essential than consumers who are struggling. You would expect the opposite. The reason shows up in what each group blames. Two-thirds (66%) of non paycheck-to-paycheck consumers point to day-to-day spending as the main driver of their financial situation. They frame it that way when the structural bills feel handled. For these households, their control levers live in the daily choices, not the locked-in commitments.

Struggling consumers tell a different story. They’re twice as likely as non-paycheck to paycheck consumers to cite unexpected events as the cause of their financial situation (61% versus 31%). Forty percent report low or no flexibility to cut expenses, compared with 18% of non-paycheck-to-paycheck  consumers. That 22-point flexibility gap is the single widest in the dataset, and it is structural that becomes behavioral.

Read More: Meet the 27 Million Americans Who Drive 8% of Consumer Spend but Struggle to Pay Their Bills

Pulling It All Together

The picture that emerges across these segments is that essential is a characteristic of the person spending, not of the expense itself.

What actually predicts how a household will behave under financial pressure is the combination of life stage, family structure and financial history. Millennials cite long-term decisions at 50%. Married parents at 59%. Boomers at 31%. That 28-point spread between younger or parenting households and boomers is about which commitments are actively running through the budget. And how willing those households are to protect the priorities behind them.

Segmenting customers by income decile or FICO band captures none of that. Segmenting by priority profile captures all of it.

The Priority Behind the Payment

Every payments company, credit issuer and bank has built its data stack around two questions. What did the consumer buy, and how much did they spend? The PYMNTS Intelligence data in this report says those questions answer less than half of what matters.

What a consumer buys is visible in the transaction record. Why they bought it, whether it is a conscious priority or a forced one, and whether they would fight to protect it under financial pressure, isn’t.

The next competitive edge in payments and financial services isn’t more behavioral data. It’s priority data.

Consider two $7,800 annual tuition payments sitting in two different customer profiles. Same category, same frequency, same payment amount. In the transaction record, they’re indistinguishable. In reality, they’re three different customers. For the 17% of households that rate private school absolutely essential, that $7,800 is sacred. Those consumers will go into debt to protect it. For the 23% who call it necessary but a choice, it’s up for review the minute cash flow tightens. For the 60% who call it discretionary, and who happen to be paying the tuition because a grandmom is paying, it’s the first thing to go. Transaction data alone can’t tell them apart.

Read More: Why the Offers Economy Is Broken

Priority data gives a view of who the customer is, what they have committed to and what they will trade off to protect those commitments. It predicts the next move in a way the transaction record can’t. The implications play out differently for different parts of the ecosystem.

For credit issuers, priority data answers the single most valuable question in the business. If this consumer’s cash flow tightens, which bills get paid and which may not? Forty-three percent of paycheck-to-paycheck consumers who are struggling say spending changes alone cannot fix their situation. They’ll miss a payment on something.

Priority data tells the issuer which something.

Read More: Your Business Has Its Payments Data. Now What?

For merchants, the lesson is that product category isn’t destiny. The same subscription service splits 14% essential, 32% necessary, and 54% discretionary across the full sample, but 17% essential and 43% necessary among single parents. That’s one product and two completely different retention fights.

The Values-Locked customer will swallow a price increase. The Lifestyle Choice customer will cancel the moment a competitor runs a promotion. The same treatment for both leaks revenue at both ends. Dynamic pricing, loyalty programs and churn prevention all need to follow where the category sits in the customer’s priority stack, not the category code.

For Buy, Now Pay Later and installment lenders, priority data points to an opportunity much larger than discretionary retail. Thirty-seven percent of married parents call childcare absolutely essential. Twenty-nine percent call private school essential. Forty-one percent call family financial support essential. These are the recurring, high-ticket, Values-Locked line items that households currently put on credit cards, take from savings, or borrow from family to cover. Pay later products built for those priorities capture a segment the retail-focused BNPL players aren’t actively addressing. The underwriting case is stronger, too, because households don’t default on what they’ve decided matters most.

For banks and personal financial management tools, the implication is that sorting spending by category is a map drawn against how consumers actually think. A single parent’s clothing spend isn’t lifestyle. A millennial’s food delivery isn’t dining out when it functions as a logistics and convenience tool for a working household. A gym membership isn’t fitness when the user joined for chronic pain.

Money management tools that let consumers tag their own priorities, or that infer priorities from which categories survive the next income shock, stop being transactional logs and start being the household’s priority dashboard. That’s a stickier relationship, and one of the few defensible positions left as transaction-data parity among competitors continues to erode.

Read More: The Next Battle in Credit Won’t Be for Top of Wallet

Across all of those use cases, the strategic insight is the same. The successful payments, credit and banking players over the next decade won’t always be the ones with more transaction data. They’ll be the ones who know who their customer is underneath the transaction.

Two consumers with identical demographics and identical purchase histories can have radically different priorities, and those priorities decide how they behave under pressure.

The question everyone should be asking is whether their data strategy reflects that. Or whether it is still bucketing customers simply by what they bought last month.

 

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.

 

That scene plays out millions of times a day across grocery stores, retail shops, restaurant apps and eCommerce checkout flows. A deal is out there. A consumer wants it. But the engine to connect them sputters and fails. The store collects the data point on an unredeemed offer and moves on. The brand absorbs the wasted promotional spend and chalks it up to the cost of doing business. The consumer, tired by all the jitter-jive associated with the three or four steps required to find and then redeem a promo code or an offer, eventually stops trying.

“So what,” some merchants and brands might say. The customer still shopped at the store, still bought the product and, bonus, paid full price. What’s so wrong with that picture?

Here’s what’s wrong.  Every one of those unredeemed offers isn’t just a missed connection. It’s a missed connection with a customer who may shop somewhere else next time.

The consumer who never saw the offer and paid full price probably wondered whether they missed out on a deal somewhere else for the same thing. The brand spent the promotional dollar and got nothing back. The merchant lost the chance to win a basket that might have looked different, even bigger, with a customer who might be more loyal had the right incentive arrived at the right moment.

New data from PYMNTS Intelligence finds that 50% of restaurant diners and nearly half of retail shoppers noticed no offer during their most recent visit. The offer was there but they couldn’t find it, didn’t see it. That’s not a rounding error in the redemption data. These are the customers brands most want to reach and are most consistently failing to convert.

A new PYMNTS Intelligence report produced in collaboration with FIS surveyed 2,754 consumers across grocery, restaurant and retail, putting hard numbers to a problem the industry has been tolerating for years. The findings aren’t a gentle critique. They describe an offers economy that’s out of step with the consumers it was built to serve, extracting data and attention at the top of the funnel while failing to deliver value where it matters most. At the bottom of the funnel, when it’s time to close the deal.

What the data reveals is that the breakdown in the offers economy isn’t about coupon mechanics or loyalty app design. It’s about whether the promotional dollars that brands, merchants and issuers collectively spend are achieving the one thing offers are meant to do. Change behavior.

PYMNTS Intelligence data shows they can, and sometimes they do. But the $42 billion gap in the offers economy points to a structural deficiency that has grown too large to ignore.

The right framework for understanding this structural breakdown is FIT: Friction, Inertia and Time. I’ve been writing about these forces for years, drawing on data from dozens of platform inefficiencies to develop it. What I’ve found is that each force operates independently. Together, they’ve created an offers economy that consumes enormous promotional investment while delivering outcomes that fall far short of what new technology, new models and new ways of thinking about offers now make possible.

Read More: Using the FIT Framework in a Digital 3.0 World

The Wanamaker Problem, Still Unsolved

The early department store pioneer John Wanamaker famously said that half of his advertising money was wasted, but he didn’t know which half. That was in the early part of the 1900s, when paper and pen were the most innovative tools available to track those outcomes.

One hundred and twenty-six years later, that dilemma defines the failure of the 2026 offers economy, even as offers have gotten richer and more plentiful and the technology available to serve them is now extraordinarily sophisticated.

Among the consumers who did find an offer, only 13% online and 10% in-store experienced the automatic application of the discount at checkout. The other 87% to 90% had to put a lot of elbow grease into redeeming them, clearing an average of more than two active hurdles just to use a deal they were already aware of, the PYMNTS Intelligence data found.

And when consumers who saw an offer but said “no thanks” were asked why, 40% said the offer wasn’t relevant. Not the obstacle course between discovery and checkout. Irrelevance. The offer had already captured their attention and in many cases their personal data. But returned a deal they didn’t want.

Read More: Personalized Offers Are Powerful — But Too Often Off-Base

This disconnect is commercially more serious than a leaky funnel.

The $42 billion sitting uncaptured in the gap between promotional dollars spent and consumer value delivered is the economic impact of a system that isn’t living up to its potential.

Seven in ten consumers who noticed an offer changed what they bought or how much they bought when the offer actually reached them.

So, every dollar disappearing into the gap between an offer that existed and a consumer who never found it is more than wasted spend. It’s a lost opportunity to change behavior, win a brand switch, build a basket or earn a payment method preference.

The outcomes the offers economy is supposed to produce.

Making the Offers Economy More FIT

My two cents is that the friction in the current offers system isn’t accidental. It’s a deliberate design choice.

Think about your own experience in finding and redeeming offers. The offers ecosystem was built to use the lure of a discount as a mechanism for data capture at the top of the funnel. Sign up for the email. Create the account. Opt into the text notifications. The consumer is asked to “pay” in personal information and future attention before they have any basis for trusting that what comes back will be relevant or worth the exchange.

When 40% of non-redemptions happen because consumers find them irrelevant, it means the data capture is often happening without the personalization payoff that was supposed to justify it. The brand got the email address. The consumer got nothing they wanted.

The “I love it/I hate it” ubiquitous promo code is the most visible symptom of this dysfunction.

A brand distributes codes across email campaigns, coupon aggregator sites and promotional partnerships, loses control of who redeems them and under what conditions, captures no meaningful attribution data from the transaction and builds no relationship with the consumer who found the deal on a third-party site and who will return to that site next time rather than to the brand or merchant. The code is not a marketing tool. It is a markdown without meaningful attribution.

That dysfunction persists not because merchants and brands can’t see it, but because inertia has made it too comfortable to ignore. And, ironically, so durable.

Consumers have adapted to the broken system rather than rejecting it. They use secondary email addresses for promotional signups. They abandon carts and wait for the recovery email that almost always arrives with a better offer. They have learned the game and play it, which creates a false signal for merchants and brands who look at redemption rates and email list growth and conclude that the system is working.

What the metrics don’t capture is the 27% of shoppers who pay no attention to offers at all because the current offers ecosystem hasn’t given them a reason to engage. Those consumers aren’t lost. They’re waiting for an offers experience worth their attention and their loyalty. The longer inertia holds the current system in place, the more the third force, time, works against everyone.

Read More: Embedded Offers: The Billion-Dollar Opportunity Inside Recent Consumer Spending

The dominant offer discovery channels — the merchant apps, checkout screens and in-store signage — all require the consumer to already be inside the purchase flow. The offer arrives after they’ve already finished their shopping. It creates no opportunity to influence what goes in the basket, consider an alternative product or increase spend.

The generational data makes the time pressure across merchants and brands more real. Gen Z is the most likely to shut the door entirely on offers, and therefore brands and stores, when presented with manual steps and promo code hunts.

Read More: The Five Rules of Engagement for Gen Z Spending and Payments

Nearly one in five Gen Z (19%) shoppers who saw an offer and didn’t use it cited too many steps as the reason, compared to one percent of boomers. For this generation, friction is not an inconvenience. It’s a reason to shop with another merchant or stick with the brand they already know rather than try something new.

What a Different System Changes

The  case for a better offers architecture starts with understanding, and then embedding offers, into the customer journey at the start, and not at the end.

Nearly nine in ten consumers say they want to see every relevant discount before they decide what goes in their cart. Before they decide they’re finished shopping and want to call it a day.

An embedded smart offer delivers it at the moment of intent. That shift does more than improve the user experience. It changes the purchase decision itself.

An offer that finds the consumer rather than waiting to be found doesn’t require effort, trust in a coupon aggregator or memory of a promo code under pressure at checkout. It arrives attached to the product in their consideration set, delivered through the card they already use, built on their actual purchase history rather than a demographic bucket.

The data finds that when offers do reach consumers, seven in ten change what they buy. Seven in ten change how much they buy. The goal is to make finding offers the default rather than the exception.

Read More: The $42 Billion Checkout Opportunity Hiding in Plain Sight

One-to-one personalization is what separates this model from every prior iteration of card-linked offers. The card credential carries transaction history across merchants and categories at a level of granularity that no email list or loyalty program can approach.

It knows not just that a consumer shops at a particular grocery chain but which brands they buy consistently, which they have tried once and abandoned, which categories they trade up in and which they treat as commodities.

A dynamic offer built on that data is much more than a discount. It is a precisely timed intervention in a known purchase pattern, served to defend a brand relationship showing signs of fatigue, to introduce a product at the moment the consumer is most likely to try something new, or to reward the specific behavior the brand wants to reinforce.

Read More: AI Pushes Personalization From Guesswork to Growth

According to PYMNTS Intelligence data, four in ten consumers say a form of smart, embedded real-time savings would be highly influential in making a payment method that supports it their default. Seventy-seven percent say it would be at least somewhat influential. The card that delivers embedded smart offers seems influential in winning in the wallet.

New Economics for Merchants and Brands

The attribution crisis at the center of the offers economy is as consequential as the consumer experience problem but gets far less attention.

Brands spread promotional budgets across email, coupons and loyalty programs and get back data that is too weak and too delayed to guide the next decision. Money goes out and redemption rates come back, but whether an offer drove new behavior or simply discounted a purchase that would have happened anyway is largely unknowable.

It is the Wanamaker problem, 126 years later, still unsolved.

Using the card credential as the delivery layer could address this at its root. The card is present at every transaction, which means a brand funding an offer at this layer knows exactly what was purchased, where, by whom and whether the behavior was incremental, such as a trial, a basket expansion or a brand switch.

The promotional dollar no longer buys impressions. It buys outcomes. Return on investment is not inferred. It is measured, attributed and immediately actionable.

For merchants, this changes the economics in a meaningful way.

Brand-funded offers become a source of incremental revenue rather than a cost center. For merchants, this reframes offers from an expense to manage into an asset that drives acquisition and retention.

The result is a reallocation of promotional spend away from impression-based marketing with uncertain outcomes and toward incentives tied to specific products, merchants and behaviors. A fundamental shift in how the offers economy works.

The Card’s Untapped Commercial Position

For issuers, this creates something new. It is a commercially meaningful position at the center of the offers economy that no ad network, email platform or coupon aggregator can replicate.

It is grounded in trust, verified transaction data and full visibility across the purchase journey. The question is whether issuers recognize the opportunity and move with the urgency it demands.

PYMNTS Intelligence data suggests the consumer is already there. Among those most willing to engage with embedded and personalized offers, 75% are willing to share data with banks and 78% with retailers.

These consumers are high frequency and high value. They skew toward millennials and bridge millennials and have established financial relationships.

This is the audience brands want to reach, and they’re ready to let the card be the platform that connects them.

Why Agentic Commerce Shortens the Window

The case for a better offers economy stands on its own in the current environment. But agentic commerce makes the window for building it shorter than most issuers and merchants currently appreciate.

AI-powered agents already assist consumers with purchase decisions. The next generation won’t assist. They’ll execute. A consumer sets preferences and constraints and the agent handles the research, the selection and the transaction. In that environment, the offers ecosystem as currently constructed isn’t just inefficient. It’s irrelevant.

Read More: What Happens to Stores When AI Agents Do the Shopping?

An AI agent doesn’t hunt for promo codes. It doesn’t sign up for email lists. It doesn’t activate loyalty IDs. If an offer can’t be accessed programmatically through a credentialed interface the agent can query and apply without intervention, the offer doesn’t exist for that transaction. The promotional dollars behind it simply don’t connect.

A tokenized smart credential is best positioned to survive this transition. An offer embedded at the card credential layer doesn’t need the consumer or the agent to do anything. It’s queried, surfaced, applied and attributed automatically. The agent selects the payment credential that delivers the most value at the point of purchase. The card with the richest offer layer wins the transaction by default, without ever competing for attention at a checkout screen.

Read More: Demystifying AI’s Capabilities for Use in Payments

The card that hasn’t built this capability will not lose that competition gradually. It will lose it at scale, as agent-mediated commerce moves from a niche behavior to a default one over time.

The Offers Economy Bet Worth Making

The argument for a better offers economy ultimately rests on a behavioral claim the PYMNTS Intelligence report makes with data-driven clarity.

Offers change what people buy, how much they buy and which payment method they use to buy it.

According to the data, these aren’t incremental improvements. They’re substantial shifts in commercial outcomes driven entirely by the presence of a relevant, timely, frictionless offer.

The current offers economy captures only a fraction of this behavioral potential because it was designed around data extraction rather than value delivery, and because the friction it introduces at every stage causes the majority of consumers to never encounter the offer at all or to abandon it before redemption.

An embedded smart offers model inverts this logic entirely. The offer finds the consumer. It’s relevant because it’s built on verified transaction data rather than demographic stereotypes.

It’s applied automatically, so the behavioral effect does not depend on the consumer remembering to act at the right moment or clearing two, three or more hurdles to redeem it, if they see it in the first place.

And it creates a data loop that allows brands to observe outcomes directly and improve targeting with each iteration.

The consumers most ready to engage with this model represent more than half the U.S. adult population.

The brands most likely to fund it are the ones watching their promotional budgets produce diminishing returns in channels that can’t demonstrate attribution.

The merchants most likely to benefit are the ones that need new sources of promotional revenue that don’t require them to sacrifice already thinning margins.

The issuers most likely to build it are the ones that understand the card credential is no longer just a payment instrument. It’s foundational for building a relationship with their customer.

 

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.

 

But for everyday things, all of the things you need reliably, relatively quickly and without a lot of rigamarole, Amazon is probably your go-to.

It didn’t take long. By the beginning of the millennium, four to five years after it launched, Amazon had secured its spot as the world’s largest online retailer. And it deepened its moat over time as Amazon added more and more to its inventory and more and more sellers with more products to its digital shelves. Books first, then music, then electronics, then apparel, then grocery, then healthcare, then used cars, then prom gowns, bridal dresses and patio furniture.

Eventually the question of where to look first was replaced by a reflex to pop open the app with the lowercase “a” on the home screen.

That reflex is now forming in GenAI. And the story of how consumers are choosing AI models is tracking the same arc.

Understanding one is the fastest way to understand the other. And to predict what the competitive landscape in AI is likely to look like not next quarter, but over the next three to five years. The data is already telling the story. The question is whether the platforms competing in this space are reading it right.

The Habit Is Already Here

A decade ago, conversations at my hair salon over the holidays were about Alexa. Not the countertop Echo device, but the personification of an assistant whose wish would become their command.

In 2026, conversations at those same salons are remarkably similar. But the main character is not Alexa anymore. It is “Chat” or “GPT.” People are asking about the best mattress to buy, whether it is faster to fly or drive from Boston to New York, how to plan the most perfect 13th birthday party and how to assemble a capsule wardrobe for spring. GenAI, and how consumers use it, has gone mainstream.

PYMNTS Intelligence surveyed roughly 15,000 U.S. consumers over the last five months about how they complete 54 personal tasks across nine categories of daily life. The goal was to document which tasks are now AI-first and where AI models have replaced traditional methods of primary consumer interface.

Today, 146 million people, 56% of U.S. consumers, use an AI model to complete at least one personal task. The share of non-users fell from 48% to 44% between October 2025 and February 2026. Mainstream users rose from 30% to 34%. There is a steady, directional move toward more AI use by more people doing more things.

Not a hockey stick. Not a hype cycle. A habit forming at scale, with no signs of slowing down.

The tasks driving this adoption are mundane, even a little tedious. Writing and communication dominate, with editing and drafting emails representing the most common activity. Product discovery follows closely. In fact, finding product links is the single most-performed task across the entire survey sample, the top-ranked activity, and the most stable trend line across the five-month dataset. The traditional method users are leaving behind is Google.

Read More: Why 30 Million US Consumers No Longer Search

More than a quarter of consumers consult AI models for health-related activities, specifically looking up symptoms and researching medications. Daily task lists, meal planning and grocery lists all show gradual upward momentum.

These are not one-off use cases but the daily slog of modern life. AI models are becoming a convenient one-stop shop to address them all.

AI habits are formed at the low-stakes, high-frequency end of the day-in and day-routine, the activities that everyone does more or less daily or multiple times a week. Exactly where Amazon built its lead in online retail.

Amazon won by being the most reliable first stop for the broadest range of everyday needs, and by making the experience of starting there so easy and reliable that choosing something else required a reason not to. It’s the same position ChatGPT is building right now, even as other models emerge to capture more specialized activities.

The More Than One Model Reality

The average active AI consumer now engages with more than two platforms; power users, the 10% of consumers who represent roughly 30 million people, engage with nearly four. On the surface that looks like a fragmented market. It really isn’t.

Think about how most Americans shop. When looking at consumers’ latest retail purchase, 55% shopped at a total of 13 stores. Amazon captures more than 56% of their online spend. The rest goes to a handful of specialists for those one-off, less routine purchases. The boutique with the better shoe selection, the wine shop where someone knows their palate, the specialty fish store worth driving to. It looks like variety. It’s actually much like the shopping hierarchy: one go-to that handles the everyday, and a short list of niche players  that earn their slots by doing something their go-to doesn’t do as well.

The multi-model AI story is the same story. Engaging with multiple platforms is not the same as dividing time equally among them. ChatGPT anchors the stack for every user segment. Other models earn specific slots based on use cases. The consumer using more than one platform is real. Parity of use among those platforms is not.

AI Task Map: Complexity vs. Frequency

What determines where a model earns its slot, and where ChatGPT holds its ground, becomes clear when you map the 54 daily tasks PYMNTS Intelligence tracks against two dimensions: how complex and high-stakes the activity is, and how frequently the consumer performs it.

The Commodity quadrant, high frequency and low stakes, is where habits form and where ChatGPT wins by the simple fact of having been there first and improving over time. Rewording a sentence, finding a product link, building a grocery list. The consumer knows in three seconds whether the answer is right. The cost of the wrong one is a paragraph with too many em dashes or radishes on the grocery list instead of radicchio. It’s where most consumers started their AI-first journeys, and where the critical mass of everyday use still lives.

Read More: How Time Became the Next Great Asset Class

The Trust Gap quadrant, high stakes and lower frequency, is where the market is still being decided. Medication interactions. Loan comparisons. Whether that non-compete clause is enforceable.

Unlike the commodity tasks, the consumer has almost no way to verify the accuracy of the output in real time. Adoption lags here not because consumers don’t value it, but because trust is harder to build when the tasks happen less often and the stakes of getting it wrong can be consequential.

This is where platform differentiation does matter, and where the AI fluency of the user becomes more apparent and adoption-defining.

ChatGPT leads all nine task categories, from 42% in shopping to 60% in writing. it looks different among power users depending on the task. Gemini gains ground in financial and health-related tasks. Claude does in complex document and contract review. Copilot holds where Microsoft Office integration makes specific outcomes more personalized.

We find that power users have already built what the rest of the market hasn’t needed to yet: an AI portfolio with a primary model for everyday tasks and others where they’ve discovered that different models deliver a better outcome.

The more subtle insight is that this behavioral pattern did not emerge from how any of these models set out to build a user base. It emerged from trial, error and engagement over time. Trust grew incrementally, activity by activity, until the habit of starting somewhere became the habit of staying there.

The Invisible Influencer: Consumer vs. Enterprise

The most important force shaping how consumers discover new AI models isn’t visible to them. It isn’t marketing. It isn’t social media. It’s the workplace.

For most consumers, the journey starts the same way Google once did. Personally. A question. A task. Something they need to figure out before the kids wake up or in the middle of a busy day. ChatGPT was the model they found first, because it was there first, so it became the model they trusted first, one low-stakes task at a time.

Read More: Gen AI: The Technology That Broke the Adoption Curve

As that habit deepens, it moves into work. The same model now shows up in higher stakes moments: drafting memos, summarizing documents, preparing for meetings. The personal habit now becomes the professional one. But as the work becomes more complex, and the stakes attached to the output become more related to job performance, the limits of that default start to show. Not enough to displace it, but enough to create a reason to consider something new.

Often, the second model doesn’t show up because the user goes looking for it. It’s introduced. A colleague uses something different. A team adopts a tool for a specific workflow. A company standardizes on a platform. Or the task itself becomes demanding enough that the user is pulled toward something more precise.

This is how models like Claude gain ground. ChatGPT expands outward from the consumer, earning trust in low-stakes, high-frequency tasks and carrying that trust into the workplace. The habit comes first; the enterprise follows.

Claude follows the opposite path. It is encountered in the context of work, where precision matters and the cost of getting it wrong is higher. Contract analysis, code review and complex research are not entry points for casual use. They are reasons to adopt something new. In this case, the enterprise is not the endpoint but the starting point.

Read More: How Leading Enterprises Really Measure Gen AI ROI

These models earn trust in high-stakes moments rather than everyday ones, which is why its growth shows up differently. Not as a default, but as a deliberate choice tied to specific use cases.

What emerges is not a fragmented market, but a structured one. One model becomes the default, the place users start without thinking, while others earn their place more narrowly, tied to tasks where performance justifies the switch. It looks like variety on the surface. In practice, it is a hierarchy.

That is the fork that matters. One path builds from habit outward. The other builds from necessity inward. Every platform in AI is now, whether intentionally or not, choosing which side of that fork to pursue.

Read More: Big Tech Faces the AI Innovator’s Dilemma

Gemini: the GenAI Dog That Hasn’t Barked

Of every company positioned to own the daily AI habit, Google started with the most built-in advantages. Gmail has more than 1.8 billion active users. Google Search handles billions of queries per day. Google Calendar, Maps, Photos and Drive are the embedded infrastructure of daily life for more than a billion people. If habit formation is about being the first stop in a consumer’s day, Google was already there. It had been there for twenty years. And yet, when consumers name the AI model they use most for daily tasks, the answer isn’t Gemini.

ChatGPT started in a different place. A conversational model that says, “ask me anything.” Not the reflex of searching for something already known, but the instinct to think out loud and work through something more complex than a few keywords could capture. Google goes and fetches what consumers have already decided they want. ChatGPT became where consumers go to figure out what to do next with options and detail.

In some ways, the familiarity consumers have with Google may actually be working against Gemini rather than for it. Consumers do not associate Google with the kind of open-ended, generative, think-it-through experience that defines how they are now using AI. They associate it with looking things up with a bunch of links received in return. That association, built over two decades, is sticky in ways that a product rebrand cannot easily overcome. And for many so far, that their experiences with it have proved disappointing.

Read More: Why AI Shopping Is Still Just a Smarter Search Bar

Two Different Tuesday Mornings

So, here’s what a Tuesday morning now looks like for more than half of Americans.

Charlotte wakes up in Baltimore and uses ChatGPT to help draft a text to her kid’s teacher. Before she’s had her first cup of coffee, she’s also asked it to find the best price on a coffee maker and pull the purchase link. By lunch, she’s used it to look up the symptoms of a rash that appeared on her arm. Not to self-diagnose, she’d tell you, just to know whether it’s worth running to urgent care. In the afternoon she used it to rewrite an email requesting a merchant refund, producing a draft that was considerably more diplomatic than her original. By evening, she’s back at the GPT prompt to plan a packing list for next weekend’s camping trip and build a grocery list based on the recipes she wants to make for the week.

Charlotte never once stopped to think about whether she was using an agent, or whether AI was even the right word for what she was using. She just did what she needed to do, at a prompt she already trusted.

Three hours later, Katie wakes up in Seattle and opens Gemini to scan her investment portfolio, asking it to summarize overnight moves in three sectors she’s watching. She switches to ChatGPT for the first draft of a memo she needs to circulate before ten, and to get suggestions for a ten-day European trip in July. At lunch she asks ChatGPT to draft an agenda and fundraising marketing plan for a non-profit board she chairs. By midday she’s in Claude, asking it to explain the meaning of a contract clause her lawyer flagged. She uses Copilot to create a travel itinerary around a business conference in two weeks, because she’s learned it surfaces hotel availability better than the others.

Katie doesn’t think of herself as a consumer using several AI tools. She just thinks of herself as someone who uses the right tool for the job: a primary model for most things, specialists for the tasks where they perform better.

Two consumers. Same Tuesday. Very different relationships with AI, and with the platforms competing for their attention.

The Habit Hardens

Let’s end how we started. With shopping.

You probably didn’t sit down one day and decide that Amazon would be your default for everyday essentials. You just used it. Then used it more. Then stopped noticing you were making a choice. The habit formed because of the friction you stopped experiencing. The comparison shopping you no longer bothered with, the store you used to drive to that now feels like too much. Amazon didn’t win your loyalty. It won your reflex by being convenient and reliable and trusted.

Read More: The First Chatbot Consumers Try May Be the One They Stick With

That is exactly what is happening in GenAI right now, and the window for platforms to shape habits may be narrower than most of them seem to appreciate.

Consumers are not evaluating AI models. They are using them. And with each use, the mental calculus of where to start shrinks a little more. Charlotte in Baltimore isn’t going to wake up one morning and reconsider her relationship with ChatGPT. She is just going to keep opening the app. The task will change. The prompt will change. The reflex won’t. As long as the output remains valuable.

For the platforms competing in this space, this is both the opportunity and the threat.

The opportunity is that habits at this stage are still forming. The commodity quadrant is not fully locked, the trust gap quadrant is wide open, and the AI-fluent consumer who becomes a power user is still building her AI stack. The threat is that every day a consumer completes a task with a model and walks away satisfied, that model gets a little harder to displace.

Amazon’s lesson was not that it built the best online store. It was that it built the most reliable first stop, across the broadest range of everyday needs, before its retail competitors understood that the first stop was the thing worth competing for. By the time they did, the habits were already set.

The GenAI market is in the first few innings of that same game. ChatGPT holds the first-stop position for the everyday. Claude is earning its ground at the high-stakes end, task by task, credential by credential. Gemini is still looking for the moment when its structural advantages convert into behavioral ones. Copilot has inroads for the tasks where Microsoft integration produces an outcome worth switching for.

What happens next will be determined less by product releases and more by how habits are formed and how embedded that reflex is in everyday life. And whether the platforms understand that they aren’t just competing for users, but for the order in which those users show up at their prompt.

The consumer who opens ChatGPT before she’s had her coffee isn’t likely to be pulled away by a better feature set alone. If she changes her behavior at all, it will be because another model earns a specific role in her routine. Much the way a great wine shop earns its place, not by replacing Amazon, but by being worth the detour. That is the competitive reality of the next three to five years. Not a single winner. Not a fragmented free-for-all. A hierarchy, with one anchor handling the everyday and a short list of niche players competing for attention at the edges.

 

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.

When I sat down with Kelly Cook, it didn’t feel like an interview. It felt like we were getting right back into a conversation that never really stopped.

Cook has always been direct about what she’s trying to do at David’s Bridal. As she approaches her first anniversary as CEO, that clarity hasn’t softened. If anything, it’s sharpened. She didn’t take the job to tweak a legacy business. She took it knowing that a company built around a single, emotional purchase had to be reimagined entirely. And quickly, with a totally different mindset — revolutionary, not evolutionary. A strategy that would also energize the team.

“I told the [team], sometimes when you’re in a dark place, you think you’ve been buried, but you’ve actually been planted.”

That line stopped me. I joked with Cook that she should put that on a cup or a mug.

But that line captures not just the ambition of what she’s doing, but the standard she’s holding herself and her team to, as they rebuild a brand that had already been through two restructurings in four years.

Rethinking the Problem Before Solving It

Cook’s approach to the transformation of David’s isn’t to start with solutions. She starts by reframing the problem.

David’s, on paper, is a retailer. But as she and her team dug into the data, what they saw didn’t look like a single-transaction business at all. It looked like a long, complex, emotionally loaded journey. One that extends over 18 months and involves hundreds of decisions.

“The Big Day has about 300 tasks,” she said. “And she’s buying 18 outfits,” she added, still sounding slightly surprised by that realization.

But it was that insight that changed everything. If the company continued to organize itself around selling a dress, it would miss most of the value. And most of the relationship.

What followed was what Cook calls the shift from “aisle to algorithm”: using data not just to understand the customer, but to stay connected to her across multiple moments. It’s also changed how decisions get made internally. The goal now is speed with accountability. Turning data into insight, insight into action, and then quickly measuring what worked and what didn’t.

That last part matters as much as the first.

How She Decides: Instinct, Data — and a Willingness to Walk Away

Cook talks about “big bets,” but what’s interesting is how often those bets involve deciding not to keep going.

We talked about live shopping, which had an early moment of promise for the company. The first event performed well with strong engagement, strong results. By most conventional measures, it was a success.

But something didn’t sit right with her.

Even as the numbers came in, she had a sense that the broader trend was already peaking. That consumer excitement around live shopping wouldn’t sustain at a level that justified continued investment.

So they stopped.

That’s not an easy decision to make, especially after an initial win. But it reflects how Cook thinks about return on investment over time, not just in the moment. She’s less interested in chasing spikes than in building systems that compound.

It’s also an example of how she balances data with instinct. The data said “go.” Her experience, and her read on where the market was heading, said “be careful.”

She listened to both, but ultimately acted on the longer-term signal.

Expansion Isn’t a Strategy — It’s a Consequence

The move into categories like prom, graduation and other formal occasions is often described as expansion. But in Cook’s telling, it wasn’t really a choice. It was an outcome of understanding the customer more fully.

“The woman who comes to us for one occasion will come back,” she said.

Prom became a particularly important entry point, not because the company set out to chase a younger demographic, but because it realized that identity, style and occasion don’t map neatly to age.

“Mom is an attitude, not an age,” she said. Another one of those lines that feels like it belongs on that same coffee cup.

What looks like category expansion is really a shift toward continuity. Creating multiple reasons to engage with the brand over time, and increasing lifetime value in the process.

Measuring What Actually Matters

For all the talk of transformation, Cook is disciplined about how she measures progress.

Financial performance is, of course, part of it. But she puts equal weight on engagement, repeat behavior and customer feedback. The signals that indicate whether the relationship is actually deepening.

And then there’s trust.

She comes back to that word often, and not in a vague way. Trust, for her, is both a brand metric and a business driver. If customers trust David’s to show up for more than just the wedding day, the rest of the model works. If they don’t, it doesn’t.

That’s why the shift to a broader ecosystem, including partnerships with Amazon, Walmart, DoorDash and independent boutiques isn’t just about reach. It’s about relevance.

The retailer’s DoorDash partnership, by the way, is crushing it in Vegas, Cook said.

“We want to serve every bride,” she told me, “whether she buys from us directly or not.” Or someone who needs a dress in 20 minutes.

As she put it in a way that stuck with me. The goal is to “be the engine in everybody’s car,” not just a single car on the road.

Moving Fast — and Accepting the Misses

Of course, not every bet works, and Cook doesn’t pretend otherwise.

Artificial intelligence is a good example. The company is using it across customer experience, operations, data analysis and content creation. But speed introduces risk. At one point, an AI-generated marketing asset went out with a visible error, the kind of mistake that reminds you how quickly things can go wrong when processes haven’t caught up to ambition.

Her response wasn’t to pull back.

“I’d rather fall forward going fast than fall backwards going slow,” she said.

That doesn’t mean being careless. It means accepting that in a period of transformation, some mistakes are the cost of momentum, and that the bigger risk is hesitation.

One Year in — and Still Pushing

As we wrapped up, what stood out wasn’t just how much has changed in the past year, but how much Cook still sees ahead.

She’s not talking about stabilization. She’s talking about building a fundamentally different kind of company. One that doesn’t depend on a single moment but participates in a lifetime of them.

And she’s doing it the same way she started: by making big bets, measuring them rigorously, and being willing to change course when the signal, whether from data or instinct, tells her to.

Not buried.

Planted.

And yes. I still think that belongs on a coffee cup.

But that wasn’t the most interesting part of my experiment.

Even if the AI had identified the right toaster, it still couldn’t confirm the actual price, verify that the item was in stock or tell me whether it could be delivered within any reasonable timeframe. It could not place the order. To do that, I was referred to a merchant site I had never heard of to navigate the checkout.

So, I ended up doing what most people still do instead.  I went to Amazon and bought the toaster there.

The infrastructure required to actually curate all of the available options, and then execute the purchase, is a big gap yet to be filled.

I have repeated my Toaster Test periodically ever since. The LLMs are smarter, faster and more conversational than they were a year ago. They have added a few better brands to the list they produce. They still cannot produce the brand I wanted to buy, confirmed in stock, at a real price, available for purchase right now. The research loop is impressive. The transaction loop is anything but.

My test is not evidence of a technology failure. In fact, the AI part worked amazingly well. It is a marketplace failure.

And it illustrates the core challenge of agentic commerce in 2026.

The models are breathtakingly amazing at helping consumers do the research and the shortlisting of options. The infrastructure required to actually curate all of the available options, and then execute the purchase, is a big gap yet to be filled.

What is billed as a revolution in commerce is, for now, mostly a highly intelligent search bar. A better one than Google. A more conversational one. But still, at its core, a tool that finds the answer and then hands the consumer off to someone else to close the deal.

The Consumer Shift Is Real

The behavioral shift behind this problem is no longer theoretical. It has gone mainstream quickly.

PYMNTS Intelligence research from January 2026 found that 41% of consumers have already used dedicated AI platforms for product discovery. More striking is that a third say they have fully replaced their prior methods. They are not layering AI on top of old habits.

They shut the door on the old way and left.

Read More: Why 30 Million US Consumers No Longer Search

AI adoption crossed 54% of U.S. adults in January 2026, up ten points in a single month.

Among millennials, two in three have used a conversational AI assistant for research.

That’s just the tip of the behavioral iceberg.  In December 2025, 34% of AI power users relied on native AI interfaces as their primary method for shopping discovery.

One month earlier that share was 22%. Among light users, reliance on AI models  jumped from 5% to 16% in the same period.

These are not gradual shifts. They are new habits forming at a pace the industry did not anticipate.

Read More: Gen AI: The Technology That Broke the Adoption Curve

The experience driving this shift is genuinely different from traditional search. Instead of scrolling through pages of links and sponsored listings, consumers receive a structured answer that explains the tradeoffs between competing products and can be refined through conversation until it matches the actual buying decision. It is something keyword searches could never deliver with any sort of precision.

But then the consumer leaves the conversation and goes somewhere else to complete the purchase.

The question that matters now is not whether agentic commerce will eventually close that gap. It is who becomes a casualty on the agentic highway, and who benefits.  And how.

Google and Shopify on the Defense

Let’s start with Google, because the damage there is real and already underway, even if their headline numbers still look healthy.

In Q4 2025, Google Search revenue grew 17% year over year. Gemini has 750 million monthly active users. Alphabet crossed $400 billion in annual revenue for the first time. This is not a wounded player. It is also not the dominant one it used to be.

Google has been trying to become a commerce destination since it launched Froogle (a play on “frugal”) in 2002. It rebranded that effort multiple times, built Shopping tabs, launched Google Express, acquired Pointy and embedded Gemini.

Despite all of it, Google Shopping remains a listing service that shows products and sends consumers somewhere else to buy. The transaction, the customer relationship and the post-purchase experience all happen in someone else’s ecosystem.

What has changed is the top of the funnel. Consumers who once opened Google to research a product are now opening ChatGPT, Claude or Perplexity instead. The PYMNTS data makes that shift quantifiable. A  third of consumers who have tried AI for shopping discovery have fully replaced their prior methods.

Read More: What Happens to Stores When AI Agents Do the Shopping?

That is not a marginal shift in behavior. It is a structural fracture of the search functionthat Google has monetized for two decades.

Consumers who switch to AI for research are already bypassing Google at the top of the funnel.

Defending a position is not the same as expanding it. The highest-value transactional queries, consumers who already know what they want and are ready to buy, may still run through Google’s standard channels.

But the middle of the funnel, the research and comparison phase where consumer intent is shaped, is moving to AI platforms that Google does not own and cannot easily monetize, despite Gemini’s headline user numbers.

Consumers who switch to AI for research are already bypassing Google at the top of the funnel. If Google cannot capture them at the bottom with a transaction, it loses the journey, and that customer, entirely.

Google’s answer is the Universal Commerce Protocol, an open standard built with Walmart, Shopify, Target, and two dozen other partners, designed to let AI agents complete full shopping journeys inside Google’s own products. The logic is that if Google can plug trusted commerce players into its AI surfaces, consumers can transact through Gemini while the actual commerce relationship belongs to the retailer. Google becomes the front door without building the back end.

Read More: The Protocol Power Struggle Reshaping AI-Driven Commerce

This is more or less what it does today with Google Shopping, except with agents. It is also an acknowledgment that Google cannot build what a commerce network requires. It is borrowing the trust it does not have from partners who do.

Then there’s Shopify. Their situation is more complicated and more consequential for the merchants who depend on it.

Twelve months ago, Shopify had the most credible claim to being the open-web alternative to Amazon. Millions of merchants, a high-profile AI commerce partnership with OpenAI that sent competitors scrambling and a narrative about the future of direct-to-consumer commerce that the industry largely accepted.

Read More: Can Shop Cash Turn Shopify Into an Amazon Challenger?

That narrative is now in trouble on two fronts.

The OpenAI native checkout partnership is gone. OpenAI pulled it after fewer than 30 merchants went live with a product that had not built the systems to collect state sales taxes. The Shopify landing page built specifically for ChatGPT now redirects to its homepage. Catalog syndication still exists, so Shopify merchants can be discovered inside ChatGPT. But discovery without a transaction is a referral, not a commerce relationship.

OpenAI is now a discovery layer that sends consumers somewhere else to buy. That is exactly what Google has always been. It’s not where Shopify needs to play.

Shopify has no fulfillment network, no consumer credentials at scale, and no native AI agent with meaningful adoption.

Read More: From Assistive to Agentic AI: Consumers Wade Into Autonomous Commerce

The second front is Amazon. Shop Direct, launched in February 2025 and expanded to more than 100 million products from over 400,000 merchants, allows Amazon Prime customers to buy from brand websites using their stored credentials through the Buy for Me capability.

Amazon is now offering independent brands something Shopify cannot match: Prime subscribers, Amazon payment rails, Amazon logistics and an AI agent already in the consumer’s pocket.

Shopify’s strategic response is to position itself as open-protocol infrastructure for agentic commerce, building integrations that make merchants on its platform discoverable across any AI surface. That is a reasonable long-term play.

Read More: Shopify Bets Big on Agentic AI

It’s also coming from a fundamentally weaker position than the one Shopify held a year ago. With an uncertain timeframe for when commerce, at scale on AI models will happen.

Walmart’s Two-Step Agentic Commerce Play

Walmart has made a deliberate and smart decision. It has opened its full product catalog online to Google’s Gemini through its Sparky assistant, which surfaces a Walmart-branded experience inside external AI platforms. When a Gemini user searches for a product, Gemini calls Sparky, which opens what Walmart’s head of AI describes as a window inside Gemini where the Walmart shopping experience takes over.

This has been packaged as a bold bet on open agentic commerce. But is it?

What Walmart is doing with Gemini is not meaningfully different from what it has always done with search advertising. It’s funneling traffic to Walmart.com through a different front-end interface, hoping to snag net new customers who might not have found Walmart through a traditional search query.

The transaction still happens in Walmart’s commerce system. The consumer relationship still belongs to Walmart. The agent is the channel, not the commerce infrastructure. Calling it agentic commerce is a generous description of what is, in practice, a search query that renders inside a chatbot window owned by Google that funnels queries to Walmart.

Read More: Why the ‘Person’ of the Year in 2025 Should Be the Chatbot

The consumer relationship still belongs to Walmart.

The more interesting and strategically significant bet Walmart is making is Sparky and the opportunity it represents to convert its 100 million per week physical store shoppers into online customers it can monetize through agents. This is where Walmart’s agentic story becomes genuinely compelling.

Walmart has a physical footprint and, with One Pay, an online wallet and payment method that’s akin to Amazon’s in the physical world.  If Sparky can move even a meaningful fraction of those in-store shoppers into a digital commerce relationship where Walmart can apply personalization, subscription economics and agentic purchasing, well let’s just say that the opportunity is substantial.

Read More: Walmart Rolls Out Agentic Advertiser Assistant

The risk in the Gemini relationship is also real. Openness means the agent can send consumers to a competitor when the competitor’s offer is better. Walmart is trusting that it wins those comparisons often enough and that Google surfaces its products without letting its own commercial interests shape the ranking. That trust has not been tested at scale. And it runs directly into the structural problem that has undermined every search-based commerce experiment for two decades.

How intent and eyeballs get monetized.

The Right-Now Winners: Amazon and the Status Quo

The reality of agentic commerce in March 2026 produces a short list of those in the pole position right now.

Amazon is at the top.

The status quo of how most people actually shop online is a close second.

Amazon didn’t wait for the industry to agree on rules. It spent three decades building out a marketplace and AI-enabled it inside its own ecosystem. It’s now extending that solution outward on its own terms.

Rufus, Amazon’s AI shopping assistant, handled 250 million shoppers in 2025, with monthly active users growing 140% year over year. Amazon says that its Rufus users are 60% more likely to complete a purchase than non-Rufus shoppers. When the conversation ends, the transaction completes immediately because the marketplace already exists behind it.

Amazon is making its marketplace bigger, more open, and more valuable, while ensuring that every transaction, wherever it originates, runs through Amazon’s infrastructure on Amazon’s terms.

Shop Direct, however, is the more significant development. By expanding to more than 100 million products from over 400,000 merchants and enabling Prime subscribers to buy from brand websites using stored Amazon credentials, Amazon is doing something strategically important.

It’s not just defending its marketplace. It’s making its marketplace bigger, more open and more valuable, while ensuring that every transaction, wherever it originates, runs through Amazon’s infrastructure on Amazon’s terms.

The Prime subscriber base is the strategic asset that makes this possible. Prime members spend significantly more than non-Prime customers. Their purchasing intent is high. Their payment credentials are stored. Their trust in Amazon’s fulfillment is established. When Amazon brings consumers to a brand’s website through Buy for Me, it is not sending a casual browser. It is delivering a buyer with an intent to complete a purchase. That is a value proposition that competing platforms cannot replicate without the same combination of payment infrastructure, logistics capability and consumer trust that Amazon has spent 30 years building.

That’s despite the recent Perplexity lawsuit, which, within the space of a few days, clarified the legal landscape in Amazon’s favor, as another judge just yesterday (March 17) reversed it pending an appeal, according to Bloomberg.  For now, the bots are there, free-riding on an ecosystem that they didn’t build, and according to Court filings, they allegedly disguised their bots in order to gain access.

Read More: Amazon Injunction Could Change the Future of Agentic Commerce

The second winner is the status quo of how online shopping actually works.

Most consumers, most of the time, in 2026, still shop the way they did before AI agents arrived. The search and purchase journey is largely disaggregated. They search Chat or Google or Amazon, then buy from wherever they can find the right price and the fastest delivery.

Read More: Legacy Business Models Break

What Still Has to Be Built

The gap between what AI can figure out and what it can actually do about it is not a small one.

Closing that gap requires things that do not yet exist in combination. It requires a real business model for catalog access, not just technical protocols. Retailers need a commercial reason to make their live inventory available to AI agents, a compensation structure that makes that exposure worthwhile, and governance that protects them from being used as price comparison tools that send consumers to competitors. The open standards being developed are the pipes, not the economics.

Read More: What Happens to Stores When AI Agents Do the Shopping?

It requires careful thinking about those economics. Every commerce network built on a discovery foundation eventually faces the same crossroads. Merchants pay for visibility. Promoted products rise above better-matched ones. The consumer notices the results look like ads. Trust flies right out the window.

Read More: Why Trust is Data’s Only Real Currency

The AI agent that tells a consumer it found the best product for their needs while receiving compensation from the merchant whose product it recommends is not working in the consumer’s interest and looks like a more sophisticated ad disguised as advice.

Closing that gap requires things that do not yet exist in combination.

That requires brands to make hard choices and some big bets.

Meanwhile, Amazon isn’t waiting.

It is using every month of industry delay to extend its position. Every new merchant added to Shop Direct, every consumer who uses Buy for Me, every improvement to Rufus deepens an ecosystem that is already the hardest thing in commerce to replicate. The open AI platforms are not competing against a static target. They are competing against a moving one.

What’s Next

I have written about the promise of agentic commerce many times. The thesis has always been the same. Consumers want an agent that works for them. One that knows their preferences, understands their constraints, does the research, makes the comparison and closes the deal on their behalf. Not a better search bar. Not a more conversational listing service. An actual agent that shops.

Read More: How Consumers Want to Live in a Conversational Voice Economy

The consumer demand is not in question. It never was. A third of consumers have already abandoned their prior shopping methods entirely. More than 70% say they want to use AI agents to shop. These are not early adopters playing around with the shiny new object. They are mainstream consumers who’ve found something better than what they had and voted with their behavior. Those habits are sticking.

What the past year revealed is that the gap between consumer readiness and commercial infrastructure is wider than most of the predictions of 2024 acknowledged. The technology moved faster than anyone anticipated. The commerce plumbing did not.

Merchant agreements, live inventory access, payment rails connected to the conversation, governance frameworks, return policies, tax infrastructure, the whole invisible machinery that makes a purchase feel effortless rather than terrifying. None of that gets rebuilt in a product cycle.

That doesn’t mean it will take decades. The pace of AI development alone collapses the timeline. The models are improving every few months, the protocols are being written, the coalitions forming, the early experiments producing hard-won, even painful, knowledge about what the market actually requires. And all of it is laying the foundation faster than any prior generation of commerce infrastructure was built.

Read More: AI Doers Drown Out AI Naysayers

The question isn’t whether the loop closes. It is who closes it, under what terms, and who is left holding the better position when it does.

This is where time matters in a way the optimists tend to underestimate. Igniting commerce networks is more science than art, more strategic than wishful thinking. Every merchant recruited makes the next recruitment easier. And more consumers more likely to give it a try. Every consumer transaction deepens the trust that makes the next one more likely. Every improvement to fulfillment raises the bar that competitors must clear. Getting to that point is a tedious slog

The players who are building real infrastructure now, actually solving the hard coordination problems, are accumulating advantages that will be very difficult to displace once the market tips.

The opportunity on the other side of this infrastructure gap is unlike anything commerce has seen since Amazon proved that consumers would trust a website with their credit card if the experience was reliable enough and the selection was wide enough.

The gap between consumer readiness and commercial infrastructure is wider than most of the predictions of 2024 acknowledged.

What it produced is a marketplace that accounts for more than 60% of all online sales and a Prime membership that reshaped how an entire generation thinks about buying things. Not to mention the analog for how online transacting must behave. Agentic commerce, done right, with an agent that is genuinely aligned with the buyer, represents a comparable reset. Not an incremental improvement on search. A different model entirely.

Read More: How Time Became the Next Great Asset Class

My Toast Test still disappoints. The answers are better but still missing my go-to brand. Yes, I am picky. And the AI still can’t buy it for me. But the day that changes is the day I’ll buy, and then do it again, and build the trust that, next time and the time after that, an agent can do all of that for me.

Until then, agentic commerce remains just a smarter search bar. A better one than what came before. And one that I will continue to use that way.

But a search bar, nonetheless.

 

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.

Top of wallet.

Issuers built elaborate rewards architectures to win it. Travel points. Cash back ladders. Category bonuses. Limited time offers designed to nudge one card ahead of another in the consumer’s mental stack.

But the data tells a different story about what consumers are actually trying to do when they reach for a card.

Many aren’t trying to maximize rewards. They’re trying to manage their money.

Every household in America is running a balance sheet, whether they think about it that way or not. Paychecks arrive on one schedule. Bills arrive on another. Unexpected expenses show up whenever they feel like it. Consumers are constantly managing the gap between those three things, sometimes consciously, often on autopilot.

The credit industry has spent decades competing for something consumers were never actually optimizing for.

Credit cards, Buy Now Pay Later, store cards, debit-linked installments, promotional financing. None of these are competing products in the consumer’s mind. They’re tools. Different tools for the same job. Keeping the household financial engine running.

The industry built an ecosystem around winning a choice that consumers weren’t ever really making the way the industry assumed they were.

And now, just as intelligent agents are about to take over the payment decision entirely, that assumption is about to be cracked wide open.

The Consumer Is Already There

The conventional narrative frames credit as a battle between BNPL upstarts and “old school” credit cards. Disruption versus legacy. New versus old.

The data doesn’t support that story.

PYMNTS Intelligence research of a national sample of 2,980 consumers in January 2026 shows that 31% of U.S. consumers used credit card installment plans in the prior three months. Just 12% used BNPL. Seven in 10 used a general-purpose credit card for at least one purchase.

That three-to-one “Buy Now Pay Later” gap has held across multiple survey waves and across every generation, including Gen Z, the generation BNPL was purpose-built for. More than four in ten Gen Z consumers report using credit card installments. Millennials show a similar pattern.

If consumers were chasing the newest, shiniest thing at checkout, BNPL would dominate. It doesn’t.

Consumers are choosing options to best manage their cash flow.

Why? Because consumers aren’t chasing novelty. They’re optimizing for flexibility at checkout and choosing options to best manage their cash flow. Installments inside a credit card account let consumers restructure a purchase into predictable payments, against an already approved credit line, while keeping the broader credit relationship intact. BNPL does that too, especially for smaller purchases or consumers who want to keep their credit lines available for emergencies. But it’s one tool in a kit, not the whole kit.

Consumers behave like sophisticated financial managers. The industry just hasn’t thought of consumers and their use of credit in much the same way.

The Myth of Top of Wallet

Here’s where the top-of-wallet strategy goes sideways in an agentic world.

Most consumers don’t have one card. They have a portfolio. A high-limit card for working capital flexibility. A corporate card for travel. A store card for promotional financing. A BNPL account used like a debit card for everyday essentials and smaller purchases that don’t need to touch existing credit lines. They move between these tools depending on the size of the purchase, where they are in the pay cycle, and what the merchant is offering. Which card has open credit to buy? Which one has the promotional rate? Is it worth opening a  a new account to get the 10% off at checkout?

Top of wallet becomes irrelevant when the agent doesn’t care which card is on top.

The industry called winning that mental game “top of wallet.” Rather than decide, just default to the card that does the best job most of the time.

That assumption is about to be upended. Once intelligent agents start executing payment decisions, they won’t optimize for habit or brand loyalty, unless the consumer puts it in the prompt. They’ll optimize for the best financial outcome. The best rate. The best terms. The best fit for the consumer’s finances at that specific moment.

Top of wallet becomes irrelevant when the agent doesn’t care which card is on top.

Consumers Already Built the Stack

Long before anyone in the industry started talking about agentic commerce, consumers had already jerry-rigged their own manual version of it.

From the industry’s perspective, credit cards, installment features, store cards, BNPL and merchant promotional financing compete with each other. From the consumer’s perspective, they solve different problems. A revolving line absorbs unexpected expenses and bridges pay cycles. Card installments convert big purchases into structured payments without opening a new account. BNPL typically handles smaller-dollar purchases with predictable repayment. A grocery purchase late in the pay cycle goes on one account to preserve cash. A furniture purchase gets converted to an installment plan. A merchant promotion routes the purchase to embedded financing.

Consumers are already doing sophisticated cash flow routing.

Consumers are already doing sophisticated cash flow routing. They’re just doing it manually, with incomplete information, under time pressure.

What looks like fragmentation to the industry is a consumer’s personal credit stack. The next phase of the ecosystem won’t replace it. It will automate it.

One Credential, Many Jobs

Here is where the real disruption sits.

Think about what a smart credential could actually do. A single account could create access to the credit optionality that consumers already collect in their physical or virtual wallets. And at the moment of purchase it could evaluate every available credit alternative and pick the right one for that moment in time. Pay now and take the cash back. Convert the purchase to an installment and protect short-term cash flow. Apply the merchant’s promotional financing and pay nothing for six or 12 months.

Draw on a revolving line when flexibility matters more than cost. Sort of like a credit waterfall, using real-time underwriting and intelligence intended to improve cash flow for the consumer.

The same checkout moment. Completely different economics depending on what the consumer needs and what the merchant, issuer or brand is willing to offer to close the sale.

This is not a thought experiment. The building blocks already exist.

This is not a thought experiment.  The building blocks already exist. Issuers already embed installment features inside credit card accounts. Merchants already offer promotional financing to drive conversion on big-ticket purchases. Acquirers are enabling agentic underwriting for BNPL players to make more predictive credit offers and decisions. Card networks are enabling issuers to create flex accounts that turn purchases into “pay now, a little later or a lot later” based on user-established parameters. BNPL players are enabling access to their rails and merchant networks for existing banks and credit union debit cards.

The next step is having a credential that becomes smart enough to assemble those options in real time, present the best one, and execute it without the consumer having to figure it all out.

So, What Happens to Rewards in an Agentic World?

Rewards programs were built on a single promise. Give consumers something they value, and they will change their behavior. Carry this card. Use it here. Reach for it first. The points and miles and cash-back percentages were about influencing behavioral preference at scale.

If agents make the payment decision, that behavioral lever largely disappears. The agent isn’t susceptible to aspirational travel advertising. It doesn’t have a favorite airline or a loyalty to a particular hotel chain, unless it is mentioned in the user prompt. It runs the numbers.

That doesn’t mean rewards go away entirely. It means they have to earn their place in the decision differently. The rewards programs that survive this shift will be the ones that convert from lifestyle marketing into real-time financial value that improves cash flow for the consumer.

For merchants, the agentic payment model could become the most direct line to conversion they’ve ever had. Today, merchants spend billions trying to influence consumers before they get to checkout, through advertising, promotions, loyalty programs, and brand building that may or may not move the needle. Most of that spending happens before the consumer has decided to buy, which means most of it simply hopes for the best, even with using the best of targeting tech.

If the agent is making the payment decision based on what produces the best outcome for the consumer, merchants have a direct path to influencing that decision through offers embedded in the transaction itself. A promotional rate that makes a big purchase affordable. An instant rebate that improves the consumer’s cash position today. A financing structure that converts a hesitant browser into a buyer. Rewards that show up as real money at checkout rather than as aspirations to be redeemed someday.

Brands get a version of this too, and something even more valuable on the side: first party data on what actually drove the purchase. Not survey data. Not modeled attribution. Real transaction data showing which offer, at which moment, converted which consumer. For brands that have spent years trying to measure whether their marketing actually works, that is worth considerably more than the cost of funding the offer.

Merchants bid to be the most attractive option at the moment of payment.

The issuer that builds that credential becomes the platform that sits between merchant intent and completed transaction. Think of it as the Google AdWords model applied to checkout. Instead of bidding to appear at the top of a search results page, merchants bid to be the most attractive option at the moment of payment, funding offers that make their product the one the agent selects.

Google built one of the most profitable businesses in history by owning that auction dynamic for attention. The issuer who owns it for transactions is sitting on something just as valuable. And unlike a traditional rewards program, this model turns merchant and brand marketing budgets into revenue. That is a very powerful place to sit.

The Cash Flow Buffer Is the Strategy

Total consumer credit outstanding has surpassed $5 trillion. The usual headlines call it a debt crisis. Look more carefully and you see something else.

Revolving credit growth has slowed, but more consumers are maintaining balances rather than paying them down completely. That’s not a sign of financial distress. It’s a sign of deliberate financial management. Consumers are keeping credit available, maintaining a buffer they can draw on when something unexpected happens. In corporate finance, you’d call that maintaining access to a working capital facility. The household version looks exactly the same.

Consumers are keeping credit available.

PYMNTS’ Consumer Expectations Index shows a roughly 20-point confidence gap between financially stable households and those living paycheck to paycheck. That gap isn’t just about sentiment. It has real-world implications.  Consumers with cash flow constraints rely more heavily on credit tools that help them smooth cash flow between pay periods.

Which means this isn’t niche behavior or a sign of financial mismanagement. Managing cash flow with credit is what everyone is doing, all the time, across every income level and every generation. Whether that person is living paycheck to paycheck or not.

Why the Future of Credit Is Cash Flow

Consumers have been running their own cash flow optimization engine for years. They do it with imperfect tools and incomplete information. They do a pretty good job of it despite the friction.

In an agentic commerce world, what’s changing isn’t the consumer’s job. The household CFO isn’t getting replaced.

What’s changing is the assistant they now have to help them.

The next era belongs to the credential they never have to think about.

For decades, the credit card industry competed to sit at the front of the wallet. Rewards were lures. Win the top of the stack, and you win most of the spend.

Agents will call balls and strikes differently. Based on how well they improve the cash flow position of the consumer making the purchase. Not where a card is in the wallet stack.

The industry spent decades competing to be the card consumers remembered. The next era belongs to the credential they never have to think about, because it is always doing the thinking for them.

 

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.

There is a young woman in Chicago who earns money five different ways before noon on a Tuesday. She drives a rideshare app in the morning, drops a delivery on the way home, sells a handmade item from her Etsy shop, completes two micro-tasks on a gig marketplace and picks up a two-hour hospitality shift that evening. She has no payroll department, no direct deposit hitting every other Friday. What she has is a continuous stream of small disbursements, real-time transfers that function, collectively, as her paycheck.

She is not an outlier. She is the emerging archetype of the American worker.

I’ve started calling what she represents the transactional worker, operating within what we at PYMNTS Intelligence now call transactional payroll. It’s an income architecture defined not by schedules but by transactions, not by employment status but by output, not by the calendar but by the clock.

The data tells a story that most corporate payroll departments, legacy banks and enterprise HR platforms are only beginning to recognize. We’ve surveyed more than 60,000 U.S. consumers across multiple waves of research to document it.

The way tens of millions of Americans earn money has fundamentally changed. The way the financial system pays them has not kept up.

Read More: From Payroll to Pay Now: How Real-Time Earnings Are Rewriting Work

This is not a payroll innovation story. It is evidence of a deeper structural change: the shift from a batch-based financial system built around periodic wages to a transactional economy built around continuous income flow in a real-time economy.

The Scale of the Shift

According to PYMNTS Intelligence research commissioned by Ingo Payments, roughly 21% of all U.S. disbursements recipients in 2020 most often received their payouts instantly. By May 2025, that share had nearly tripled to approximately 41%.

But the headline figure spans every category of payout. Insurance claims, gaming winnings, loan proceeds, investment returns. The story that matters for the future of work lives inside one specific slice.

Wages.

When we isolate transactional payroll, the disbursements workers rely on as earnings and income, the shift is even more pronounced. Today, 46% of U.S. workers report receiving payouts instantly most of the time. Among those who depend on earnings from gig platforms, tipped shifts, construction sites and long-haul routes, the demand isn’t a preference. It is a structural requirement.

The cohorts driving this shift aren’t hard to identify.

Among Gen Z workers, 84% received at least one instant disbursement in the past year, and 40% now rely on instant methods more than any other payout channel. For this generation, waiting several business days for earnings to hit their account doesn’t feel like a minor inconvenience. It feels like a design failure, a relic of a system built for a different kind of work.

Four Use Cases Where This Hits Hardest

The wage use cases driving this shift are concrete. Our research has examined the specific industries where transactional payroll is most acutely felt, and where the gap between when work is done and when money arrives creates the most friction.

Four stand out.

Construction. Construction workers, particularly day laborers, trade subcontractors and project-based crews, have long operated in an environment where the gap between work completed and wages received can stretch days or weeks. Paper checks and ACH batch cycles are still the norm on many job sites, creating liquidity pressure for workers carrying out-of-pocket costs for tools, transportation and materials before a paycheck clears.

Our research found that construction is among the industries where instant wage access produces the most immediate lift in worker retention and shift acceptance, precisely because the cash-flow timing risk is highest.

Read More: Real-Time Payments Give Contractors Back 2 Workweeks a Month

Hospitality and Tipping. As digital payments have replaced cash at restaurants, hotels and event venues, tip disbursement has become a genuine operational problem. Workers who once left a shift with cash in hand now wait days for tips to be batched, processed and cleared. Our research tracked hospitality as one of five key industries rapidly adopting instant payment rails to resolve this mismatch.

Among tipped workers, 65% say they need their earnings instantly or within a single day. This reflects the financial architecture these workers live in, where the shift-by-shift tip cycle is the primary income cadence and delayed access disrupts every downstream obligation.

Transportation and Trucking. Trucking represents one of the clearest cases for per-trip, per-load transactional payroll. Owner-operators and independent drivers carry upfront costs such as fuel, tolls and maintenance before they’re paid for the haul. When settlement takes days, those costs create working capital gaps that small operators cannot easily absorb. Instant disbursements serve a dual purpose here: they reduce driver turnover for carriers and fleets, and they eliminate the paper check dependency that has historically characterized the industry.

Gig Platforms. The gig economy remains the most digitally advanced sector for instant wage disbursement. Nearly six in 10 payouts made by corporate senders to consumers for gig work now move via instant methods, the highest rate of any industry in our research. But the critical insight is that gig platforms aren’t the outlier. They are the template. The same logic driving instant wage adoption in the gig economy is now migrating into property management, event staffing, healthcare support, retail and logistics.

Read More: Why Gig Economy Companies See Payables As Key To Success

Many of the workers across these four use cases sit within what we call the Labor Economy: approximately 60 million U.S. workers concentrated in hourly, hands-on roles across logistics, hospitality, retail, construction and healthcare support.

These workers account for nearly 37% of the entire U.S. workforce and drive more than $1.7 trillion in annual consumer spending. But their savings sit roughly 40% below the national average. When their income is delayed or unpredictable, the consequences don’t stay contained to their households.

A 1% wage change across this population translates to approximately $17 billion in GDP impact. That is not a rounding error. That has a non-trivial economic impact.

Transactional Payroll: A New Architecture of Earning

It’s worth pausing on what transactional payroll actually means, because it captures something that the phrase “gig economy” doesn’t fully convey.

Transactional workers are not workers who are scraping by, stuck between real jobs or failing to find traditional employment. In many cases, they’ve actively chosen a different relationship with income, one built around autonomy, flexibility and the ability to match earnings to obligations in real time.

For one-third of millennials, transactional work is now the primary source of cash flow, not a supplement to a traditional salary. For roughly half of Gen Z, core income arrives through online selling, freelance services or platform-based work. These workers are not fringe cases of the labor market. They are increasingly its center.

Read More: Instant Payouts Become the New Paycheck in a Real-Time Economy

And the financial system’s continued insistence on treating them as exceptions, routing their earnings through legacy payroll cycles, ACH batch processes and clearing windows designed for a different era, is creating friction with measurable economic consequences.

Consider the math. A worker who earns $200 on a Tuesday and can’t access those funds until Friday or the Friday after that isn’t merely inconvenienced. That worker may pay a late fee, miss a payment, borrow at high cost, or decline a subsequent shift because they can’t front the gas money to take it. Each of those outcomes has a price, to the worker, to the platform that depends on their availability, and to the broader economy that depends on her spending.

Our behavioral data, spanning multiple waves of study across the U.S. workforce, makes the stakes tangible.

Among gig workers, 54% say they need same-day access to their funds. Among tipped workers, that urgency climbs to 65%. Freelancers and contractors, whose work tends to follow longer billing cycles, still show meaningful urgency, with 48% requiring same-day access once work is completed.

For these workers, real-time access to pay is not a nice-to-have. Given thin savings, variable income and fixed obligations, same-day access is a liquidity requirement.

The Infrastructure Gap

If the demand for transactional payroll is so clear, why does the infrastructure to support it remain so incomplete?

The answer isn’t primarily technology. The rails for instant payments exist and are operational. The issue is adoption, incentive and institutional inertia on the sender side.

Read More: Fee Sensitivity and the Opt-In Economics of Instant Payouts

Despite overwhelming worker demand, only 36% of gig platforms currently offer instant payments consistently. The remaining 64% offer them only sometimes, rarely or never. That gap represents both a significant friction point in the labor market and a significant commercial opportunity for providers and platforms willing to close it.

The adoption data reveals something important about how workers respond when platforms do offer instant payments. It is not gradual. When instant payouts become available, 59% of all disbursements on that platform go instant almost immediately. Workers don’t need to be educated or incentivized. They’ve been waiting for the option.

And the stickiness is equally compelling.

Once a consumer receives an instant payment, 57% make it their primary payout method, up from 39% in 2020 and continuing to rise. For transactional payroll workers specifically, that stickiness reaches 68%. Once the expectation of real-time access is set, it does not reset.

The competitive implications are direct. Payout speed is a platform selection criterion. Workers operating across multiple transactional platforms, which is increasingly common, allocate their time toward platforms that pay faster. A platform that pays on Friday competes at a disadvantage against a platform that pays Tuesday night.

Solving for Employer Liquidity

There is a structural tension in the move toward on-demand pay that the business case for instant disbursements does not always address directly. Traditional biweekly payroll is not merely an administrative convention. For employers, it is a working capital tool. When wages are batched into a single disbursement every two weeks, the employer retains the use of that capital during the intervening period. Moving to on-demand disbursement replaces that predictable float with a continuous, variable draw on operating cash. Smaller platforms, staffing operators and hospitality groups may struggle to absorb that shift.

This is not a theoretical concern. It is one of the most consequential and underappreciated reasons that the adoption gap in instant payroll persists. Employers and platforms that want to offer real-time pay often find themselves caught between genuine worker demand and real treasury constraints. Solving for one without accounting for the other produces a half-measure that the market won’t sustain.

The model gaining traction, and the one that unlocks widespread adoption without forcing employers to restructure their working capital management, is the earned wage access intermediary. In this architecture, a third-party provider advances earned wages to workers in real time, fronting the capital the employer has not yet disbursed. The employer continues to settle on its existing payroll schedule. Workers receive access to funds when work is completed. The timing gap is bridged, but the employer’s cash flow cadence is preserved.

Read More: From Perk to Necessity: On-Demand Pay Brings Predictability to Paychecks

This decoupling is the key structural insight. Workers get real-time access; employers retain their working capital rhythm; and the intermediary earns revenue through a combination of small per-transaction fees, subscription arrangements with platforms, or employer-funded access offered as a workplace benefit. In the employer-funded model, the cost is typically offset by measurable reductions in turnover, higher shift acceptance rates, and improved worker availability. These outcomes carry their own return on investment.

The model also resolves a credit and underwriting question that has historically slowed adoption. Because the capital being advanced represents wages already earned rather than credit extended against future income, the risk profile is fundamentally different from consumer lending. The intermediary is not underwriting the worker’s ability to repay. It is bridging a timing gap on income that already exists, and recouping from the employer’s scheduled payroll settlement. The advance is not a loan. It is an acceleration.

What has lagged behind the model is integration. Earned wage access capabilities should be embedded directly into the payment infrastructure that platforms use to manage disbursements so that instant pay becomes a default feature rather than a separate product the worker must seek out. The platforms and financial institutions that achieve that integration, and offer it at the point where work is completed rather than as an add-on enrollment flow, will define what modern transactional payroll infrastructure looks like for the decade ahead. The employer liquidity problem is solvable. The business model to solve it already exists. What remains is execution.

The Real-Time Cash Flow Mindset

The behavioral shift among transactional workers isn’t just about speed. It reflects a fundamentally different relationship between earning and spending, one in which income management is an active, continuous practice rather than a passive, periodic event.

PYMNTS Intelligence research into how transactional payroll workers describe their own financial behavior reveals a pattern economists and financial planners would recognize as sophisticated real-time cash flow management.

These workers don’t smooth income across a pay period. They match income to obligations directly, often on a same-day or next-hour basis. They’ll describe choosing a particular shift because they need money tonight, or completing a specific delivery run because a bill is due in the morning.

Read More: America’s Workers Got Left Behind in the FinTech Boom

That mindset creates a set of product requirements that traditional banking and payroll infrastructure was not designed to meet. The worker managing cash flow in real time needs real-time visibility into earnings, real-time access to those earnings and financial tools that operate at the same cadence as their income.

The generational profile matters here. Workers who have grown up in this real-time income environment, primarily millennials and Gen Z, have not experienced traditional biweekly payroll as a default. For them, the expectation that income arrives when it is earned is not a request. It is an assumption.

That product requirement is now getting a serious answer from both banks and FinTechs. As I wrote at the start of the year, Buy Now, Pay Later is stepping in as an emerging working capital tool for the modern middle class, migrating out of its original discretionary-spending niche and into the everyday financial plumbing of households that live close to the edge of their checking accounts.

Read More: BNPL’s Next Act Is as Consumer Working Capital

For transactional workers managing variable income against fixed obligations, the fit is intuitive. BNPL’s installment structure lets workers authorize a necessary expense today and match its repayment to income they know is coming, without incurring the penalty economics of an overdraft. It converts the timing gap from a crisis into a schedule.

I see BNPL as a direct complement to the real-time payment rails that transactional payroll depends on. Instant disbursements get money into workers’ hands faster; BNPL tools help them extend that liquidity intelligently across their obligations.

The Side Hustle as Financial Infrastructure

The phrase “side hustle” entered mainstream vocabulary as a way to describe supplemental income, something workers did in addition to a primary job. That framing no longer captures reality for a significant and growing share of the workforce.

Recent PYMNTS Intelligence data, in collaboration with WorkWhile, shows that nearly one in five hourly workers now reports regular side work. And critically, the majority of that income is used not to build savings or fund discretionary spending, but to cover essentials: rent, groceries, utilities, transportation. What was once supplemental has become foundational.

Read More: Wage to Wallet Index: Side Work Patterns in the Labor Economy

This has real consequences for how the financial system should think about risk, liquidity, and the definition of a customer. A worker who earns $42,000 a year across three platforms and a tipped hospitality job is not a gig worker in the colloquial, marginal sense.

That worker is running a complex, multi-channel income operation that requires the kind of financial tooling that used to be reserved for small business owners. And risk profiling and underwriting that reflects their financial reality.

The financial stress that characterizes this population is not primarily about the level of their earnings. These workers are not earning nothing. They are earning enough, or close to enough, but that income is arriving at the wrong time, in the wrong cadence, with the wrong access characteristics. The stress is liquidity stress, not income stress. It is the stress of having earned money that is not yet available.

PYMNTS Intelligence research has found that the gap between salaried and hourly workers in terms of financial confidence and perceived mobility is driven less by income level than by access to liquidity tools.

Salaried workers experience the economy as a set of expanding options because they have predictable, smoothed income that gives them a planning horizon. Hourly and transactional workers, even those earning comparable amounts on an annualized basis, experience it as a treadmill, because each pay gap is a potential crisis.

Late fees, overdraft charges and clearing delays consume a disproportionate share of transactional workers’ income. A $35 overdraft fee on a $200 paycheck is not a minor inconvenience. It is a 17.5% penalty on earned wages, imposed not because the worker is financially irresponsible but because the infrastructure around them doesn’t sync with their reality.

This is precisely why I believe BNPL will become a structurally sound replacement for overdraft and late-fee dependency. Overdrafts are backward-looking: the penalty arrives after the worker has already stumbled. BNPL is forward-looking: the obligation is priced, scheduled and disclosed before the transaction is authorized. For transactional workers running their finances like a small business, that predictability will become a basic planning tool.

What the System Owes the Transactional Worker

The business case for faster payments to transactional workers is well established.

Platforms that pay faster attract more workers, retain them longer and compete more effectively for labor supply. Workers who receive instant payouts are far more likely to continue using the same platform. Workers who are not managing acute financial stress make fewer avoidable decisions, fewer missed shifts, fewer last-minute cancellations. And workers who are paid in real time have more money to spend in the real economy, because they’re not handing it over in late fees and penalties.

Read More: Instant Payouts Pull Workers Toward Rival Platforms

But the business case, while important, is not the whole argument. There is something more fundamental at stake in how the financial system treats earnings that belong to workers who have already done the work.

For the transactional workforce, timing is not an administrative detail. It is the difference between financial stability and financial fragility. The same earnings, arriving one day sooner, produce materially different outcomes for a worker managing thin savings against fixed obligations. The technical capability to deliver those earnings in real time exists. The rails are built.

The question is whether the institutions and platforms that sit between the worker and their money will treat that delivery as a baseline expectation or a premium feature.

The woman managing five income streams before noon on a Tuesday is already functioning as her own CFO. What she needs from the financial system is a product stack that matches that sophistication. Instant disbursements on the earning side, and intelligent short-term liquidity tools on the spending side. BNPL, offered through the debit card tied to the checking account where her income lands, gives her a way to smooth obligations without incurring the blunt-force penalty of an overdraft.

Those who bring these tools together, pairing real-time payroll with embedded installment options, will define what transactional banking looks like for the next decade.

For the banks and FinTechs that see around that corner, the upside is significant. Workers who experience real-time access to their earnings form durable relationships with the platforms and financial tools that provide that access.

Loyalty, it turns out, moves at the speed of money, too.

 

Until NEXT time.

Join the 19,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.

In March 2020, government officials and epidemiologists were telling Americans to brace for a two-month shutdown. PYMNTS Intelligence went into the field and asked 1,923 consumers directly what they thought. They said five months. In May 2020, the experts said recovery would come by fall. Consumers said February 2021. In December 2020, when officials were projecting a return to normalcy by March 2021, consumers put it at January 2022. They were right every single time.

And no, this wasn’t luck. These weren’t epidemiologists with access to better models. And they didn’t have Dr. ChatGPT to help advise them.

They were ordinary people doing something that spreadsheets can’t. They were factoring in their own fear, their changed habits, their distrust of public spaces and the fact that no one had given them a reason to go back. They knew where they were going to spend: online, away from the physical world. And they knew it was going to be for a lot longer than anyone in Washington or on Wall Street wanted to accept.

Most of those habits stuck.

The inflation story has played out the same way. When the Fed and most economists were calling price pressures “transitory,” PYMNTS Intelligence consumer surveys were already telling a different story. In late 2022, consumers predicted that inflation wouldn’t return to pre-2021 levels until late 2024. And then pushed that forecast out further with each passing month of data. Again, they were closer to right.

And then there are tax refunds.

Forecasters and retailers reliably predict that every spring consumers will use their refunds on discretionary purchases. A little treat, something nice. PYMNTS Intelligence data kept showing otherwise.

If consumers have been right, repeatedly, ahead of schedule and across multiple economic cycles, why does conventional wisdom about consumer sentiment keep getting blindsided?

Among households living paycheck to paycheck and struggling to pay their bills, which represents about 68% of American households, roughly two-thirds of any refund goes straight to everyday expenses or debt repayment. About 16% save or invest it. That “treat yourself” shopping spree doesn’t happen for most American families.

So, here’s the obvious question. If consumers have been right, repeatedly, ahead of schedule and across multiple economic cycles, why does conventional wisdom about consumer sentiment keep getting blindsided?

The answer may be that the instruments we use to measure sentiment were designed for an economy that no longer quite exists.

Three Blind Spots in How We Measure Consumer Sentiment

The foundational index for measuring U.S. consumer sentiment goes back to 1946, when University of Michigan psychologist George Katona started surveying 500 consumers monthly about their economic outlook, not just their spending behavior. It was a real methodological breakthrough. For the first time, economists had a systematic way to track the mood behind the money movement. The University of Michigan Index of Consumer Sentiment has been doing that job ever since.

But the economy has changed in ways the original methodology couldn’t have anticipated. Three structural gaps have opened up between what the classic measures capture and what’s actually driving consumer behavior today.

Blind Spot 1: The Damage From Inflation Was Permanent, Not Transitory

Economists will point to the data and say that inflation has cooled, wages have recovered, unemployment is low. On the surface, things look fine. So why do so many people still feel squeezed?

Because prices never went back down.

When experts say inflation “slows,” that means goods are getting more expensive at a slower rate, not that they’re getting cheaper.

When experts say inflation “slows,” that means goods are getting more expensive at a slower rate, not that they’re getting cheaper. Groceries, rent, utilities, and car insurance are  still dramatically more expensive than they were in 2020. A household that budgeted around 2019 prices, with no extra savings to cushion the blow or with wages that haven’t kept pace, has absorbed years of cumulative damage that the current CPI reading glosses right over. CPI doesn’t reflect the prices people actually remember paying.

READ MORE: Why Consumers Don’t Care About Monthly CPI – and Why it Matters

Traditional sentiment surveys ask consumers whether things are getting better or worse right now. They’re not designed to capture how deep the hole already is after a multi-year price reset. A consumer whose grocery bill has climbed 25% over four years, and whose wages recovered slowly, isn’t confused when they say they still feel financially stressed. They’re accurately describing their situation. The survey instrument is the one missing the finer points.

This is why PYMNTS Intelligence data kept showing consumers pushing their inflation recovery timelines out further, even as pundits were declaring victory. Consumers weren’t ignoring macroeconomic data. They were describing a lived experience that macroeconomic data simply wasn’t built to capture.

Blind Spot 2: Income Has Stopped Being a Reliable Predictor of Spending Behavior

Most economic models are built on a sensible assumption: higher income means more spending. In a world where household balance sheets are stable and monthly obligations are modest relative to what people earn, that holds up.

But that relationship has weakened. And it’s become far more conditional on liquidity, debt load and fixed monthly costs. All these things determine whether a paycheck actually has room in it, regardless of the number on the stub.

In December 2021, PYMNTS Intelligence found that 61% of Americans reported living paycheck to paycheck. By August 2025, that figure had climbed to 71%. What makes this more than a poverty story is who’s inside that number. Nearly half of people earning over $100,000 a year now report the same condition, stretched thin by fixed obligations, debt service and the cost of maintaining a lifestyle that looked affordable before the inflation surge.

It’s the structure of someone’s financial life that determines whether there’s actually room for discretionary spending.

Income still matters, of course, but it’s the structure of someone’s financial life that determines whether there’s actually room for discretionary spending. Whether their fixed obligations leave any margin. Whether an unexpected expense would mean going into debt.

Two people earning identical salaries will make completely different spending decisions if one has three months of savings and the other has none. The number on the pay stub looks the same. The behavioral response to any economic signal does not.

This also explains why the tax refund narrative keeps failing. For a household with a financial cushion, a $2,000 refund is a windfall, something to enjoy. For a paycheck-to-paycheck household, that same $2,000 is a partial offset against overdue bills and high-interest debt. Aggregate income models can’t tell those two households apart. Their spending decisions couldn’t be more different.

Blind Spot 3: Employment and Job Security Are Two Entirely Different Things

A low unemployment rate tells you what percentage of people who want jobs have them. It tells you almost nothing about how secure those people feel in the jobs they hold, whether they believe they could find comparable work if they were let go, or whether technology is quietly eroding the value of their skills in ways that haven’t shown up yet in anyone’s data.

That distinction matters enormously for how people spend.

Decades of research consistently shows that people begin pulling back the moment income feels uncertain, often well before anything bad actually happens. It’s not job loss that triggers spending contraction. It’s the anticipation of possible job loss and the perceived difficulty of replacing it that does. By the time a drop in consumer spending shows up in sales data, the behavioral shift has often been underway for weeks or months.

It’s not job loss that triggers spending contraction. It’s the anticipation of possible job loss.

There’s also a specific kind of worker who represents a real but underappreciated vulnerability here. Someone who feels personally secure in their current role, but who doesn’t believe they could quickly find equivalent work if they had to. They’re not spending freely. They’re spending cautiously, keeping one eye on an exit they’re not sure exists. A headline employment rate captures none of that.

The rise of AI in the workplace makes this more urgent.

Workers whose skills are adjacent to automation are experiencing something with no real precedent in postwar labor economics: a slow, uncertain degradation of occupational value that doesn’t trigger unemployment benefits, doesn’t show up in job-loss statistics and doesn’t register on any existing confidence measure. But it’s absolutely shaping how they think and spend.

A New Instrument for a Different Economy

These three gaps are what drove the development of the PYMNTS Consumer Expectations Index, or PCEI.

On March 2, 2026, PYMNTS Intelligence will launch the PCEI, a monthly, census-balanced survey of more than 2,000 consumers designed to measure U.S. consumer sentiment in a way that’s directly useful for business decision-making. PYMNTS Intelligence Senior Analyst Matt Albrecht has led this effort, bringing three years of experience overseeing the Florida Consumer Sentiment Index at the University of Florida’s Bureau of Economic and Business Research.

The PCEI isn’t meant to replace the University of Michigan Index or the Conference Board Consumer Confidence measure. Those instruments have decades of data behind them and serve real purposes. What the PCEI does is go deeper, and specifically on the structural constraints and risks that classic measures don’t capture as directly.

It builds on traditional sentiment measures such as household finances, economic outlook and purchase climate while extending into the structural constraints that determine whether consumers can actually act on their confidence: debt manageability, savings capacity, emergency readiness and labor-market security. The index maps these across 11 dimensions on a 0 to 100 scale, with 50 as the neutral baseline.

The core idea is straightforward.

Sentiment only drives behavior when households have room to act on it.

Sentiment only drives behavior when households have room to act on it. Optimism without financial capacity isn’t really optimism, it’s a feeling that won’t translate into spending. The PCEI is designed to capture both how consumers feel and how free they actually are to follow through on those feelings.

The framework draws on more than six years of PYMNTS consumer research, spanning hundreds of thousands of respondents across dozens of studies. That body of work is where the pattern became undeniable. Consumers aren’t confused about their own situations. They’re accurately describing conditions that existing measures weren’t built to detect.

What the February Data Says

The headline number for February is 51.5, which is technically above the neutral reading of 50, technically pointing toward cautious optimism. Stop there, though, and you miss the real story.

The more revealing finding is where that aggregate number breaks apart. The spread between consumers who live paycheck to paycheck and those who don’t is more than 10 points. The spread between the highest and lowest generational cohorts is about 7 points.

In a single data point, that comparison captures the whole thesis of the PCEI.

Generation explains how high you’re sitting. Financial structure explains how far you can fall.

The job security numbers deserve a close look because they illustrate the blind spot perfectly. Workers score their personal job security at 83.5. which is solidly positive. They’re not lying awake worrying about layoffs next month. But when asked how quickly they could find a new job at the same pay, that score drops to 48.0, just below neutral.

Secure in place. Not confident about what happens if they need to move.

Today’s spending is holding up not because consumers feel genuinely resilient (a word that the media bandies around freely), it’s holding up because they feel safe staying put. The moment that changes, the calculus shifts fast.

The takeaway isn’t panic, but another measure of fragility. Consumers can feel secure enough to stay in place while still lacking the confidence to take discretionary risks. That’s how a headline number can look stable while the underside remains tight.

Debt confidence at 71.4 is the strongest reading in the index, and it’s worth being precise about what that does and doesn’t mean.

Consumers feel they can manage what they owe, which is actually pretty good in an environment where credit card balances have been climbing steadily. But current financial conditions score only 51.5. People feel like they’re keeping up with obligations. They’re less sure they’re actually getting ahead. That’s stability through management, not through progress. And there’s a ceiling on that kind of confidence.

Generationally, Millennials lead at 60.7, the most optimistic cohort by a meaningful margin. Baby Boomers and Seniors sit at 53.5, closest to neutral. But the more notable pattern is how synchronized the movement has been across all generations. Every cohort softened in November, rebounded in December, eased in January, and improved again in February. The same rhythm, at different altitudes.

When Consumers Know Best

February’s data describes a consumer who is holding on. They’re managing debt, keeping their job, not panicking, but doing so carefully, selectively and with a clear eye on what could go wrong.

Their resilience is real. So is the shaky foundation underneath it. Something that has less to do with what people earn than with how their financial lives are structured.

Whether they have a cushion. Whether their obligations leave room to maneuver. Whether one unexpected expense tips the balance.

Consumers feel the constraints in real time.

Consumers told us the pandemic would last longer than the scientists projected. They told us inflation wouldn’t vanish on the Fed’s timeline. They told us the tax refund wouldn’t become a shopping spree. They were right every time. Not because they’re better forecasters, but because they feel the constraints in real time.

When permanent structural shifts and short-term policy uncertainty are reshaping how people think and spend at the same time, asking only how the consumer is “feeling” may not be the most useful question.

The more practical one is how well the consumer is positioned to act on those feelings.

The data suggests consumers have understood their own situation clearly all along. The question has always been whether the tools used to measure them were designed to actually listen.

 

Until NEXT time.

Join the 19,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.

Fifty years ago, Susan was one of more than a hundred million Americans who drove to a shopping mall an average of once a week to buy things. By 1976, a third of all retail sales happened at the mall.

Susan shopped exactly the way the mall designers intended. The department store anchors were the magnets motivating her to make the trip in the first place. The smaller shops lining the corridors between them turned her walk from point A to point B into a series of serendipitous purchases that kept the whole ecosystem humming, including a stop or two at the food court to grab lunch or dinner.

Fifty years before that, Susan’s grandmom was one of the millions who found the lure of the department store irresistible. A New York Times article published on February 2, 1927 cited Federal Reserve data showing department store sales setting new retail records in 1926. (If you do happen to click on the article link, check out the ad for life insurance on the same page.)

For the first time, curated retail items could be found under one roof for granny to touch, feel and try on. Knowledgeable salespeople demonstrated products and answered questions, turning store browsers into buyers. Stores offered credit to make buying easier, and retail sales took off. And like shopping at the mall, going to the department store was an experience that was part social, part shopping, but mostly lots of fun.

Julie Satow’s book When Women Ran Fifth Avenue documents the role department stores played in shaping commerce and influencing fashion in their heyday. It’s a great read.

Today, Susan’s granddaughter, Ellie, starts her shopping trip with a scroll.

Her smartphone and apps give her access to the equivalent of every mall and every department store ever built with a tap and a swipe. PYMNTS Intelligence data shows that nearly two-thirds of American consumers start their shopping trips that way, making purchases 12 days each month on average and window shopping for another 12. For retailers, this is pretty good news. Mobile window shoppers convert to a purchase at a rate three times higher than the casual mobile phone user. Younger consumers and parents skew that number even higher.

Read More: The 2025 Global Digital Shopping Index: The Rise of the Mobile Window Shopper and What It Means for Payments

The digital world gives Ellie tons of options to buy whatever strikes her fancy without ever visiting a store. She has certainty about product quality and inventory availability even if she decides to make the trip just to see the item for herself. If she doesn’t, the logistics of getting her items delivered are transparent, predictable and efficient.

The latest PYMNTS Intelligence data finds that 48% of clothing and accessories, 61% of electronics purchases and 75% of sporting goods and hobby purchases made online now happen at Amazon. Overall, almost two in every ten retail purchases now starts and ends online.

Read More: Walmart Aims at Closing Amazon Online Sales Gap

But Ellie’s generation may be the last to make even that digital journey themselves.

Do Consumers Still Want to Shop at a Store?

Warren Buffett said in a 1977 Wall Street Journal interview that the best business to own is a toll bridge. His thesis: Once capital is invested to build one, you can keep collecting revenue and raising prices since you control access to the places people want to go.

If you can’t be the destination and monetize it, the next best thing is to own the metaphorical bridge everyone has to cross to get to one. Platforms are a great example of the art and science of being a massively successful toll booth. Or a complete disaster, if you get the platform economics wrong.

Buffett’s aphorism is particularly appropriate to shopping in an age of AI agents.

The retail industry is in the throes of a heated debate about whether to open storefronts to AI models so that consumers (or their agents) who start their product journeys there can find their merchandise and buy. That’s the OpenAI, Stripe, Adyen, Fiserv, Perplexity, Shopify, Google, PayPal and Walmart thesis.

Read More: Department Stores of the Future Are AI Agents

The other side of the argument: if you’re already the destination, what’s the point of the bridge? If consumers already come to you to search, discover, compare and buy, then connecting to someone else’s toll bridge is not in your best interest. As the destination, you’re the bridge and the toll booth all wrapped up into one. That’s the Amazon and eBay thesis. (Amazon, I get. eBay is the ultimate head-scratcher, if you ask me.)

But there’s a more fundamental question at the heart of this issue even before getting to the strategic question of open versus closed.

In a world where agents can buy, do consumers still want to shop at a store?

Will Ellie, her kids or her grandkids need to start at “the store” in order to gather product information, read reviews, evaluate options and decide what, when and whether to buy so that they will have confidence the order will be right and show up on time? If so, consumers won’t be willing to fully delegate to an agent because too much can go wrong.

Or will shopping, as we know it, shift from stores and apps to a collection of hundreds of billions of digital SKUs that agents shop and buy on a consumer’s behalf? In that scenario, consumers say shopping across endless aisles and multiple drives to physical stores is too much of a hassle. Agents can do it smarter and more efficiently.

This turns out to be the $5.5 trillion question for retail in the U.S.

Because if AI agents do the shopping, nobody (or many millions of nobodies) will “go” anywhere. Instead, shopping’s theoretical bridge leads to another bridge and another bridge and another with their own tollbooths.

The destination that was once the store becomes irrelevant because it becomes invisible. The new destination becomes the prompt and the agent dispatched to do the shopping and buying. The bridges and the tollbooths connect to and from this new agentic storefront.

The Shift That’s Already Here, Sort Of

The data says this isn’t so much of a theoretical debate anymore.

PYMNTS Intelligence research shows that more than six in ten U.S. consumers used AI in the past year to do something. More than a third of Gen Z consumers and power users now start their daily tasks on dedicated AI platforms first, including content discovery. And not in addition to Google search, but as a replacement.

As of January 2026, not only have 41% of consumers used dedicated AI platforms for product discovery, but 33% say they have fully replaced their prior methods. They’re not layering AI on top of old habits. They’re shutting the door and leaving them behind.

Read More: Smart Agents Replace Super Apps

This behavior is more predominant among the early adopters who are all in on AI and agents, with 51% having replaced their old methods of product discovery. True to their early adopter roots, they are willing to tolerate untold friction to try something new. Among AI power users, the 24 million Americans who use AI and agents to do everything from building shopping lists to doing research on what stocks to buy, the share who reported replacing their previous search and discovery methods continues to increase since November 2025, when it stood at 46%.

For them, the front door of commerce is already in motion.

But here’s the caveat.

The growth is being driven overwhelmingly by use cases that are not yet letting agents make high-stakes, complex purchasing decisions on their behalf.

In part that’s because the inventory of products and merchants to shop remains nascent. In part it’s because the experience is largely circa the early days of internet commerce: functional, but not exactly one-click amazing. And mostly it’s because consumers still have to trust that their purchase discovery and outcomes are as good as the content discovery and outcomes when the starting point is the physical or virtual store.

For the moment, the interest in doing these things is higher than the reality of actually doing them.

And yet about 82% of these power users, who are the most likely to have replaced their old discovery methods, say they would use AI agents for big, complex purchases where the stakes are high and getting it wrong has financial consequences. These are not impulse or everyday buys. They are the high-consideration, high-research decisions that used to take hours, even days or weeks. And where a lot of spend hangs in the balance.

That’s consumer intent at the very leading edge of AI and agents. And it’s pointed directly at the heart of how retail works today.

Why Shopping is Not a One-Size Fits All Experience

The mistake in the current debate about agents versus stores is treating “shopping” as a one-size fits all activity. It isn’t. There are several distinct reasons for how, why, when and where people buy. And the relevance of AI agents will vary tremendously based on those buying triggers.

Take replenishment, AKA subscriptions.

The dog food. Laundry detergent. Paper towels. Toothpaste. The purchases where the consumer made the real decision once, maybe even years ago, and has been on autopilot ever since. There’s no discovery here. No joy. Just the mild annoyance of remembering to place an order before it runs out. And the catastrophe of running out when they forget something important.

This is where an AI agent doesn’t just help, it can take over entirely and reinvent the subscription experience along the way.

No more boxes of paper towels that pile up in the basement because someone in the household forgot to pause an order. Instead, an agent can notice and nudge.

“It’s been six weeks since you ordered dog food. Time to reorder?” “You haven’t refreshed your white short-sleeve t-shirts for the summer. Want the same ones you bought last year, or take a look at the five most popular styles in your size?” “Time for new flip flops. Same pair as last June, or want to see what’s out there?”

In truth, consumers don’t want to spend their time shopping for these things. They want the buying process to disappear. That’s what the agent can do, with a prompt and one-tap confirmation.

Read More: Why 30 Million US Consumers No Longer Search

Amazon already does a version of this with Subscribe & Save, which I live by. Alexa+ attempts to take it further by adding context, knowing that summer is coming, that the dog food bag lasts roughly this many weeks, that last year’s flip flops ran a half-size too small because of a return and a reorder.

Early access data from Amazon on Alexa+ finds that users tripled their shopping activity and had two to three times more conversations compared to the original Alexa. Amazon made Alexa+ fully available to all U.S. users this month (February 2026), and it’s free for its 250 million Prime members. That’s a lot of people who already live with an agent on their kitchen counter or inside an Amazon app.

Walmart’s play with its One Pay banking, credit, shopping and rewards app tries to capture a piece of this layer by turning the grocery trip into a financial flywheel that expands into replenishment. It remains an aspirational goal. Walmart’s ecommerce business has seen strong growth over the last year (from 16.4% to 19.9%), with much of that growth coming from groceries, which drive nearly 60% of their sales. Its Subscribe and Save service, which launched in 2023, is positioned as a counter to Amazon’s in capturing recurring sales for groceries and essentials.

In an agentic world, whoever owns this layer owns the most frequent, most predictable and most invisible transactions in a consumer’s life.

Then there are the bigger-ticket, once-every-so-often, more considered purchases.

A new camping tent. A stroller. A laptop. A dishwasher. A new car. These are the decisions where people currently spend hours reading reviews, comparing specs, toggling between browser tabs and going back and forth over whether the extra hundred dollars is worth it. And this is exactly the layer where the PYMNTS Intelligence data gets most interesting. It’s precisely these complex, high-stakes decisions where consumers are most eager to hand the tedious task of evaluation those options to AI.

Read More: From Assistive to Agentic AI: Consumers Wade Into Autonomous Commerce

It makes sense. Typing in (or speaking) a detailed prompt is just easier than scrolling through forty-seven open tabs. A consumer can describe what they need in plain language and the agent does in seconds what used to eat up an entire afternoon or more.

Amazon’s Rufus is the most visible example of what this looks like inside a closed ecosystem. The numbers Amazon reports tell you why Amazon is pushing so hard on it.

They report that some 250 million shoppers used Rufus in 2025, with monthly active users growing 140% year over year. Amazon says it’s on pace to drive more than $10 billion in incremental annualized sales with it. Rufus users are 60% more likely to complete a purchase than non-Rufus shoppers. During Black Friday 2025, sessions involving Rufus that ended in a purchase doubled compared to the trailing thirty-day average, while non-AI sessions grew just 20%, they say. That doesn’t feel like a marginal improvement. It suggests that a fundamentally different shopping behavior is happening inside its ecosystem.

Then there are the LLMs such as ChatGPT, Claude, Gemini that want to be the destination where that journey starts and ends, routing consumers to the right merchant with the right product at the right price. It’s also where the strategic tension begins to get, well, a little tense.

Read More: Why the ‘Person’ of the Year in 2025 Should Be the Chatbot

Amazon can end the shopping journey because it owns what I’d call logistics certainty. A consumer knows when the product is arriving: same hour, same day, next day. They know what the shipping costs are: mostly free with Amazon Prime, which is a bonus and eliminates the uncertainty of not knowing the final cost. A consumer knows how to return an item if there’s a problem. And where to do it for free. The entire post-decision experience is part of their destination’s appeal.

Perhaps one of the most underrated variables in the entire agentic commerce debate: What exactly happens once an agent clicks “buy.”

Then there are the highly complex, product/service blended purchases that support some of life’s biggest moments.

The 25th anniversary trip. Having a baby. Sending a kid to college. Buying a first house. Getting a puppy. Planning a wedding. Moving to a new city. Each of these involves dozens of purchases, but none of them is really about filling a cart. They are about making a life passage happen.

It’s also where the question of whether consumers still want to go to stores gets interesting. The answer might be yes, but in a different way than it happens today.

Nobody wants to spend a week researching car seats, strollers, nursery furniture, baby monitors, bottle warmers and the thirty-five other things your friends, relatives and in-laws insist you absolutely must have before the baby arrives. Then there’s the physical infrastructure necessary to support the baby. The diaper service. The pediatrician. The day care. The preschool. It’s exhausting, often contradictory and largely anxiety-producing.

The agent’s role in these moments isn’t to shop for those parents. It’s to be the smart and efficient concierge that helps to simplify the massive complexity around this important life moment.

Today preparing for the new baby (or any of life’s biggest moments) looks like parallel processing: researching across endless tabs, juggling competing recommendations, stitching together a plan from fragments.

Tomorrow it might look like this.

Congratulations! You’re having a baby! Here’s a curated registry across eight retailers based on your budget, your house or apartment size, and what parents with similar lifestyles actually used and loved. Here are the retailers that have most of these items in stock if you want to go in and check them out IRL. Here are the three pediatricians near you accepting new patients with strong reviews. Here’s a timeline of what to buy to avoid the panic-ordering purchase of any car seat in stock at 2 a.m. three days before the due date.

The agent handles the things that take time and create frustration. The hassle of research, comparison and logistics coordination is stripped away, and what’s left is the part that actually matters to people. The choosing, experiencing and satisfaction of making a good decision in the context of something very important about to happen.

Who Plays Where and Why It Matters

So, who owns the toll bridge and who becomes the destination? That depends on which version of shopping you’re talking about. And the four most consequential players in this space, Amazon, Google, Walmart and the LLM platforms, are each building from very different starting positions.

Amazon: The Everything Store wants to become the Everything Concierge.

Amazon pulls in roughly 2.5 billion visits a month to its site, commands roughly 9.1% of all U.S. retail spending according to the latest PYMNTS Intelligence report, serves more than 300 million active customers, with 250 million-plus Prime members locked into an ecosystem that has become the defacto starting point for product purchases. Amazon reported Q4 2025 revenue hit $213 billion, up 14% year over year. These are not the numbers of a company that needs to reinvent itself. These are the numbers of a company that can afford to make big bets from a position of massive retail strength.

That appears to be what Amazon is doing.

Subscribe & Save handles replenishment. Alexa+ is hoping to extend that by contextualizing it, learning the consumer’s purchase velocity, anticipating what they need before they think of it and eventually ordering ahead of time with its auto-buy feature.

For more considered purchases, Amazon has built the perhaps most complete environment in retail: search, reviews, product information, price comparison, and then the killer, logistics certainty, all in a one-stop shop. Consumers start and end the journey on Amazon and reliably know that the product will arrive when promised. Rufus and Help Me Decide don’t replace the shopping journey, but instead compress it.

And at a nearly 10% advertising conversion rate, roughly five times Google Shopping’s rate, the math for brands selling on Amazon is hard to argue with.

Amazon’s bigger bet with Alexa+ is to push beyond shopping and into life. The ambition is not to be just a shopping assistant but a life operating system.

Alexa+ already integrates with Uber, Grubhub, Ticketmaster, Vagaro for spa and fitness, Thumbtack for home services, Square for merchant services, Expedia for travel, Yelp for local discovery, and Amazon Autos for car sales. Amazon’s Panos Panay has described a vision where Alexa becomes the consumer’s personal shopper, butler and home manager. The more she understands about the consumer’s life, the better she can serve the customer.

Add Amazon’s healthcare play through One Medical and pharmacy, its grocery infrastructure through Whole Foods, Fresh, and the Uber Eats delivery partnership, Prime Video for entertainment, and Alexa embedded in hundreds of millions of homes, and you start to see the outlines of a company that could plausibly become the life concierge, not just the destination as the store. What we once called the Super App.

Here’s the catch. Amazon may make more trips to my home in a week than I’d like to admit, but Alexa+ still has a long way to go, at least in my experience, before she can live up to the claim of a personal assistant. She is what most consumers say they’d trust, according to PYMNTS Intelligence data. For me, it’s a crap shoot as to whether she can reliably turn on Bloomberg TV in the morning on my Fire TV, never mind organizing my day-to-day.

The conundrum is that Amazon is trusted today for efficiency, reliability and price. But no one is thinking of Amazon curating their wedding registry. The Everything Store may struggle to become the Everything Concierge precisely because life moments require warmth and judgment that an efficiency machine may not inspire. And Alexa+ hasn’t proven she can deliver.

Then there’s Google, which is playing a fundamentally different game.

Google is a massive advertising machine that generated more $400 billion in revenue in 2025, with advertising driving the lion’s share of those numbers. Google says its Shopping Ads drive 76% of all retail search ad spending. Google is, by a wide margin, the world’s largest digital advertising platform.

But Google’s position in commerce is that it is really nowhere. It’s always been the bridge, never the destination. Consumers search on Google, discover products, and then leave to buy somewhere else. Google gets paid for the referral.

That’s what Google is trying to change.

In January 2026, CEO Sundar Pichai unveiled the Universal Commerce Protocol at NRF, an open standard designed to let AI agents navigate the full shopping journey from discovery through checkout, all within Google’s own ecosystem. Google co-developed UCP with Shopify, Etsy, Wayfair, Target and Walmart, and got endorsements from Visa, Mastercard, Stripe, American Express and Best Buy. The message was clear.

Google wants to stop being the bridge and become the destination.

Read More: The Protocol Power Struggle Reshaping AI-Driven Commerce

This announcement came shortly after Google rolled out agentic checkout in November 2025, letting users set a target price and authorize Google to auto-purchase via Google Pay when the price drops. Its Shopping Graph now indexes more than 50 billion product listings, with 2 billion updated every hour. The new Business Agent feature lets retailers like Lowe’s, Michaels, Poshmark, and Reebok deploy branded AI assistants directly inside Google Search to chat with shoppers and close sales. Direct Offers, a new ad pilot, offers exclusive discounts to high-intent shoppers in AI Mode.

It’s nothing if not ambitious. But Google’s challenge remains closing the gap between intent and conversion by turning itself into the internet’s marketplace.

Google Shopping Ads convert at roughly 1.91%. Amazon’s marketplace converts at nearly 10%. That’s a five-to-one ratio, and it tells you everything about the difference between a platform where people go to browse and a platform where people go to buy.

Open standards and broad retailer partnerships are a compelling pitch, but until more shoppers are completing purchases inside Google rather than clicking away to finish somewhere else, the toll bridge metaphor still holds. And merchants need to consider how much, and when, to put effort into exposing their entire product catalog to Google, without a clear understanding of how it plans to monetize those sales. And who owns the customer relationship. More on that point later.

Walmart is the wildcard with one truly irreplaceable asset: its physical storefront.

Roughly 100 million people walk into Walmart for the most frequent, most habitual, most non-discretionary purchase in retail: their groceries. That foot traffic is massive. But it’s also their greatest Achilles’ Heel.

Few people who go to Walmart for groceries buy anything else. The foot traffic is enormous, but their slice of the retail basket is narrow. PYMNTS Intelligence data shows Walmart’s share of retail declining in nearly every category except food.

Walmart’s AI strategy seems two-fold. With its AI partnerships, Walmart appears comfortable being a destination where the LLM toll bridges send traffic because its physical infrastructure is something no AI platform can replicate. And digital is where they lack meaningful share.

Physically, the One Pay credit and rewards play is an attempt to expand what the in-store shopper buys — to turn a grocery trip into a broader financial relationship. If you are already in the store buying food and the Walmart app knows you need new school supplies for your 10-year old,  it can offer you credit and rewards to buy them right then, and might expand the basket. It’s a theory, for now.

Like Amazon, the challenge for Walmart is in playing the life concierge role.

Walmart is not a lifestyle brand. It is not where a consumer goes to curate their baby registry or furnish their first apartment or plan their wedding. It is not a Super App.  Unless Walmart can leap from “where I buy groceries” to “where my Walmart agent manages my life,” it remains powerful but confined to a single, if highly defensible, position as the world’s biggest grocery store. With a little retail eCommerce on the side.

And then there are the LLM platforms that are building the toll bridges that want to own the road. And become the destination.

Consumers are increasingly starting their discovery journeys on these platforms, and the agents can route them to any merchant. That brings with it enormous disintermediation power. The PYMNTS Intelligence data confirms it. Consumers are replacing traditional search and discovery methods with AI-first approaches, and the replacement rate is accelerating.

Adobe reported that AI-driven traffic to retail sites surged 693% during the 2025 holiday season, and shoppers arriving from AI services were 38% more likely to convert. This sounds amazing. But transaction volume is miniscule; this is still very early days.

OpenAI has already launched its own Instant Checkout feature. Microsoft Copilot is partnering with Shopify for embedded checkout. Perplexity was the first to launch one-click checkout within its app. Everyone wants in on commerce.

But the LLMs have a structural vulnerability that the current hype obscures. They have no logistics. They can send you to a store, but they cannot guarantee when the product arrives, what shipping costs, or how returns work. Every merchant on the other side of their toll bridge is a different experience with a different uncertainty profile. And the selection of those merchants is pretty limited right now.

Being discoverable by an agent does not solve the fulfillment gap. And that is why Amazon’s position is structurally stronger than the LLM toll bridge position, at least for now. Amazon is the one destination where the post-agent experience is already solved.

My sense is that the LLMs’ real aspiration is the life concierge.

Not to sell you things, but to be the agent that manages complexity across your entire life.

Getting a puppy? It finds the vet, orders the food, finds and books the trainer and the dog walker,  schedules the groomer, arranges for doggie daycare and the transportation to and from. The LLM becomes the relationship layer. Merchants, including Amazon, become the inventory.

Their advantage is being platform agnostic, smart and a massive time saver. The disadvantage is that a concierge without logistics and without fulfillment is ultimately dependent on others to deliver. And without access to all of the right products, creates buyer uncertainty.

In a world where the uncertainty tax on fulfillment determines which agent recommendation the consumer actually trusts, that dependency may be the LLMs’ defining limitation.

And  why “just open your storefront to the AI agents” is not the simple answer it appears to be for merchants.

Who Owns the Relationship When No One Goes Shopping Anymore?

Let’s close with how we started. The destination and bridge/toll booth metaphor assumed a world with fixed destinations and fixed paths to get to and from. The destination was valuable because there were a finite number of destinations, and lots of  people wanted to get there. There were only a few ways to  get in and out. The bridge was valuable because it made the trip possible. And people and businesses paid the tolls.

The agentic world makes the destination and the bridges with tollbooths pretty fluid. For the replenishment shopping example, the destination is invisible, handled by an agent on autopilot that noticed a consumer was running low before they did.

For considered purchases, the destination is whoever collapses the research journey fastest while maintaining the certainty that the product will arrive when promised and can be returned if it’s wrong.

For life moments, the destination is wherever the people feel a connection and a sense of trust to start the conversation.

The real strategic question comes down to how consumers view the shopping experience for each of those use cases.

Do they want an agent that makes shopping disappear? Or do they want an agent that makes the shopping experience at a store they know and trust  better?

The answer, almost certainly, is both. The company that figures out how to deliver both will own the most valuable thing in commerce.

Not the product, not the price, not the storefront, and not the toll bridge.

The relationship. The real starting point of any shopping journey.

 

Until NEXT time.

Join the 19,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.

For most of the twentieth century, the sound of work and study in America was the clickety‑clack of typewriter keys in offices, newsrooms and dorm rooms as hundreds of thousands of machines moved ink to paper using typewriter keys.

By the late 1980s, Smith Corona sat at the center of that world. It was reported to have a 50% share of the typewriter market. Its reputation as a reliable, well‑managed manufacturer could be measured by strong sales, robust distribution channels and steady margins.​

Smith Corona’s leadership team did exactly what great management teams with a dominant market position are trained to do. They listened to their best customers, doubled down on what they asked for  and operated the business with a financial model that was familiar, predictable and profitable. Innovation consisted of incremental improvements at the margin: quieter machines with less clickety-clack, more and better fonts and typos that could be more easily corrected.

When PCs came onto the scene in the late 1970s and early 1980s, the Smith Corona team sort of shrugged it off. They were new, expensive and hard to use, regarded mostly as “luxuries” for engineers and hobbyists. Pouring capital into such a new and unproven PC market seemed like a bad idea, especially since not a single one of their customers asked for one. So why risk the distraction?

A solid base of long-time, loyal, high-margin customers asking for more of the same but better made PCs look like a sideshow. That was, until the typewriter itself became the sideshow less than two decades after those so-called luxury toys came into the market.

The late Clay Christensen would give that management logic a name. His 1997 book, The Innovator’s Dilemma, found its way on the reading list of most every MBA syllabus. In that book, Professor Christensen describes how customer‑centric, financially rational behavior can lead great firms into irrelevance when challengers with new technology enter the market.

His thesis is that those leaders can become structurally blind to the threat of emerging technologies. New technologies, when introduced, are typically clunky — like the first iPhone that couldn’t actually make and receive any calls — and filled with friction. Those early adopters don’t always signal a rush of fast followers.

Read more: What Electric Vehicles, Impossible Foods and Buy Now, Pay Later Teach Us About Early Adopters

Yet history shows that less-profitable, initially clunky innovations  frequently become the catalysts for market redefinition, as emerging technologies introduce something better than the status quo to customers who never directly asked for something new.

That same dilemma now faces the Big Five technology giants as they bring AI and agents into their business.

Amazon, Apple, Google, Meta and Microsoft each have world-class technical talent and a track record of innovation. Each is already weaving AI into its existing products; many have been using machine learning for decades to improve the user experience. Each tells a compelling story about how AI will make its core businesses better. Each has pioneered new and better ways for consumers and businesses to engage.

The more pressing question is whether any of them will treat AI the way Christensen said disruptive technologies must be treated.

By allowing it to challenge and potentially disrupt the business models, P&Ls and organizational structures that pay the bills today. And leave a lot left over to fill the piggy bank.

Google: The Innovator’s Dilemma in Real Time

If there were ever a company facing its own modern day version of the innovator’s dilemma, it’s Google. It is the rare example of a firm that both invented the core technologies behind the current AI and agentic disruption and controls one of the most profitable business models in modern history.

Google’s AI bona fides are impressive.

First, its technical contributions to AI are foundational.

Google Brain and DeepMind did not merely adopt deep learning early, they helped define it. The 2017 Transformer architecture described in the “Attention Is All You Need” paper fundamentally rewired how machines process language, vision and sequence data. Nearly every frontier model — Gemini, GPT-4 and 5, Claude, Llama — is built on that architecture.

DeepMind’s AlphaFold solved a grand challenge of biology at scale, reshaping drug discovery and molecular science. Nobel Prizes in 2024 in Chemistry and 2025 in Physics tied directly to Google’s AI research team underscore genuine scientific leadership.

Alphabet today still makes its money the same way it always has: by selling and monetizing eyeballs through advertising. In Q3 of 2025, Google Services, which includes Search, YouTube, network ads and subscriptions, generated 85.1% of total revenue. Search alone contributed 55%. Even Google Cloud, now exceeding $43 billion in annual revenue, remains a minority business that has not changed the company’s fundamental dependence on advertising.

This matters because AI does not merely improve the search and discovery experience. AI and agents threaten to compress or eliminate the very behaviors that make search ads so lucrative. Agentic systems answer questions directly. They summarize, compare, recommend and increasingly act. Each step collapsed by AI removes page views, links, ads, keywords and bidding opportunities. The same technology Google invented to understand language better can dramatically reduce the opportunity for the monetization on which its ad business depends.

Read more: When Chatbots Replace Search Bars, Who Wins at Checkout?

This fork in the road to new and different ways of monetizing its platform is nothing new to Google, particularly in commerce. But prior attempts haven’t ended well.

For more than two decades, Google has tried and failed to move beyond being a click to a page on a website. Google Shopping never became a destination. It became just another version of advertising. Google Pay, after more than fifteen years, finally scaled distribution but remained infrastructure for banks and networks rather than the foundation for a merchant ecosystem. Time and again, when forced to choose between building a new economic engine and reinforcing advertising margins, Google stuck with the latter.

Gemini represents the most serious test yet of whether that pattern can change.

Today, Gemini sits atop a rebuilt Shopping Graph with roughly 50 billion product listings Google says are refreshed hourly. AI Mode allows conversational discovery that feels different from traditional search. Users can ask for recommendations, constraints, trade-offs and comparisons at the prompt the way they’d talk to a sales associate: a conversation rather than a string of keywords. With the Universal Commerce Protocol and agentic checkout flows announced at NRF, Google has moved one step closer to closing the discovery/purchase loop, allowing discovery, selection and payment without leaving the Google ecosystem. Chrome integration and tokenized GPay credentials push this further, reducing friction to near zero. In theory.

Read more: Why 30 Million US Consumers No Longer Search

This has all of the structural underpinnings of  a marketplace.

But organizationally and economically, Google still behaves like an advertising platform. Ranking logic, monetization incentives and merchant relationships are all optimized around performance marketing. Even zero-click conversion still functions as a pay-for-click model rather than a reimagining of commerce. To truly become a commerce platform, Google would need to govern transactions, manage disputes, handle returns and accept responsibility for outcomes. Not just route demand more efficiently.

Doing so would mean shifting revenue from CPC and CPA toward transaction fees, merchant software and financial services. It would also mean accepting lower margins in exchange for structural diversification. That is the choice that Christensen warned incumbents struggle to make.

Technology makes both paths possible. Google’s incentives make one far more comfortable than the other. At least for now.

Meta: AI as Advertising Optimizer

Meta offers a much more direct version of the same story. Where Google at least seems to flirt with commerce, Meta has spent the past decade demonstrating their lack of success in escaping a business model that works very well.

Meta is, at its core, an advertising platform. In 2025, roughly 98.9% of revenue came from what is called its family of apps, including Instagram. Reality Labs contributed barely more than one percent and continued to generate massive losses, with cumulative operating deficits approaching $80 billion by late 2025. No amount of spin can obscure the Meta reality. Advertising pays the bills and then some. And nothing else has come close to replacing it.

What makes Meta’s case interesting is not a lack of ambition. The company has repeatedly tried to invent its way out of advertising dependence. Hardware initiatives like Portal failed to gain traction. Payments efforts stalled. The Libra/Diem stablecoin project collapsed under regulatory pressure, partner unease and a flawed business premise. Betting the company (and name) on the metaverse vision consumed tens of billions of dollars but never translated into mass-market behavior.

Read more: What We’ve Learned From Libra

Read more: How Facebook Turned the Metaverse Into a Buzzword

Even commerce features inside Instagram and Facebook Shops ultimately fed back into ads rather than standing alone as transactional platforms.

AI and agents have not changed that dynamic. They seem to have only intensified it.

That said, Meta’s AI assets are impressive. Llama is an open foundation model family, with recommendation systems that analysts say operate at unmatched scale, along with sophisticated ad-ranking and creative tools. But nearly all of this capability is aimed at one objective: maximizing the efficiency and yield of advertising. And the result is hard to dispute. AI-powered ad products reached roughly a $60 billion annualized run rate in 2025, with measurable lifts in click-through, conversion and pricing.

Smart glasses illustrate the same pattern. Once positioned as a step toward augmented reality, they are now framed primarily as an AI interface that keeps users inside Meta’s ecosystem. The revenue model ultimately points back to monetizing eyeballs (literally) through ads, engagement and new ad formats rather than an independent platform business built to sustain commerce.

For Meta, AI and agents do not threaten the core. They only strengthen it. Who knows, maybe we will see another name change soon: MetAI.

Apple: Hardware as the Center of Gravity

Apple’s innovator’s dilemma is a puzzle because it looks so different from Google’s or Meta’s. Apple is not dependent on advertising. It does not monetize eyeballs. Its margins come from selling slick hardware at scale and embedding high‑margin services within its operating system to keep its users there. But that strength has also created a pattern of repeated, expensive near‑misses in AI‑driven categories where Apple should, on paper, have won.

In fiscal 2025, Apple generated roughly $416 billion in revenue. The iPhone alone accounted for just over half. Services reached a record $109 billion, with gross margins north of 70%, making it the most profitable segment in the company. Yet nearly every dollar of Services revenue remains tethered to the installed base of devices. Apple’s economic center of gravity still runs through the handset, and everything else exists to protect, enhance or extend that core.

That thesis helps explain Apple’s uneven history with AI.

Read more: Why Generative AI Is a Bigger Threat to Apple Than Google or Amazon

Take Siri.

Launched in 2011, Siri had an early lead over Google Assistant and Amazon Alexa and had every opportunity to nail voice and become a trusted voice assistant. But Apple treated Siri as a feature rather than a platform for connecting the user with services and apps within its ecosystem.

And a pretty poor one at that. Not surprisingly, Siri stagnated. By the time large language models reset user expectations for conversational intelligence, Siri had become shorthand for the not-so- smart assistant that users largely rejected.

Apple’s AI struggles extend well beyond Siri and voice. The company spent nearly a decade and billions of dollars on Project Titan, its autonomous and electric car initiative, only to shut it down in early 2024. The project cycled through leadership changes, shifting goals and strategic resets, never resolving whether Apple was building a vehicle, a self‑driving system or a broader mobility platform. Ultimately, it produced no new revenue stream and no new platform.

Home and ambient computing tell a similar story. HomePod never became the control panel for the home. Apple TV remained a content endpoint, not a broader services hub. Despite tight hardware integration and a loyal customer base, Apple failed to turn the home into a meaningful AI surface where assistants, commerce and automation converged.

Payments show both Apple’s strengths and its limits. Apple Pay and Wallet achieved massive global penetration as credential and tokenization layers. Apple Card and Apple Pay Later extended that footprint modestly. But Apple stopped short of building a full merchant services stack, a commerce marketplace or an AI‑driven purchasing layer that could compete with Amazon. Payments remained infrastructure, not a platform.

Read more: The One Big Thing Apple’s Project Breakout Needs but Doesn’t Have

Search is another glaring absence. Apple controls one of the most valuable distribution platforms in consumer computing, yet it never seriously attempted to build a general‑purpose search engine or AI discovery layer of its own. Instead, it has relied on multibillion‑dollar payments from Google to keep Google Search as the default on Safari. That arrangement is enormously profitable in the short term, but it left Apple without deep institutional experience in search, just as AI began to redefine those functions.

GenAI  exposed the cumulative effect of these choices. As large language models became the new interface for search, assistance, and conversion,  Apple lacked a frontier‑class general‑purpose model of its own. We see this deficiency playing out now in real time as Apple continues to lose key AI talent and multiple attempts to launch Apple Intelligence have turned into an oxymoron for this $4 trillion company.

Read more: Apple’s $10B AI Crisis. 3 Bold Moves To Reinvent Its Future

The decision to partner with Google and use Gemini as the backbone for a dramatically upgraded Siri is regarded as a Hail Mary move designed to shore up Apple’s weakest flank and preserve the relevance of the iPhone and Services ecosystem. But it also confirms that Apple is renting intelligence rather than owning it. AI, in this configuration, is a defensive layer wrapped around an existing business model, not a force reshaping it.

Apple’s dependence on hardware, notably smartphones, is clear. And its innovator’s dilemma is a strategic challenge that Tim Cook’s successor will inherit.

Amazon: Commerce-Native And AI-Powered

Amazon is often described as the best‑positioned company for an AI‑driven future because it already operates a global commerce, logistics and cloud infrastructure at extraordinary scale. And it has a long history of using AI and machine learning to make the engagement with its platform stakeholders more engaging and profitable. But Amazon’s starting point in commerce could end up being both its greatest advantage and its greatest constraint.

By 2025, Amazon generated roughly $715 billion in annual revenue. Retail remains the largest top‑line contributor, but AWS and advertising account for a disproportionate share of operating income. And its $68.2 billion advertising revenue is almost pure profit. Amazon already rebuilt commerce around data, automation and algorithms long before generative AI arrived. AI is fully embedded in pricing, search ranking, inventory placement, fulfillment routing, fraud detection and advertising.

And there there’s Alexa.

When Alexa launched, it defined the modern voice assistant category. I was one of its earliest and most enduring fan girls. It seemed like such a powerful platform and launch pad for moving Alexa into the physical world and expanding the Amazon footprint.

Read more: How Consumers Want to Live in a Conversational Voice Economy

Amazon, with Alexa, was first to scale dedicated voice hardware into tens of millions of homes and hundreds of millions of devices. Buick once advertised the value of its car around its integration with Alexa in the cockpit. The ambition was explicit: Amazon’s Alexa would become the consumer’s virtual personal assistant for shopping, services and daily life.

As a voice assistant, Alexa was pretty good at telling the time, the temperature and the occasional bad Dad joke. It was great at taking orders like setting timers, turning the lights on and off and opening and closing the blinds. Yet its ability to cross the proverbial chasm to shopping and commerce proved disappointing, even when Alexa was embedded in a device with a screen. Consumers began to lose trust. The skills ecosystem failed to mature into a vibrant marketplace to be monetized. Alexa lost momentum.

Internally, Alexa became a costly initiative with weak direct revenue, reportedly generating tens of billions of dollars in cumulative losses. Amazon began cutbacks in its hardware and Alexa business units in later 2022 and continuing into the Spring of 2025.

Hardware complicates the picture further. Amazon has never cracked consumer hardware economics. Echo speakers were subsidized to drive adoption and engagement of Alexa. As AI assistants and agents become embedded into phones, cars, wearables and operating systems, the standalone smart speaker will become the technological equivalent of the dodo bird. AI and agents will be ambient and live where screens, sensors and identity already exist. That’s not in plastic cylinders on kitchen counters.

Now, GenAI and agentic have given Amazon and Alexa a different opportunity to reclaim lost ground.

Rufus, embedded in the Amazon app as a shopping assistant, plays to mixed reviews. With 250 million users and claims by Amazon of improving conversion by 60%, users seem to either love it or hate it. Alexa+, announced as a generative upgrade, promises richer conversation, task execution and orchestration across services. The jury is still out. My own experiences with Alexa+ have been mostly mixed.

Both Rufus and Alexa+ are structurally optimized for Amazon’s own commerce rails. Alexa and Rufus both route demand inside of Amazon’s ecosystem, not across and outside of it. And, wisely in my opinion, Amazon has shut off access to its ecosystem from AI models. Amazon is a destination with consistent access to billions of SKUs. And it has a track record of connecting AI and agents to purchase and conversion within it, even if Rufus can be annoying at times.

So, this is where Amazon’s innovator’s dilemma becomes interesting.  To become the Super Agent for daily life, Alexa would need to operate across ecosystems, booking services, sourcing products, managing tasks and executing payments whether or not they are inside of Amazon’s ecosystem. Monetized in some way, either tied back to Fulfillment by Amazon or using its Amazon Pay wallet to monetize the transaction.

Read more: Smart Agents Replace Super Apps

Read more: Commerce Finds Its Voice

Amazon’s challenge is not technical. The company knows how to build and scale new business units and monetize them. And they have done so well with AWS. But doing so required separating AWS from retail economics and allowing it to serve a broad ecosystem of companies, including competitors. Alexa and Rufus have not been given that freedom.

Microsoft: Strengthening the Core Without Rewriting the Model

Microsoft’s position with AI and agents is often described as the strongest on paper. It owns Azure, the enterprise stack, and is paid massive amounts of money by LLMs to connect it to other models. It holds a big stake in OpenAI.

And yet Microsoft’s AI story, so far, is one of more of the same rather than reinvention. Copilot sells more feature-rich Microsoft 365 bundles. GitHub Copilot sells more subscriptions. Azure sells more compute.

Read more: The Existential Threat That Microsoft Missed — and Could Put Its GenAI Future at Risk

Microsoft has laid AI tracks everywhere. It has good, powerful engines and an installed base of users. But it is still running trains on routes defined by old workflows. AI makes those workflows faster and better. But is does not yet reshape how value is created or captured.

Investors sense this tension. Massive capital expenditure raises questions about whether AI is generating new demand. Microsoft has the assets to do more. What it lacks, so far, is the willingness to let AI redefine its business boundaries and the business models to support it.

Read more: Why Measuring the ROI of Transformative Technology Like GenAI Is So Hard

The Fork in the Road

Smith Corona didn’t disappear because it ignored innovation. It disappeared because it optimized relentlessly for the business and margins it already had. Their innovator’s dilemma was not having the stomach needed to change course and a plan to creatively destruct itself, even though its best customers weren’t asking for anything different.

The innovator’s dilemma for Big Tech looks different. Then again, it may not be the right lens at all.

None of them have behaved like incumbents afraid to disrupt themselves.

Google bet early and heavily on the core technologies behind modern AI, and is winning big.  Amazon took a real swing at ambient computing with Alexa and Echo, a bet that looks less like complacency and more like a BlackBerry or early Microsoft moment where execution determined the less-than-stellar outcome. Apple swung for the fences as well, from Siri to autonomy to ambient computing, but struggled to turn its ambition into an enduring commerce platform.

Maybe the lesson here isn’t whether the Big Five are willing to change. But whether they can adapt agents and new AI flows fast enough to reshape how commerce in an agentic world works. And how they make money.

 

Until NEXT time.

Join the 19,000 subscribers who’ve already said yes to what’s NEXT.

PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.

She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.