As Al shopping agents take on more of the ordering, Tom Fitz-Walter, Managing Director UK & Europe at TMX Transform, looks at what it means for supply chains built to serve human shoppers, and how retailers can rebuild them around data rather than persuasion.
A family of four needs feeding for the week. Someone has a dairy intolerance, someone else is training for a marathon and wants more protein, and nobody has time to think about it.
Increasingly, the person doing the ordering isn't a person at all. They tell an Al assistant the constrains - household size, dietary needs, budget, days to cover - and the assistant builds the basket, checks what’s already in the fridge, and places the order.
This isn’t a hypothetical. Tesco is piloting exactly this: an Al assistant, built through partnerships with Adobe and Mistral AI, that takes a shopper's dietary preferences, purchase history and current fridge contents and generates a meal plan with a basket attached.
In Australia, Coles CEO Leah Weckert expects roughly 30% of online orders to be placed by Al agents within five years. In the UK, 58% of retailers told the Payments Association that agentic transactions are already hitting their platforms.
The retail conversation has understandably focused on what this means for marketing, loyalty and conversion. But there’s a more structural question underneath it: what happens to the supply chain when the person doing the choosing is a machine?
The second customer
For most of retail history, supply chains have been built to serve a browsing, comparing, occasionally impulsive human. Packaging sold the story. Delivery windows were a nice-to-have. Brand loyalty smoothed over the odd stockout or late parcel.
Al agents don’t browse and they don’t forgive. They read APIs, not shelf-talkers. They compare delivery speed and stock accuracy against dozen retailers in the time it takes a human to open one tab, and they have none of the emotional pull – the loyalty card, the familiar aisle, the ad they half-remember – that keeps a person choosing the same retailer out of habit. If the data says a competitor can deliver the same order a say sooner, the agent moves on. There’s no complaint, no lost customer conversation to have. The order is just gone.
That’s the shift supply chains now have to plan for: a second customer, sitting alongside the human one, that evaluates retailers entirely on machine-readable performance.
What retail looks like when the shopper isn't shopping
Once ordering shifts from browsing to delegation, the competitive battle moves upstream. Winning a human shopper is about persuasion – price, promotion, presentation. Winning the agents' order is about being selectable: having delivery terms, stock levels and product data structured clearly enough that an agents can evaluate the offer without a human every seeing the page.
Grocery is arguably where this bites hardest, because it’s high-frequency and low-consideration – exactly the kind of repeatable, planning-heavy shopping agents are best suited to.
What an agent-legible supply chain requires
Meeting this second customer means rebuilding parts of the supply chain around data, not persuasion:
- Real-time, API-exposed stock accuracy. An agent won’t wait for a retailer to ‘check and get back to you.’ If inventory data isn’t live, the agent rules the retailer out
- Verifiable delivery ETAs. Agents compare promises against actual performance history, not marketing copy. A retailer whose ‘next day’ is unreliable loses orders it never gets to fight for
- Structured, machine-readable product data. Spec sheets, dietary and nutritional information, certifications – these need to sit in query able fields, not buried in a PDF or marketing paragraph
- Traceability and provenance data. As agents start weighing sustainability or origin claims on a customer’s behalf, that information needs to exist in structured form
- Genuine API/integration readiness. Agent-to-agent commerce depends on a retailer’s backend system being able to hold that conversation at all – not just present a website
Retailers who get this right are already seeing operational gains beyond the sale itself: agentic AI is helping supply chains react faster to disruption and unlock working capital that used to sit tied up in buffer stock and manual reconciliation.
Where the infrastructure work is already happening
As part of its £800 million transformation programme, British department store John Lewis has extended its partnership with commerce tools – an agentic commerce platform – specifically to make its product catalogue transactable inside AI apps like Gemini and ChatGPT. That’s a data and integration decision, not a marketing one: it’s the retailer restructuring its catalogue so an agent can act on it.
Debenhams Group’s PayPal partnership gets the attention for letting customers discover, get recommendations and check out entirely inside the PayPal app. But the more telling move sits alongside it: a parallel partnership with Peak AI to sharpen real-time forecasting across stock, sales and pricing – the data lineage an agentic checkout needs to avoid quitting stock or delivery promises it can’t keep. Get that backend wrong, and the agent simply won’t complete the order.
Both point to the same lesson: agent-facing shopping experiences are only as good as the supply chain data feeding them.
The businesses that will win the machines’ orders
None of this replaces the human shopper – most people will still want to choose their own dinner some of the time. But a meaningful and growing share of orders are now being placed by something that reads data instead of packaging, and rewards infrastructure instead of brand feeling.
Most supply chains are not yet built to serve that customer. Retailers that treat stock accuracy, delivery reliability and structured product data as core infrastructure – not IT projects – will be the ones agents keep selecting. The rest may simply be ignored.
This article was originally published CILT's September issue.
