“Agents Will Be Doing The Buying” And Then Nothing Happens On Monday
The line turns up in every deck now. Agents will shop for people, the funnel changes, we should be ready.
I’ve stopped flinching at it. What I still can’t do is turn it into something I’d put on a sprint board, because the same week somebody says it, the revenue report shows no agent traffic at all and both of those things are true at once.
So the conversation collapses into which half you believe, which is a conversation nobody wins.
But there’s a survey that turns it back into a question about timing, which is the kind of question a marketer can actually answer.
How Many Shoppers Would Actually Let An AI Buy For Them
Four in ten, if it’s groceries. In a June 2026 study by Checkout.com, 41% of consumers said they’d be comfortable letting an AI agent buy their groceries, and 31% said the same about everyday essentials — while agents are actually involved in 3% of transactions today.
Sit with those two numbers next to each other for a second.
The willingness is already built, and the buying hasn’t followed it yet. Which means the moment to do anything about it is now, while the gap is still open — waiting until the intent number matures is waiting until everyone else has finished.
The Survey Put Shoppers And Merchants In One Study
Checkout.com sells payments infrastructure, and the design of the study is refreshingly plain: put consumer willingness and merchant readiness inside one piece of research, so the two sides can be read against each other rather than stitched together from separate press releases.
The spending-cap questions run across six markets; the merchant figures cover the US and UK only. The report says so itself.
Total sample size isn’t published, which is the sort of thing worth subtracting before you quote any of this in a meeting.
What comes out of it is a clean sort of the catalogue:
- Comfortable letting an agent buy groceries: 41%
- Comfortable letting an agent buy everyday essentials: 31%
- Average spend they’d allow without approving it first: £177 per transaction
The £177 is the line that ties the other two together. Cheap and repeated goes to the agent; expensive and occasional stays with the human. People want to decide what’s for dinner and they’d quite like somebody else to remember the milk.
If what you sell is low-ticket and bought again next month, you’re on the delegated side of that line.
Merchants Can Feel The Clock, And Say So
Here’s the part I didn’t expect from the sell side. 72% of merchants agreed that consumers will adopt agent-led shopping faster than most merchants are ready for.
They know. And 89% of them also said they’re actively preparing, which reads as a slightly uncomfortable sentence when you put it next to the first one: we’re getting ready, and we expect to be overtaken anyway.
Set that against the 3% of transactions agents touch today, and the space between 3% and the 41% who say they’re willing is the preparation window, stated as a number rather than a vibe. Nobody gets to have that twice.
The useful work in that window sits earlier than the checkout plumbing. Before an agent can buy from you it has to know what you sell and be able to say it back accurately, so the cheap move right now is counting how the models currently describe you — while the answer still has time to change.
How Far These Numbers Travel
Four things to hold before you carry any of this into your own forecast.
- Checkout.com sells payments infrastructure, so it has a commercial interest in agent-led buying becoming normal
- Total sample size isn’t published, and the merchant figures come from US and UK data only
- The line that 10% of purchases will be AI-led within a year (33% of consumers said so) is a consumer prediction, not a measurement
- Agents are in 3% of transactions today; the willingness figures are not a market size
The trust side deserves the same precision. 24% said they’d never delegate a purchase to AI, and 27% said they don’t trust any organisation running AI shopping agents at all.
That’s a real floor, and I don’t think it moves quickly. But a quarter refusing and four in ten willing can both be true, and the four in ten go first — which is the bit that decides what you do this quarter.
Place Your Products On Price And Repeat Purchase
If I keep one thing from this, it’s the order of the questions rather than any single percentage.
Start by plotting what you sell on two axes: ticket size and how often somebody buys it again. Land toward cheap and repeated and you’re in delegation territory, near enough that £177 to plan around it.
For anything sitting on that side, what an AI says about you when it’s assembling a basket is the thing that decides whether you’re in the basket. With agents at 3% of transactions, there’s still slack in the schedule to go and fix what they say — which is not a sentence I expect to be writing in a year.
One step earlier than “does the agent pick us” is “does the agent know we exist”, and I’d sort that out first. Challenger Brands Convert Best Once The AI Shortlists Them is the companion piece on that stage, and reading the two together makes it easier to work out which problem you’ve actually got.
Anyway — no rush, but not no hurry either.
