Agentic commerce: what AI shopping agents need from your delivery data

Shoppers are starting to ask AI shopping assistants to find, compare and shortlist products before they visit a single retailer. AI shopping agents read what they can find, weigh the options and recommend the retailers that appear to meet the shopper’s requirements. Delivery is part of that shopper’s request more often than retailers expect: “Can it arrive by Friday?”, “Can it be delivered to a locker near my workplace?” or “Can I return it for free if it doesn’t fit?”

Use this audit to check whether the delivery information on your product pages and in your policies matches what customers are offered at checkout for their basket, postcode and order time. It covers what agentic commerce changes and why it matters to UK retailers now, then gives you seven checks on the delivery information an agent relies on, a three-stage readiness model and a 2027 checklist with suggested owners.

What is agentic commerce and what does it change about the buying journey?

Agentic commerce is shopping where an AI agent acts on the shopper’s behalf: the shopper describes what they want in plain language, and the agent returns a shortlist, answers follow-up questions and, on some platforms, completes the order. Chat-based shopping in ChatGPT, Google’s AI Mode and Gemini and a retailer’s own AI shopping assistant all sit on this spectrum.

The change for retailers is where the decision happens: an AI shopping assistant narrows the options before anyone lands on a site, so the information a retailer publishes and how precise it is does more of the selling.

It helps to be precise about what an agent can do. Some AI assistants read retailers’ product and delivery policy pages. Others can also complete checkout on the shopper’s behalf. Those existing web pages still matter, while structured data and supported integrations can provide more precise, up-to-date information.

Why it matters for UK retailers now

Agent-led discovery already shapes which products are recommended to UK shoppers, even though agentic checkout is still rolling out. Retail Economics’ Future of Search and Discovery report, based on a survey of 6,000 consumers across the UK, USA and France in November 2025, found that 73% had used an AI assistant in the previous twelve months and 38% had used AI shopping tools for product research, comparisons and recommendations.

Checkout is arriving at different speeds and in different forms. Google’s agentic checkout inside AI Mode and the Gemini app is live for eligible merchants in the USA; in May 2026, Google said it would extend it to Canada and Australia in the coming months, with the UK to follow. In March 2026, OpenAI said ChatGPT shopping would focus on helping shoppers find and compare products, while letting retailers handle checkout through their own websites or apps within ChatGPT.

UK retailers are not waiting: JD Sports announced in January 2026 that customers would be able to buy through ChatGPT, Microsoft Copilot and Google Gemini, starting in the USA. Even retailers that do not yet support purchases through AI assistants should review their delivery information, because agents may already use it when recommending where shoppers should buy.

Why product data alone is no longer enough for an agent to recommend you

Product feeds answer the first questions an agent asks: what the item is, what it costs and whether it is in stock. But they rarely answer the next one: whether this retailer can get it to this shopper, at this address, in time and at an acceptable price.

Consider a shopper with a £60 basket who needs an item by Friday. The delivery page advertises next-day delivery, but the shopper orders on Thursday evening to a postcode with a longer transit time:

Question What the delivery page says What applies to this order
Can it arrive by Friday? Next-day delivery available No: the 2 pm cut-off has passed, so the earliest despatch is Friday
Which service applies? Next day from £5.95 Next day is not offered to this postcode; two-day tracked is
When will it arrive? Order before 2 pm for next-day delivery Tuesday: Friday despatch plus two working days in transit, with no weekend deliveries or bank holidays
What will it cost? Free delivery over £50 Standard charge waived on the £60 basket; the £3.95 remote-area surcharge remains, so the total delivery charge is £3.95

Therefore, retailers need to provide more than product details and a general delivery policy. They should make clear which delivery services are available, what they cost and when the order is expected to arrive, based on the shopper’s basket, postcode and order time. This helps shoppers and AI shopping agents assess whether the retailer can meet their needs.

On retailers’ own sites, delivery already decides a large share of purchases: the UK Delivery Benchmark 2026, produced by Retail Economics with GFS, found that three in five shoppers had abandoned a purchase or switched retailer because the delivery options did not match their needs. It also found that 61% of high-value shoppers put delivery choices ahead of delivery cost. An agent comparing a retailer’s eCommerce delivery options on a shopper’s behalf is likely to consider similar criteria before the shopper visits the retailer’s website.

What delivery information do AI shopping agents need

The six delivery inputs below, plus a seventh check that applies to all of them, are a practical starting point for assessing whether your delivery information is accessible, accurate and usable by AI shopping agents. The table turns each one into a check a retailer can run this week with its own website and a few test baskets:

Delivery input What the retailer should establish A practical test
Cost The charge, free delivery threshold and surcharges for the actual basket and destination Compare the advertised charge with the checkout result for several basket values and destinations, including a remote one
Speed A date or range per service, calculated from despatch time rather than a generic two to five working days Place the same basket before and after the despatch cut-off and check that the quoted date moves
Postcode eligibility Which services apply to each destination, including the Highlands and Islands, Northern Ireland and overseas Test representative postcodes and confirm the options offered match the carrier services that serve them
Cut-offs The latest order time for same-day despatch and how it changes at weekends, on bank holidays and during Peak Check that the cut-off on the delivery page, the product page and the checkout is the one the warehouse works to
Out-of-home availability Whether the parcel can use the collection point offered and which locations are near the shopper Test a standard parcel and an item that may exceed a locker size limit; confirm the locations offered are near the test postcode
Returns The return window, method, cost and any product exclusions Compare the product page, the returns policy and the live returns flow for the same item and look for disagreement
Freshness When each answer was produced and what should trigger a recheck Change the basket, destination or available stock and confirm the delivery answer updates

Missing or inconsistent information in any of these areas makes it harder for an agent to establish whether a retailer meets the shopper’s requirements. These seven checks are a practical starting point rather than a rule every platform applies in the same way. Outdated information can undermine the other checks, so check freshness first.

Out-of-home availability is the easiest to leave out because it depends on where the shopper is. GFS’ out-of-home delivery options cover 75,000+ Click and Collect, PUDO and locker locations in the UK and internationally; an agent can only suggest them if it can see which are near the shopper and whether the parcel is eligible.

Why static delivery pages and free-text policies are not enough for real-time recommendations

A clear delivery policy page explains the retailer’s general offer: a table of services, a line saying orders placed before 2 pm are usually despatched the same day and a paragraph of exceptions. However, for an agent, the question is whether that information is accessible, precise and up-to-date enough to answer for a specific postcode, basket and deadline. “Usually”, for an AI shopping agent, does not provide enough information for a shopper’s unique request.

It helps to think of readiness in three stages, because each one asks something different of the retailer:

  • Discovery: can the relevant delivery and returns information be found and understood? This is the policy page, the product page and the product feed, with delivery and returns details stated plainly and marked up in supported structured data, so an agent has machine-readable delivery information to work from.
  • Validation: can the retailer establish the applicable cost, eligibility and estimated date for the actual basket and destination? This is the job of the checkout’s delivery options engine, and it is where most of the audit table is answered.
  • Execution: can a supported integration carry the chosen option into an order, with the necessary checks and authorisation? This depends on the platform as much as the retailer; an API is only part of it, because Google’s checkout programme, for example, requires configured product, shipping and returns information, a supported integration and its approval.

Accurate delivery options are often calculated at checkout after the shopper enters a postcode. These options may be displayed using JavaScript. Retail Economics reports that many AI crawler bots cannot read JavaScript-rendered content (Retail Economics, 2026), so some crawlers may be unable to access those delivery details. An integration also needs accurate underlying data: connecting a system that provides incorrect delivery information does not make that information reliable.

How delivery options should connect to stock, warehouse and carrier availability

A delivery option is a promise built from three facts: the item is in stock, it is at a location that can despatch it in time and a carrier service will collect it before cut-off and deliver it to that postcode. If any of those facts is stale, the option is wrong, however well it is formatted. Publishing a delivery answer is not the same as proving the operation can fulfil it.

The options a retailer exposes should therefore be generated from live data rather than typed into a page. The data should include stock and despatch location from the warehouse or order management system, service availability and transit times from the carriers and cut-offs from the collection schedule. An existing delivery options engine at checkout is a starting point, but it should use up-to-date stock, carrier-service and cut-off data to confirm the available delivery options before the shopper places the order.

Orders that split across locations need particular care: show the expected arrival for each shipment and use the latest arrival only when describing when the complete order will have arrived. Peak can be a test: when a carrier caps volumes in December, a next-day option still advertised on a static page becomes a promise the retailer cannot keep.

What happens when an agent promises a service the retailer cannot fulfil

The shopper is unlikely to blame the agent. The order was placed with the retailer, so it must handle complaints about missed delivery dates, refund requests, “where is my order?” (WISMO) enquiries and customer reviews.

The second cost is easier to miss. The platforms behind these agents already use the quality of a merchant’s product, delivery and returns data when deciding what to show, although none has set out how, or whether, missed delivery promises feed back into recommendations. The safer assumption is that accurate promises help and missed ones do not, which is why the fallback rule in the checklist matters: an honest “not confirmed” beats a confident wrong answer.

A 2027 agentic commerce delivery-readiness checklist

Use this checklist to prepare your delivery data for agentic commerce now, as AI-assisted shopping and agentic checkout develop. Each item carries a suggested owner to confirm in your own organisation, because delivery promises go wrong when nobody owns the rule behind them:

  • Structured product data: Delivery cost, delivery time and returns fields completed in the product feed and the structured data on each product page (owner: eCommerce)
  • Live delivery options: A service that returns the real options, prices and dates for a basket and postcode through an API, not only in the browser (owner: Technology)
  • Postcode rules: Surcharges, exclusions and longer transit times for remote and international addresses held as data, not as footnotes (owner: Operations)
  • Cut-off calendar: Same-day despatch cut-offs by service, with weekend, bank holiday and Peak changes published in advance (owner: Operations)
  • Stock and location: Delivery promises generated from live stock at the despatching location, with out-of-stock items removed (owner: Operations)
  • Carrier availability: Service availability, capacity limits and transit times fed from the carriers so promises change when the network does (owner: Operations)
  • Out-of-home locations: Parcel shop, locker and store collection points exposed with opening hours, size limits and distance for the shopper’s postcode (owner: eCommerce)
  • Returns data: Return methods, drop-off locations, cost and window stated in the same machine-readable form as delivery (owner: Customer Service)
  • Fallback rule: Where current eligibility or timing cannot be confirmed, state the uncertainty or hand the shopper to checkout rather than presenting an estimate as a guarantee (owner: eCommerce)
  • Promise monitoring: The promise made in each channel compared with the delivered date, with misses reviewed monthly by channel and carrier (owner: Customer Service)

Assign an owner and a review date to each gap you find. A correct API response is not enough if nobody owns the service rules behind it. 

How structured checkout and carrier data make a retailer easier to recommend

In agentic commerce, the retailers an agent can recommend with confidence are the ones whose delivery answers are structured, current and consistent, including in the AI shopping assistant a customer is talking to. That is a data problem before it is a marketing one, and it starts with a single source of machine-readable delivery information, which means getting the validation stage right.

GFS is a managed Multi-Carrier delivery partner, bringing together carrier services, delivery expertise and pro-active customer support through one relationship. Retailers gain access to 1,000+ delivery services and 75,000+ collection points, supported by competitive carrier rates, capacity planning and contingency arrangements. GFS helps manage delivery from checkout through to returns, with a dedicated Customer Care team tracking parcels and resolving issues before they affect customers. Its supporting technology enables retailers to offer relevant delivery choices at checkout and adapt services as demand and carrier capacity change.

Behind the data, GFS’ managed Multi-Carrier delivery services give retailers access to those services through a single integration, backed by a pro-active customer care team that works to resolve delivery issues before they reach the customer. Connecting those delivery options to an AI shopping platform is a separate integration task that depends on the retailer’s systems and the platform’s supported capabilities. GFS can support retailers in getting the delivery data behind it right.