Agentic Commerce Needs Its Febreze Moment

In the mid-1990s, Procter & Gamble made a stunning claim. They said they’d invented a miracle in a bottle.

A chemist working there found a compound that completely got rid of really bad smells instead of covering them up with a fancy perfume-y spray. Stinky bathrooms, wet dogs, last night’s burnt fish dinner, that pile of sweaty workout clothes. 100% gone. P&G reportedly spent millions perfecting it, named it Febreze and sent it into test markets expecting nothing less than a blockbuster hit.

It turned out to be P&G’s version of a big stinker.

Pulitzer-prize winning author, Charles Duhigg explains why in his book, The Power of Habit. It’s one of my favorite books about how our brain is wired to create habits and stick with them.

As you can imagine, the P&G marketing team was flummoxed. Their slam dunk star case study (or so they thought) was a young park ranger in Phoenix who trapped skunks for a living. Her couch smelled like skunk. Let’s just say that became a real problem for her social life. According to the focus group researchers, Febreze, she said, changed her life. So, what the heck?

Then the researchers visited a woman with nine cats. As Duhigg tells the story, the smell in her living room was nothing short of gag inducing. Meanwhile, the cat owner didn’t notice a thing.

That was P&Gs “a-ha” moment. The skunk trapper knew her house stank. The cat lady didn’t. Turns out, most of us can actually be the cat lady in our own homes.

Every habit runs on a three-part loop. A cue kicks it off. A routine follows. A reward tells your brain the loop is worth repeating.

That’s where Duhigg says the habit-forming lesson kicks in. Every habit runs on a three-part loop, he writes in the book. A cue kicks it off. A routine follows. A reward tells your brain the loop is worth repeating. Do it enough times and you stop having to consciously decide every time. It just becomes, well, a habit on autopilot.

Febreze had a routine, spray to get rid of the smell, and nothing else. The vast majority of people who needed it most couldn’t smell the stink in their own houses, so there was no cue to reach for the bottle. And when they did spray, a smell they hadn’t noticed went away. Since they didn’t notice any difference, there was no reward.

A loop that doesn’t pay off never becomes a habit. Or in the case of P&G, a product that didn’t become a blockbuster out of the gates.

The turnaround came after the P&G team watched what people already did when they cleaned their homes. A few satisfied users sprayed Febreze at the very end of cleaning, after all of the chores were done. The spray at the end was more of a little cherry-on-top-of-the-sundae moment instead of a fix to a problem.

A reward.

So, P&G stopped shifted gears. They stopped selling the technology and started selling the reward. It added perfume to a product initially designed to smell like nothing. It borrowed a cue people already had, a room that was spic and span, and put a nice smelling reward at the end of it.

As the marketer who ran the Febreze launch is said to have told Duhigg, “No one craves scentlessness.”

P&G relaunched in 1998. Febreze banked more than $230 million in its first year and became a billion-dollar business in 2011.

But take a look at what P&G didn’t mess with. The tech that was the cornerstone of the product. What changed was everything around it. A cue people already had — clean the house — and a reward they could smell.

Nearly thirty years later, some of the smartest companies in the world built a new way to shop and buy with AI. The tech works, and the pitch is the tech. Agents that are smart and a checkout that is seamless. Consumers had the cue. They picked up half of the routine. But the reward never showed up.

So, they ask AI what to buy when the cue is “I need” or “I want” something. And then they buy it the way they always have.

On Amazon.

The Search Box Changed and the Reward Stayed on Amazon

As all of you Agentic Commerce history buffs already know, last September, OpenAI launched Instant Checkout. The pitch was a brand-new habit loop.

Cue: I want to buy something.
Routine: ask the AI, then buy it right there in the chat.
Reward: the best offer from lots of merchants, at your door with no hassle.

The cue showed up. Half of the routine showed up. The reward didn’t.

New PYMNTS Intelligence data shows how far that first half has moved since. We surveyed 2,191 U.S. consumers in August. Nearly a quarter of them, 23%, now start their retail shopping research with AI. That’s 61 million people, nearly one in four. Fifty-nine percent of AI shoppers say they now research on AI what they used to search for on Amazon. The Google stats are more sobering.

It’s tempting to call that a broken habit and a new one forming. But it’s really not. Consumers have searched in one place and bought in another for more than twenty years. Maybe they used to Google it and then buy it on Amazon. The AI models are a better search box since the prompts can be so much more complex. But swapping search boxes is easy because the consumer risks nothing by trying.

The half of the routine that matters, where the money goes, never moved. That’s because the reward never materialized.

But the half of the routine that matters, where the money goes, never moved. That’s because the reward never materialized.

OpenAI promised more than a million Shopify merchants. Only about 30 ever went live, along with Etsy and Walmart. And the biggest store on the internet sat it out. Amazon took a seat in April on the council that steers Google’s Universal Commerce Protocol, but it still hasn’t opened its own checkout to outside agents, and it has blocked dozens of outside agents from its site while it builds its own. Including Meta’s new AI assistant.

It’s hard to call an offer the best when it’s missing the store where these same U.S. consumers already make about half of their online purchases.

So, consumers did what people do when a habit loop doesn’t pay off. They went to where they got the reward.

Six in ten AI-assisted purchases are bought on Amazon, according to the study. Big box and department stores get 20%. Brand sites get 6%. Average AI-assisted spend went from $889 in June to $1,039 in August, up 17% in two months, and most of it landed in one place.

Figure 1

By March, about six months after launch, Instant Checkout was gone. People involved told reporters it didn’t drive sales. Walmart said purchases made inside ChatGPT converted at a third of the rate of those sent to its own site.

The search box may have changed. The reward stayed with Amazon.

Everyone Says Trust is What Agentic Commerce Needs, but Almost Nobody Says What That Means

Go to any conference this Fall and you’ll hear it from everyone and his mother-in-law. Trust is the foundation of agentic commerce. Heads nod vigorously.

Ask what that means and the hands start waving just as vigorously. Secure credentials. Verified agents. Consumer confidence. A protocol. A badge.

Here’s my simple definition. Trust is what’s left after a habit has paid off so many times that you stop checking and second-guessing. It’s a track record you carry around in your head. In retail, it gets built one reliable product experience, one same day or overnight delivery and one simple return at a time.

In retail, trust gets built one reliable product experience, one same day or overnight delivery and one simple return at a time.

Trust is the name we give to a reward that never misses.

I wrote about Amazon’s loop last October. Cue: I need to buy something. Routine: go to Amazon. Reward: found it fast, got it faster, shipped free. I argued then that AI was rewriting that loop.

A year later, the data says I was half-right. The search step did get rewritten. But a new loop only sticks when it pays off, and this one hasn’t yet. The old reward is still what people value and what drives their shopping behavior.

The Reward Lives Where Something Can Go Wrong

So, why was the search box so easy to swap while the actual purchasing stayed put? It comes down to what it costs a consumer to be wrong.

A bad answer costs nothing but time and maybe a few expletives deleted. If it didn’t produce the right results, you ask again. But a bad purchase costs money. A box has to show up. It has to be the right box with the right stuff in it, not broken or squished or wet, and at the right time. Sometimes it has to go back, and then the money has to come back too.

Keep the cue. Keep the reward. Swap the routine in the middle.

The search loop is one cue, one reward, a few seconds. The buy loop is a chain of rewards that is the routine and plays out over days. The last link, the refund, only proves itself on a bad day.

Duhigg has a golden rule for changing a habit. Keep the cue. Keep the reward. Swap the routine in the middle. That’s exactly what consumers did.

The cue is still “I need to buy something.” The reward is still fast, free shipping and easy returns. They swapped the middle. Less searching on Amazon, way less search happening on Google, more asking AI. Then they collect the same reward from the same place.

Everything Was Equal, and Consumers Didn’t Believe It

In our August 2026 survey, PYMNTS Intelligence ran a controlled test. Imagine an AI assistant doing the buying for you, we posited. It finds the exact same item at the same price, with the same shipping and the same returns, from three places. A large marketplace, a department store or the brand’s own site. “Where do you want your agent to buy it?” we asked.

Forty percent said the marketplace. Twenty-two percent said the department store. Sixteen percent said the brand. The rest said let the AI pick.

Figure 2

Then we asked why. The top answer was easier or more reliable returns, at 37%.

Figure 3

In the study, we told them returns were identical. They didn’t buy it. They believed their own lived experience over what we asked them to believe about a new routine.

The rest of the list reads the same way. Better selection, 35%. Habit, 33%. Better customer service, 32%. Payment details they trust or already have saved there, 30% each.

A third of consumers named habit explicitly. I would say that the whole list is the habit. Returns, selection, the saved card, the membership. Those are the rewards. “Habit” is the one line where consumers happened to call it by name.

Figure 4

Now, a skeptic will point out that 40% is a lot less than the six in ten whose latest purchase was AI-assisted and closed on Amazon. Fair. Though that gap, roughly 20 points, is the prize. It’s what comes into play when consumers believe everything is equal. Today they don’t, and a survey can’t make them. Only experience will.

There’s one more interesting detail. Sixty-four percent of consumers check who the seller is when they buy on Amazon, and 28% check every time. So, they don’t trust everything on Amazon, but they do trust Amazon to fix it.

That means that trust is attached to whoever owns the bad day.

Price Is What People Say and Habit Is What They Do

A skeptic will also look at this data and say it’s a price story. Best price was the top reason for 27% of AI-assisted purchases. When we asked marketplace shoppers what would get them to buy direct, 54% said a lower price and 50% said shipping as fast as the marketplace.

A reward that sits outside the routine goes uncollected. An agent can put the lower price inside the routine.

Consumers do want the best price, and frankly, who doesn’t? My read is that the best price, for most, is the all-in cost of doing business with the merchant. What it costs, what it costs to get it here, the value of getting it today or tomorrow, and what it costs to be wrong.

Brands have been free to undercut the marketplace for years, and plenty do. The share of consumers who used to complete purchases there before AI was available (10%) is still lower today for those whose latest purchase was AI-assisted (6%). The lower price was out there. But opening the Amazon app is the routine, and a reward that sits outside the routine goes uncollected. Nobody goes looking for it. There’s no reward waiting for them if they do.

The Amazon Shopper Is Like the Cat Lady

As most of us know, it’s really, really hard to change a habit – good or bad — or form one. In his book, Duhigg cites research from Duke showing that more than 40% of what we do every day is habit, with no decision involved. That research also suggests that the brain never deletes an old habit loop. The best anyone can do is write a new one over it, and the new one has to pay off every time, or the old one comes raring right back.

Breaking or creating habits also comes down to inertia, the resistance to change. It comes in two kinds, and I’ve watched both for nearly two decades across payments and the platform economy.

The first kind protects a status quo that’s broken. That’s the skunk lady. She knew her couch stank, tried other stuff and she wanted a change. Paper checks that took days to clear. Taxis you couldn’t find in the rain. Checkout forms that made you type sixteen digits with your thumbs. Nobody loved any of them. But show people something that fixes the friction and they move. Most of the big shifts in payments and commerce have been this kind.

Every one of those shifts required two things. First, someone took the risk out of trying, and someone offered a reward big enough to beat inertia. In the late ’90s and early aughts, typing a card number into a website felt reckless until the card networks promised you wouldn’t be on the hook for fraud.

Then Amazon Prime arrived in 2005 and made shipping fees disappear. The prepaid membership became its own cue. You’ve already paid, so why not start there since shipping and two-day (back then) delivery was free. Zappos made returns free both ways and turned the living room into the fitting room. Uber made payment invisible. You just got out of the car. Those rewards were the second job. Nobody ran a trust campaign to convince consumers the alternative was better.

The Amazon shopper doesn’t smell anything wrong. The consumer knows exactly what happens after they press Buy Now, because it has happened hundreds of times.

The second kind of inertia protects a status quo that’s very good. That’s the cat lady, and that’s what agentic commerce is up against. The Amazon shopper doesn’t smell anything wrong. Buying there is certain. It’s consistent. It’s the standard every other online store gets measured against and has for years. The consumer knows exactly what happens after they press Buy Now, because it has happened hundreds of times.

Now, run that through my FIT framework. Friction asks whether the status quo is enough of a hassle that people want out. Inertia asks whether the new thing is so much better that a critical mass will leave what they know. Time asks whether it gives them enough of their day back.

For consumers and agentic commerce, discovery was the easy hurdle to overcome. Hours of tabs became one good answer to a more detailed query. Checkout hasn’t. The friction in buying on Amazon today is close to zero. And buying on an AI model isn’t much of a thing.

For shopping, buying through Amazon is a sure thing that’s already very good. Excellent even. On the other side, buying through an AI model is a lot of maybes. Maybe the price is better. Maybe it shows up on time. Maybe the return is easy. Maybe when I get to the page there will be inventory in my size and favorite color. Maybe customer support will answer the call when it goes wrong.

Behavioral economists have been measuring these maybes for decades. People feel a loss about twice as hard as they feel a gain of the same size. This is why claims of “smart and seamless” won’t be enough. Smart and seamless describes the status quo. It’s odor removal for someone who can’t smell the odor. And introduces a downside risk that they’d rather not take a chance on just yet.

The Bet Is That the Loop Can Close Without Amazon

It would be easy to call Instant Checkout the agentic commerce final verdict. I think it was the test market.

The relaunch is underway, and it has bigger names behind it. The biggest players in payments, the AI platforms and a long list of merchants are connecting agents to merchant checkouts so the same agent that finds the product can buy it, start to finish.

The pieces are familiar. A way for merchants to tell a real agent from a bad bot. Tokenized credentials. Guardrails the consumer sets. How much the agent can spend, where it can shop and when it has to ask first. And behind all of it, decades of fraud tools and dispute rules.

Lots of merchants are in. Amazon’s store still isn’t.

Merchants and platforms believe they can close the loop without Amazon, with a reward good enough to break the Amazon habit. Then again, they’ve been trying that for years. This time they are touting a product catalogue that is cherry-picked for this new front door.

My payments friends should hear this next part. I’ve written that top of wallet becomes top of agent, and I still believe it.

Like Febreze, a safe payment is scentless. Consumers expect it, the way they expect a clean house to smell like nothing. The PYMNTS Intelligence data shows it. Trusting the site with payment details was the top reason for 9% of AI-assisted purchases. Saved details was 7%. Paying with a saved digital wallet on a brand site would move 24% of marketplace shoppers. An AI assistant handling the checkout would move 12%. And our controlled test assumed the agent works end to end. The marketplace still came first.

The cue is there. The routine is there, and this work lets the agent finish what it starts. What nobody has attached yet is a reward the consumer can feel.

So, is end-to-end on a frontier model good enough? It’s enough to “maybe” make the first try safe, and that doesn’t count for nothing. A spend cap and an “ask me first” rule are training wheels. They take the risk out of the first try. That’s the first job. The craving has to come from somewhere else.

Here’s the good news for the people building it. Remember the lower price nobody went looking for? Look again at what would move a marketplace shopper. A lower price, 54%. Shipping as fast, 50%. Returns as easy, 35%. An agent with a checkout it can use anywhere is the first tool that could deliver all three, inside the routine. It can put the best all-in price from any merchant in front of the consumer at the moment they’re ready to buy. And with the right promise behind it, it can make every store’s bad day look like Amazon’s. Especially since what consumers are asking for is a store that behaves like Amazon.

The 12% measures the agent as a shopping feature. The 54% measures what the agent could deliver.

Read More: 2026 Global Digital Shopping Index

P&G borrowed its cue. It found a routine people already had and put the reward at the end of it. Agentic commerce has that routine. Sixty-one million people already ask AI what to buy. The answer on the screen is the freshly cleaned room. The reward has to show up right there.

Agentic checkout today is more like Febreze before the perfume.

Stores Need More Than a New Intermediary

Here’s the part that has to pass the sniff test, pun intended.

The stores are losing. In our controlled test, with everything equal, the department store got 22% and the brand got 16%. You can understand why merchants are lining up to connect to agents. A new front door feels like a second chance.

We’ve seen this movie before. For twenty years, Google has been the front door for product search, and the stores paid for the clicks. Consumers searched there and bought on Amazon anyway. Google spent a decade trying to close the loop itself, with a buy button and a shared cart across merchants. It shut Buy on Google down in September 2023 and said only a limited number of merchants ever used it. Social commerce made the same promise. In our data, its share of purchases has been cut in half. Jet.com bet that a clever pricing engine could pull shoppers off Amazon. Walmart paid $3.3 billion for it in 2016 and closed it four years later.

Every one of those was a new intermediary. Every one moved discovery. None moved the buy, because none of them changed what happened after the order was placed. The box still came slower. The return was still a hassle. The consumer still wasn’t sure. The available inventory wasn’t good enough.

The intermediary can open the door. The reward has to be the merchant’s own.

An agent is a much better intermediary than any of those. But it can only carry the reward a merchant actually delivers. It can’t make a slow shipper fast or a painful return easy. An agent working for the buyer when comparing the all-in offer will often point right back to the marketplace.

So, the ask of the buyer is big. Merchants are asking consumers to give up a sure thing for a maybe. Duhigg’s golden rule says the reward has to stay when the routine changes. That means match Amazon on the bad day. Beat it on the all-in price. The intermediary can open the door. The reward has to be the merchant’s own.

What Has to Break and What Habit Has to Form

For agentic commerce to deliver, three habits have to break. The reflex to start at Amazon when it’s time to place the order, which is anchored by a saved account and a prepaid membership with free same-day or overnight delivery. The belief that only the marketplace makes a bad day painless. And the need to check. Consumers still look at the seller. They still want to press the button themselves.

Then two new habits have to form. Letting software spend the money without looking over its shoulder. And believing someone will make it right when the agent gets it wrong. Wrong size, wrong seller, wrong week.

The door is open a tiny crack. Twenty-two percent of consumers PYMNTS Intelligence studied would let the AI pick the store. And among those who used AI for their purchases, 45% say shopping with AI has already led them to buy from a store they’d never used before.

My guess is that the new habit starts where a miss is cheap. Clothing is 28% of AI-assisted purchases and beauty is 12%. Those are repeat buys, and repeat buys are where habits get their reps and trust can be built.

Someone Has to Own the Bad Day

So, who closes the agentic commerce loop?

Amazon could close it for itself, and that seems to be what it’s doing. Its own agent inherits the cue, the reward and the returns counter. It’s also bringing outside merchants into its closed ecosystem with Shop Direct and Buy for Me. It’s third-party seller network already drives 60% of physical product sales on Amazon, so the selection is there. Cue, routine and reward all live under one virtual roof.

Some big merchants understand that they have to bring their own reward with them. Walmart’s answer to the end of Instant Checkout was an app inside ChatGPT with account linking, loyalty and its own payments. It carried the old habit into the new room.

The other candidates are the networks and issuers, and they start with something nobody else has. They already own the bad day. Zero liability made a generation comfortable shopping online. The chargeback is the original “someone will make it right.” That comes along for the ride every time an agent pays with a card.

What’s missing is the part the consumer can see. Today those protections are the plumbing. The agent version is a promise that travels with the credential wherever the agent shops. Wrong item, wrong seller, refund that never shows. That makes a brand site, or a store nobody’s heard of, as safe to try as the marketplace. And why people shop online and in-app without a second thought.

That promise is what makes the first try feel safe. The perfume is what the consumer gets for making it. The best all-in price from any store on the web, with the bad day covered. Picture a shopper saying, “I paid less, and when it was wrong, I had my money back before I asked.” That’s a reward a consumer can feel.

It’s also the big bet. Six in ten of today’s AI-assisted orders land at a store that isn’t playing. A new intermediary gets the stores a new front door. Only a better reward, delivered every time, gets them the habit.

A new intermediary gets the stores a new front door. Only a better reward, delivered every time, gets them the habit.

P&G learned the hard and expensive way that no one craves scentlessness. No one craves checkout either. People crave the box on the porch, the right thing inside it that arrives on time in perfect condition and their money back without a hassle when it isn’t.

Whoever attaches that feeling to an agent rewrites the next habit loop. Until then, consumers will keep asking AI what to buy.

And then they’ll buy it on Amazon.