Agentic commerce: delivery and returns as discovery layer

What happens when an AI agent goes shopping for your shopper? It might be ChatGPT comparing three retailers for a winter jacket, or Perplexity finding the fastest delivery for a last-minute gift. Protocols such as the Universal Commerce Protocol (UCP) and OpenAI's Agentic Commerce Protocol (ACP) now give agents a standard way to check out with retailers.

Most shoppers still buy for themselves: 38% of European shoppers rely on AI to research purchases, while 1–3% have bought through an AI platform, according to McKinsey and EuroCommerce. As agents take on more of that work, they'll judge retailers on what they can read. Brand storytelling carries less weight with an agent than structured signals do, and delivery and returns are among the clearest of those signals.

From human browsing to machine evaluation

Humans shop with emotion, brand loyalty, and visual appeal. They linger on product pages, read reviews, and can be swayed by photography and copy. AI agents shop differently. They operate on API-first delivery infrastructure, structured data and performance signals. They compare options across multiple retailers simultaneously, and they optimize for the criteria their user has set: fastest delivery, lowest total cost, best return policy, or highest reliability.

This means that the signals AI agents evaluate are fundamentally different from what traditional e-commerce optimization targets. Beautifully designed UX is invisible to an agent parsing an API response. What matters instead is the clarity, accuracy, and machine-readability of the data behind it.

What AI agents actually look for 

When an AI agent evaluates a retailer, it examines a set of operational signals that many retailers still overlook as competitive differentiators even in traditional e-commerce shopping. These include delivery promise accuracy and specificity — not vague ranges like ‘3-5 business days’ but precise dates like ‘arrives Thursday’ — and delivery reliability based on historical performance data. 

They also include pricing transparency, where the total cost including delivery is available upfront with no surprises at checkout. Add to that return policy clarity, where structured, API-accessible terms beat policies buried in FAQ pages or PDFs. It’s exactly the same agentic shopping principle as with the product data, in which ‘100% organic cotton, 200 GSM’ wins over ‘delightfully soft to the touch’.

Delivery as your new discovery layer

If agentic commerce changes what gets evaluated, delivery is where the consequences are most concrete. When AI agents compare retailers in real time, your delivery capabilities become a ranking factor before the sale.

Delivery promise as a machine-readable trust signal

Consider how an AI agent handles a simple shopping request: “Find me a pair of running shoes under €120, delivered by Friday.” The agent queries multiple retailers simultaneously. Retailer A responds with “arrives Thursday by 6pm” — specific, calculated from real carrier performance data, dynamically adjusted for this particular order. Retailer B responds with “3-5 business days” — a static estimate set on a quarterly basis that may or may not hold. The agent recommends Retailer A.

That’s the decisive gap between adaptive and static delivery infrastructure. Retailers using AI-predicted delivery windows — calculated from real-time and historical fulfillment data, accounting for carrier behavior, destination and seasonality — produce precise, reliable promises that AI agents can trust and recommend. Retailers still relying on manually configured delivery rules that have to be widened during peak season ‘just to be safe,’ produce the vagueness that agents penalize.

Delivery pricing as a conversion signal

AI agents also evaluate the total cost of the transaction, and they are particularly sensitive to pricing opacity. A retailer that shows ‘free shipping’ but adds a surcharge at checkout would frustrate a human user, but it also sends a negative signal to an AI agent that evaluates pricing consistency and transparency across every step.

That’s where delivery pricing aligned with real delivery economics outperforms flat-rate models. When pricing reflects the actual cost of each delivery — staying competitive where carrier costs are low and adjusting appropriately where they are higher — it creates the consistency and transparency that agents interpret as reliable. Flat pricing, by contrast, creates unpredictable variation that erodes agent trust over repeated interactions.

Carrier performance as a competitive moat

An agent that tracks outcomes across orders builds a reliability profile for each retailer. Orders that arrive when promised strengthen it, and late deliveries weaken it. That makes carrier performance part of how a retailer gets recommended. Retailer A whose orders consistently arrive on time accumulates a compounding trust advantage, while retailer B with inconsistent fulfillment sees its agent recommendations decline over time. It definitely pays to measure what was promised against what actually happened.

Returns are the silent ranking factor

Most discussions of agentic commerce focus on the purchase. Almost none address what happens when something goes wrong. That’s a significant blind spot, because an AI agent optimizing for customer satisfaction must factor in a critical question — what happens if this does not work out?

Do AI agents check return policies?

Think about it from the agent’s perspective. Its job is to make the best possible recommendation for its user. That means considering not just price, speed, and availability, but also the risk of the purchase. How easy is it to return? How long is the return window? How fast is the refund processed? Can the item be exchanged instead? For an AI agent evaluating risk-adjusted value, return policy clarity is a first-order signal.

Return policy clarity, return window policy, refund speed, and exchange flexibility all become machine-evaluable data points. The retailer with structured, API-accessible return terms gives the agent exactly what it needs to make a confident recommendation. The retailer whose return policy lives in a PDF or a buried FAQ page is effectively invisible in this dimension.

Operational gap most retailers haven’t closed

Online return policies should be agent-readable. At this point, they exist as human-language text on a webpage, full of conditions, exceptions, and qualifications that require interpretation. An AI agent cannot reliably parse a paragraph of legal text to determine whether a specific product is returnable, what the window is, or who pays for return shipping. Again, AI agents compare based on data, not design, so structured data is the new foundation.

Simpler wording helps people read a policy. Agents need return terms as structured data they can query directly: whether an item is returnable, how long the window is and who pays for return shipping.

Connect returns to checkout intelligence

The most forward-thinking approach goes further: it connects return outcomes back to checkout decisions. When return data feeds into the same intelligence layer that shapes delivery promises and pricing, retailers can dynamically adjust return terms based on product category, customer behavior, and historical return patterns. This creates a closed loop where operational data from every completed transaction — whether it ends in a kept purchase or a return — strengthens the intelligence that powers the next one.

In many setups, checkout and returns run on separate tools, so what happens after the purchase never reaches the next checkout decision. Retailers who connect the two give agents a more complete picture of what buying from them is like. The retailers who unify these systems are building the operational excellence that AI agents will increasingly use to differentiate recommendations.

Checklist: are your delivery and returns ready for AI agents?

An agent reads delivery in a set order, and each step can end the comparison before the next one starts. The questions below follow that order, from whether an agent can reach your delivery information at all to what happens after the parcel ships. Most European retailers would answer "not yet" to several of them, because most still run delivery on static rules.

1. Can an agent reach your delivery information?

• Is your delivery offer available outside the checkout page? Options, prices and dates come back as values an agent can query through an API, rather than as text that only appears once a page loads.
• Does each product page show delivery information in a machine-readable form? The delivery promise sits next to the price in structured data. In Adobe's analysis of US retail sites, product pages were only 66% machine-readable to AI, the lowest score of any page type.

2. Can you deliver to this shopper, and how?

• Can an agent check whether you deliver to a specific address before it adds you to a shortlist? Availability is answered per address or postcode, rather than per country.
• Is each delivery method listed as its own option with a clear name? Home delivery, pickup point, parcel locker and in-store pickup are named the same way wherever they appear.
• Is the order of your options a deliberate choice? Options are ranked for this order. The default carries weight with shoppers: 68% of orders go out on the option the retailer pre-selected, according to Ingrid data.

3. Is the delivery promise specific, and does it hold?

• Do you give a date for this order, or a range such as three to five business days? A date calculated from warehouse cut-off times and carrier performance to that address.
• Do you measure how often the promised date holds? The on-time rate is tracked per carrier and region and informs the next promise. According to Ingrid data, 55% of shoppers name inaccurate delivery dates as their single biggest frustration.
• Do you widen delivery windows in peak season "just to be safe"? Peak-season promises are based on how carriers actually performed in past peaks, so any buffer reflects evidence.

4. Is the total delivery cost clear before checkout?

• Is the price of each delivery option visible before checkout, including surcharges? The price an agent reads while comparing is the price the shopper actually pays. In Ingrid's survey of UK shoppers, 82% say unexpected delivery costs at checkout make them reconsider a purchase.
• Are free delivery thresholds stated as conditions an agent can check? A rule such as "free delivery on orders over €50" is available as data that can be checked against the cart, as well as in a banner.

5. Are your return terms structured?

• Can an agent tell whether a specific item can be returned? Returnability is set per product or category, including exceptions such as final-sale or hygiene items.
• Are the return window and return cost available as data? The number of days, any return fee and who pays for return shipping are each a value, not a sentence in a policy page or PDF.
• Do you state how fast refunds are processed and whether exchanges are possible? Refund timing and exchange options are stated per return method, so an agent can weigh the risk of the purchase.

6. Does every touchpoint give the same answer after the order?

• Can an assistant find out where an order is and when it will arrive now? Order status and any updated date come from the same source as the original promise.
• Is the date an agent reads the same date the shopper sees at checkout and on the tracking page? One source of delivery data sits behind every surface, so the promise, the checkout and the tracking page agree.

Infrastructure shift to delivery intelligence

Delivery promises, pricing and returns all point to one question: does your delivery setup adapt to each order or does it run on static rules set once a season? The answer shapes what an agent sees when it compares you with other retailers.

See where your delivery setup stands on the path to agentic-ready: Delivery in Agentic Commerce: 2027-2030 →

Retailers who win in agentic shopping

Delivery has always mattered to shoppers. Agentic commerce makes it visible earlier, before anyone reaches the checkout. Retailers who make their delivery promises, prices and return terms readable to agents give themselves the best chance of being recommended. Ingrid Agentic DeliveryOS is built for that shift, bringing shopper, retailer and carrier intelligence into one operating system.

‍Learn more about Ingrid Agentic DeliveryOS for retail →