The Agent Took Comparison. It Handed Back the Checkout.The Agent Took Comparison. It Handed Back the Checkout.

In March, OpenAI stepped back from what everyone said would kill retail.
Instant Checkout let you buy inside ChatGPT. It lasted about six months.

Merchant adoption remained limited, and Walmart found that purchases inside chat converted at roughly one-third the rate of click-outs to its own site.

OpenAI changed course. Merchants would use their own checkout while ChatGPT concentrated on product discovery.

Back in April, I said your brand was becoming a spreadsheet row and the transaction was already gone.

The spreadsheet was right. The transaction wasn’t.

One 2026 market study puts AI usage at 62% for product comparison vs 23% at checkout and 19% post-purchase.

Look at where it stalled and a pattern emerges.

Comparison can tolerate weak accountability. A transaction cannot.

A bad comparison costs confidence. A bad transaction creates a claim.

Somebody has to own the wrong size, the late delivery, the refund.

The funnel is not splitting by task.

It is splitting by accountability.

Which puts two things retailers had started writing off back in play: preference formation and recourse.

Preference formation comes first.

Agents need a specification from humans. And a human cannot specify what they have never experienced.

You cannot ask for a coat that drapes if you have never felt one that does.

Stores have an advantage here because physical experience creates the vocabulary the agent later optimizes against: fit, weight, texture, proportion, feel.

Recourse comes last.

An agent can initiate a return. It can track a parcel. It can probably negotiate a refund.

But when something goes wrong, consumers still want an identifiable party capable of making it right.

Trust is not manufactured at checkout.

It is manufactured when somebody is accountable after checkout.

Checkout.com found in June that 27% of consumers trust no organization to operate a shopping agent, while 24% say they will never delegate a purchase at all.

That is not a rounding error.

It is a trust constraint the industry cannot simply engineer around.

Agents can take friction out of commerce. They cannot take responsibility out of it.

So the transaction has come back to retailers on loan, not on merit.

That gives them a window they did not earn and probably will not get twice.

The difficulty is that the two things now becoming more valuable, preference formation and recourse, are two things stores have historically struggled to measure.

What is a store worth when the purchase happens three weeks later, somewhere else, through a machine acting on a preference that was formed on your floor?

How do you attribute:

store exposure → preference formation → agent recommendation → transaction → recourse?

If the honest answer is “we cannot measure that,” you have not found a measurement problem.

You have found the project.

#AgenticCommerce #RetailStrategy #CustomerTrust #FutureOfRetail

Contact

bruno.gentil@sherpaconsultingasia.com

© 2026. All rights reserved.