How AI Shopping Agents Remember Shoppers Across Retailers

10 min read

AI shopping agents remember desires. Retailers log receipts. The race to bridge that gap is rebuilding e-commerce.

TLDRTap for the short version

General assistants may remember stated preferences, while retailer assistants can use account and order history inside their own stores.
eComID tries to carry a shopper identity across participating brands. True Fit supplies specialised fit answers to retailer-owned and outside agents.
The open question is whether these separate layers can improve an independent agent’s decision across retailers. Public evidence has not established that outcome.

I asked ChatGPT to buy me a pair of trousers. Given our chat history, it already knew I prefer relaxed clothes and usually avoid anything too formal. Not bad for something that’s only ever read my messages. It was all helpful until the real problem of size showed up. I wear medium at one brand and large at another, and I once sent back a pair that measured fine at the waist and still sat wrong on the length. Every store I’ve ever bought from is sitting on its own slice of my history, and none of them talk to each other.

The shops know different versions of us

Anyone who shops fashion online regularly already gets this in their bones. Head to any fashion subreddit and you’ll find people swapping measurements, reviews and return horror stories like war stories, because that’s what it takes to shop cross-brand without wasting money.

Amazon’s Andy Jassy on the company’s October 2025 earnings call said, that third-party agents showing up on Amazon get “no personalization” and “no shopping history”. Amazon’s own assistant can use Amazon’s data. Everyone else starts from zero. OpenAI says its shopping research can pull preferences from earlier chats. That helps. It still cannot replace a verified order, a documented return reason or a record of which size worked on my body.

Retailers build a picture of you from purchases, searches, sizes and returns, but only for their own store. Walk into a new one and you’re a stranger again, no matter how good their AI stack is. Sharing customer data with a competitor means new headaches around consent, security and who’s liable when it’s wrong. Shopping agents raise the stakes here, because now we’re expecting software to do that comparison shopping for us. It can talk a great game, tell me why a product suits my taste, and then land on the one call most likely to get the whole order returned. It either stops and asks, or it guesses and hopes.

The information is scattered across systems that weren’t built to talk to each other. Chat history knows what you said. Order systems know what you bought. Return systems log that something came back, though rarely why in a useful form. Product catalogues cannot even agree on what a size label means. An agent needs permission to reach those records and a reliable way to reconcile them.

Who remembers the shopper?

Assistant remembers preferences Retailer remembers purchases Context layer tries to connect them

Each system holds part of the shopper. None holds the complete picture.

A Swedish study tracked nearly 500,000 items. Shoppers who used a size finder were 0.65 percentage points more likely to return something, although the tool ranked among the weakest predictors of returns. Those shoppers also generated 7.5% more customer lifetime value the following quarter.

Two very different bets on fixing it

Stockholm-based eComID is going after cross-store recognition directly. Its Shopping Passport is meant to carry your size, preferences and return signals between brands that sign up for the network. Its assistant, Vera, lives on retailer sites, and other tools in the stack nudge shoppers away from ordering three sizes just to be safe.

The model becomes more useful as more retailers join. Brand B benefits from what Brand A knows only when both are in the network, the shopper has opted into a shared profile and the information applies to the purchase. This is no universal memory layer plugged into ChatGPT or Gemini. As of August 2026, eComID’s own sign-up page still said direct account creation was coming soon. True Fit, out of Boston, is playing a narrower game. Retailers can bolt its recommendations onto product pages, hook into its API, or run its Fit Agent directly. Its newer MCP layer is built specifically so an outside agent can ask for a size and a confidence score without getting handed the shopper’s raw data.

These services could sit beside a general assistant. ChatGPT remembers that I like loose trousers, True Fit calculates the size and eComID recognises me at the next store in its network. The chain works only when the retailer participates, the shopper grants access, the product has fit data and the agent knows which service to call.

eComID says it works with more than 60 brands and reaches 20 million shoppers a month, with users returning 30% less and participating brands converting 10% more. These are company-reported aggregates. There is no independent audit or named retailer result to check. H&M Group is an investor and its brands appear in eComID’s materials, but I could not find a public figure linking an H&M outcome to the product.

True Fit’s evidence is more specific, even if it’s older and predates the current wave of shopping agents. UK retailer M&Co reported a 1.5 percentage point lift in sitewide conversion and a 27% jump in average order value among registered users, with a 9.8% drop in returns for shoppers who bought the recommended size, tracked over six months.

Lands’ End Europe reported a 5.4% incremental sitewide revenue lift, with registered users converting roughly twice as often as other visitors. A separate Google-published case says Fit Analytics produced conversion gains of up to 11.5% and reduced returns by 4.4%. Every performance figure in these cases came from the company providing the service.

Together, these cases support a narrow claim. Fit guidance can improve results once a shopper is inside a participating store, although people who use a size finder may already be more engaged than the average visitor. They do not show an outside agent carrying someone’s history into a new store. I found no published case combining conversational memory, permitted transaction history and fit intelligence across several retailers, then measuring whether it produced a better purchase.

Evidence	What it supports	What it cannot support
eComID network aggregates	Possible network-level commercial value, and the network model most directly aimed at cross-retailer context	Named, verified retailer-level outcomes
M&Co (True Fit)	Fit guidance can improve store-level conversion and returns	Shopping-agent performance across retailers
Lands’ End Europe (True Fit)	Personalization can increase a retailer’s own revenue	Independent verification of the figure
Fit Analytics (via Google Cloud)	A second vendor reports similar retailer-level conversion and return gains	Independent verification because the figures originate with the provider
Jassy’s earnings-call comments	Third-party agents currently lack shopping history and personalization	A measured cost of that gap to any specific retailer

The evidence is useful only when we know who used the tool, how the comparison group was chosen, whether the return window was long enough and whether the lift held for first-time visitors. An agent integration adds a harder test.

A bad memory can travel too

Shared context can spread a useful signal or a bad assumption. A larger size might be intentional. A return could reflect late delivery rather than fit. An unusual purchase might be a gift. Retailers outside a network remain strangers, and a profile built from fashion purchases has little to say about groceries or furniture. We are unlikely to get one complete shopper profile. Several companies will hold overlapping, partial versions instead.

The larger stake goes beyond returns. If Amazon, Google, OpenAI or another platform combines conversational memory with activity across several stores, it may understand the shopper better than any retailer does. A store sees the final visit and purchase. The platform can see the wider search, the products rejected and the reason one item won.

That gives the platform leverage over which details reach a store and which products enter a recommendation. Retailers may receive a better-informed shopper while giving up part of the customer relationship. The platform also has an incentive to favour its own products or the retailer offering the best commercial terms. No general shopping agent has yet shown it can manage this reliably.

GDPR gives people a conditional right to data portability. It does not give retailers an automatic route into one another’s customer data. A shared profile still needs a lawful basis, a stated purpose, meaningful controls and a way for someone to withdraw consent.

Permission granted at one store doesn’t automatically cover a different retailer, a new product category, or ad targeting down the line. Ask for consent too often and people just tune it out. Ask too broadly, in one blanket swoop, and shoppers lose any real control over what’s happening with their data. People deserve to see when an old return is quietly shaping a recommendation, the ability to strip it out, and a say in which service gets to use it at all.

A useful test would give several agents the same measurements, preferences, cross-brand sizes and return reasons, then vary the amount and order of that context. Because the correct answer is known in advance, we could see whether more context improves the recommendation or makes a wrong answer sound more confident. Proving an effect on conversion or returns would still require real shoppers and an independent analysis.

Partial recall is the market, for now

Shopping agents know us in fragments. Retailer assistants know our accounts. General assistants remember our conversations. eComID’s trying to carry an identity across its network. True Fit’s solving the fit call specifically. The real business opportunity is in the handoff between all of these pieces, not in any one of them alone.

I found no published case in which an independent assistant entered a new store, drew permitted context from these different layers and demonstrated that the shopper made a better purchase. The strongest businesses in this market will connect those memories without allowing bad assumptions, weak consent or excessive platform control to create a larger problem.

Agent product ratings, on how shopping agents interpret reviews and ratings.
Returns confidence gap, on why sizing uncertainty still drives avoidable returns.
Agentic checkout gap, on the distance between retailer readiness and shopper adoption.

FAQ

What context can shopping agents remember?

General assistants may remember preferences stated in conversations. Retailer-owned assistants can use account and order history inside their own stores. Neither source automatically includes verified purchase, return and fit history from unrelated retailers.

Why can’t retailer history follow the shopper?

Retailers store transactions in separate systems and use different identifiers, product data and return codes. Sharing those records also requires permission, security controls and agreement over how another business or agent may use them.

How do eComID and True Fit differ?

eComID is building a shared shopper identity and on-site assistant across participating fashion retailers. True Fit provides specialised size and fit recommendations through retailer tools, APIs and its MCP layer. eComID depends on network participation, while True Fit remains focused principally on apparel and footwear fit.

Has cross-retailer agent shopping been proven?

Public cases show that fit guidance can improve outcomes inside participating stores. No published case yet shows an independent assistant combining conversational memory with permitted purchase, return and fit context across several retailers and producing verified commercial gains.

Contact [email protected] with corrections or reporting leads.

Partial Recall and the Business of Teaching AI to Remember

The shops know different versions of us

Two very different bets on fixing it

A bad memory can travel too

Partial recall is the market, for now

FAQ

What context can shopping agents remember?

Why can’t retailer history follow the shopper?

How do eComID and True Fit differ?

Has cross-retailer agent shopping been proven?

Like this:

Partial Recall and the Business of Teaching AI to Remember

The shops know different versions of us

Two very different bets on fixing it

A bad memory can travel too

Partial recall is the market, for now

Related reading

FAQ

What context can shopping agents remember?

Why can’t retailer history follow the shopper?

How do eComID and True Fit differ?

Has cross-retailer agent shopping been proven?

Share this:

Like this:

Discover more from Industry Contents