What is agentic commerce? A practical guide for UK retailers

Around 45% of UK respondents in McKinsey's December 2025 consumer survey said they use AI tools when shopping for fashion, ahead of Germany at 30% and just behind France at 50%. That is happening in a market where online sales accounted for 28% of all UK retail in the first quarter of 2026, according to the ONS.

Imagine a customer looking for waterproof walking boots under £100. A few years ago, they would probably have searched Google, opened several retailer websites, read reviews and spent 20 minutes comparing prices, stock availability and delivery dates before deciding where to buy.

Now, an AI assistant does most of that comparison instead. The customer asks it for a recommendation, adds a few constraints such as price, size and delivery window, and receives a shortlist in seconds. Most of the evaluation has already happened by the time they click through to a retailer. The competition has moved from ranking on a search results page to being selected by an AI system that has already filtered the market on the customer's behalf.

What is agentic commerce?

Take that shopper looking for waterproof boots again. Ask an assistant to find one under £100, in stock, able to arrive within the week, and it checks stock across several retailers, rules out the ones that cost too much or cannot deliver in time, and comes back with two or three real options rather than a single answer. That is the difference between a chatbot and an agent. A chatbot would answer one question and stop there. An agent keeps going until the job the customer actually wants done is done, buying included, if that is what it has been asked to handle.

This is what is meant by agentic commerce, an AI agent carrying a purchase from intent through to completion, rather than a person doing that work manually. The industry has not settled on one tidy definition of it. McKinsey has estimated that between $3 trillion and $5 trillion of revenue in global business-to-consumer retail could be orchestrated through agentic commerce by 2030, a range wide enough to show how much is still being argued over. Customers are handing over more of the buying decision than they used to, whichever definition a retailer prefers.

How does agentic commerce work?

An assistant needs to do three things before a purchase actually happens: understand what someone wants, place an order with a retailer it has never dealt with before, and prove to a payment network that it genuinely has permission to spend the money. Each of those has been built by a different company, and the past year has been a live test of whether they hold together.

OpenAI and Stripe took on the ordering problem with something called the Agentic Commerce Protocol. By September 2025, ChatGPT users in the US could buy directly from Etsy sellers without leaving the chat window, and Shopify merchants followed. Google shipped its Agent Payments Protocol the same month, with more than sixty partner companies signed up at launch, including Mastercard, PayPal and American Express. The point of a protocol like this is that an assistant can place an order with a retailer running completely different systems and have it processed.

Payment is the harder problem, because every card network assumes a person is physically there to approve a purchase, tapping a card or typing a one-time code. An assistant cannot do either. So Visa built something called the Trusted Agent Protocol, which lets a retailer check that the assistant is genuinely who it says it is, and that the customer actually agreed to the purchase, a check that exists because bots already turn up at checkout pretending to be shopping assistants when they are really just scraping prices and stock. Adobe data cited by Visa put the growth in AI-driven traffic to US retail sites at more than 4,700% over the previous year. Mastercard built something similar earlier, in April 2025, called Agent Pay. It builds on the card tokenisation Mastercard already runs, so an assistant holds a limited credential instead of the card number.

OpenAI withdrew Instant Checkout from ChatGPT in March 2026 and moved purchases to retailer apps instead. Take-up had been limited, product selection remained narrow, and the information customers saw was not always up to date. Development on the protocol carried on without the in-chat checkout it was originally built to support.

Why this matters for UK retailers now

The UK is already one of Europe's most digitally mature retail markets, which makes it one of the first places where changes in shopping behaviour become visible. The same McKinsey survey found that 84% of respondents across the UK, France and Germany already use AI somewhere in their daily lives, and 38% use it to research products or help decide what to buy. UK usage runs ahead of Germany across most categories, consistent with a market where a larger share of retail has already moved online.

The research also points to where this goes next. McKinsey expects adoption to expand first in bounded cases such as decision support and narrowing options, before extending into ongoing or fully autonomous execution. OpenAI's retreat from in-chat checkout points the same way. AI is changing the research stage well before it changes the checkout, and the work that once involved searching Google, opening several retailer websites and reading pages of reviews can now happen inside a single conversation.

That puts more weight on product information than it used to carry. An assistant can only recommend what it can read and compare, so a retailer whose data is patchy or inconsistent may not reach the shortlist at all. Everything else that retailer does well stops mattering at that point, because the customer never gets far enough to see it.

Where can retailers use it today

The easiest place to begin is with product information, because every AI shopping assistant depends on it. Before it can recommend anything, it has to understand what the product is, what it costs, whether it is in stock, how quickly it can be delivered and how it compares with similar options. Most retailers already maintain much of this information for Google Shopping, marketplaces and comparison sites, so the work itself is familiar, though the standard is higher. A customer can look at a few photographs, read a description and work out that a jacket is waterproof. An AI assistant cannot. If the attribute is missing or buried in inconsistent data, the product is less likely to be recommended in the first place.

The rules for being discovered inside AI assistants are much less settled. Different assistants draw on different combinations of retailer websites, marketplaces, reviews and other public sources, and those combinations continue to change. Anyone claiming to know exactly how to rank inside AI search is claiming more certainty than the industry has today. The safer approach is to make sure product information is complete, accurate and consistent wherever it appears. Conflicting prices, outdated marketplace listings or missing product attributes make it harder for an assistant to recommend a product with confidence.

Retailers are already making different choices about where AI belongs in the shopping journey. Should it become part of the experience on a retailer's own website, or should it meet customers wherever they are already shopping? Some retailers are investing in conversational shopping within their own websites and apps, helping customers search, compare products and make buying decisions without leaving those channels. Others are making their products available through assistants such as ChatGPT, Gemini and Perplexity, reaching customers before they ever arrive on the retailer's own website.

Frasers Group has taken both routes. In October 2025 it announced that customers of Sports Direct, FLANNELS and FRASERS would be able to discover and buy products through ChatGPT, Gemini and Perplexity, making it one of the first European retailers to launch a full agentic commerce experience. By April 2026 it had also launched Ask Frasers, an AI shopping assistant on the FRASERS website, and reported conversion rates up by as much as 25% compared with traditional search. JD Sports announced its own move at NRF in January 2026, becoming the first retailer to adopt Stripe's Agentic Commerce Suite, connecting AI-driven discovery through large language models and assistants to its existing pricing, inventory and payments systems.

The two approaches are trying to achieve different outcomes. One improves the experience for customers who have already found the retailer. The other helps retailers become part of the customer's shortlist much earlier in the buying journey. Most retailers are likely to end up doing both. The immediate decision is where AI can have the greatest impact.

What retailers need to prepare

Agentic commerce relies on systems retailers already use every day. Product data, payments, website policies and the customer journey all influence how AI assistants discover products, compare options and help customers buy. They are a good place to start when preparing for AI-assisted shopping.

1. Product data

OpenAI has already published the information it expects merchants to provide, including product identifiers, descriptions, pricing, inventory, images and fulfilment options. Most retailers already have this information somewhere in the business. The focus is on making sure it is complete across the catalogue, consistent wherever a product appears and accurate when an AI assistant retrieves it. Customers can often work around missing information or vague descriptions. AI assistants are much less likely to.

2. Payment providers

A conversation with the payment provider is usually the best place to begin. It is worth asking which agent protocols they already support and what enabling them would involve. Under the Agentic Commerce Protocol, the retailer remains the merchant of record, continuing to handle payments, tax, fraud checks, fulfilment and returns in the same way as today.

3. Website policies

Many website policies were written when most automated traffic came from scrapers or bots that retailers wanted to block. AI shopping assistants introduce a different type of visitor: software acting on behalf of a customer. Reviewing those policies helps retailers distinguish between the two, allowing legitimate assistants to access product information while maintaining protection against unwanted automated traffic.

4. Customer journey

The choice between owned channels and third-party assistants becomes a practical question at this point. Each involves different work on different timelines, so it helps to be clear internally about which one comes first.

Fynd's agentic commerce solution works on both ends of this, in one platform rather than four separate systems. AI PIM, AI Photoshoot and the AI Storefront Builder deal with the information side, making sure catalogue data, imagery and the storefront itself are accurate and legible to an AI assistant. The AI Commerce Agent handles the journey, bringing conversational shopping into a retailer's own website and app.

Looking ahead

Rich, consistent catalogue data improves search visibility, marketplace listings and the experience customers have on a retailer's own website, whatever happens to agentic commerce. The same goes for talking to payment providers and reviewing website policies. Customers, meanwhile, still want the right product at a fair price, delivered reliably, with decent service when something goes wrong. Agentic commerce changes how products get discovered and compared, and everything after the order is placed matters as much as it did before.

The harder question is how much of that work can happen in one place. Product data, imagery, storefront content and conversational shopping often sit in separate systems maintained by different teams, which is how a price ends up differing between a retailer's own website and its marketplace listing.

Want to see what that looks like in practice?

Talk to us