Every grocer will end up with an AI shopping channel. Very few are choosing who owns it.
That sounds dramatic. It is really just arithmetic. Shoppers are starting to describe what they want instead of searching for it, and something has to answer. That something is either a surface you control or a surface someone else controls. The technology question gets all the attention. The ownership question is the one that will still matter in five years.
The market is moving faster on the retailer side than the shopper side
Two numbers frame the moment, and they point in opposite directions.
On the retailer side, adoption is already mainstream. 68% of food retailers told FMI’s 2026 Food Retailing Industry Speaks study that they use AI in the business, up from 47% a year earlier.
On the shopper side, it is early. Only about 15% of US consumers say they use AI when buying groceries, according to Dunnhumby research reported by Grocery Dive. Most of the rest say they do not see the need, or prefer to decide for themselves.
The gap between those two numbers is the whole opportunity. Retailers are investing ahead of demand, which means the channel that shoppers eventually adopt is being built right now, mostly out of view. Morgan Stanley puts the size of the prize at $190 billion to $385 billion of US e-commerce by 2030, with groceries and consumer packaged goods among the largest drivers. Those are forecasts, not facts, and forecasts of new channels are usually wrong in the details. The direction is the useful part.
Renting is a real option, and it has real advantages
We are not going to pretend that renting is foolish. It is often the fastest path to reach.
A marketplace already has the shoppers. An assistant embedded in your current storefront ships in weeks and requires almost nothing from your team. Both are legitimate answers to the question of how a grocer participates in AI shopping, and for some operators they are the right answer this year.
What renting costs is not speed. It is position.
When the intent lands somewhere else, you see the order and not the reasoning. You learn what was bought, not what was asked for, swapped, or given up on.
When the interface belongs to someone else, so does the ranking. The rules that decide which products a shopper sees are set on the other side of the table, and they can change without you.
When the relationship is mediated, the loyalty attaches to the mediator. Shoppers get loyal to the thing that made the week easy.
When the channel is a feature of someone else’s product, its roadmap is their roadmap.
None of that shows up in the first year. All of it shows up in the fifth.
Owning does not mean building
The reason most grocers rent is that owning sounds like a multi-year platform project. It usually is not, and conflating the two is the most expensive mistake in this category.
Owning the channel means the AI storefront carries your brand, connects to your catalog and your pricing, hands the order to your checkout, and reports to you. It does not mean writing the agent, the grocery logic, and the evaluation harness yourself. Those are different decisions, and only the first one is strategic.
It also does not mean replacing the store you run today. A standalone AI storefront can sit alongside your existing online store as a second front door to the same catalog. Nothing is ripped out. The shopper who wants to browse still browses. The shopper who wants to say “dinners for the week, no dairy, under a hundred dollars” finally has somewhere to say it.
The uncomfortable part
Here is the honest state of the evidence, including ours.
Agentic grocery shopping is a contested space. Several companies are building versions of it, some of them at large national banners, and anyone telling you they are the only or the first is selling you something. We are not going to make that claim.
Our own proof is one pilot, with one grocer, and we describe it as directional rather than proven. In that pilot, shoppers reached a finished cart in a median of roughly two and a half minutes, with add-to-cart around 88%, and session-to-order conversion that ran far above the store’s normal e-commerce baseline. That is a genuine signal. It is not a promise about your traffic, and we will not present it as one.
Which is exactly why the ownership decision should be made through a contained pilot rather than a committee. You do not need to believe a vendor’s numbers. You need your own.
What we would actually do in your position
Run the cheapest experiment that answers the ownership question rather than the technology question.
Pick a defined slice of traffic. Put an AI storefront in your brand in front of it. Agree up front what success means: conversion against your existing baseline, time from stated need to finished cart, basket value, whether anyone comes back. Then look at what shoppers asked for, which is the part you never get when you rent.
If it works, you own a channel. If it does not, you spent a pilot and learned what your shoppers actually want from AI, in their own words. Both outcomes beat waiting to see what the market decides for you.