What agentic commerce looks like in practice: Insights from five leading retailers

Many brands are already tracking referral traffic from AI interfaces and watching it grow month on month. Absolute volumes remain small, below 1% in most cases, but agentic traffic is converting at up to six times the rate of traditional organic search.

The opportunity is clear, but the question is how retailers can act in practical terms.

All five retailers we spoke to have completed some of the groundwork. They've assessed product feeds, reviewed APIs, and aligned teams. When it comes to the actual build, their approach can be split into the following three categories:

1. Building on their own terms Retailer A (European-founded fashion brand) runs a proprietary ecommerce platform, giving them flexibility. For example, their CTO estimates they could adapt their product feed to a new spec like ACP within a day of development. They've already built internal AI agents for translations and product descriptions, and developed an MCP to expose order and tracking data to customer service agents.

2. Waiting on their commerce platforms Retailers D and E (a large US apparel retailer and a global footwear brand) are working within their commerce platform ecosystems, waiting to see what these platforms build before committing significant engineering resources of their own.

3. Building in-house, watching the market Retailers B and C (a large US sports and outdoor retailer and a global luxury fashion brand) sit in the middle, building in-house while monitoring what partners can offer. In most cases, that’s as far as they can go. Agentic commerce platforms are still in closed beta and timelines for broader access remain unclear.

One of the most common questions we hear when talking to retailers is: How do I make sure my products show up on ChatGPT or Gemini? The answer has two parts: What you can control, and what’s still evolving.

The part you can control starts with the feed. AI platforms need access to accurate, structured product data to surface your products.

Most brands already send feeds to Google, Meta, and other partners, but agentic commerce requires more. The two main protocols emerging, ACP (the Agentic Commerce Protocol, associated with OpenAI) and UCP (the Universal Commerce Protocol, associated with Google), each require fields that existing feeds weren't built to include. They want richer product descriptions, additional metadata, and structured data that machines can query reliably.

To complicate things further, both protocols are still evolving. What counts as “agent-ready” today may look different six months from now. So it’s important to prioritize flexibility across specs rather than optimizing for any single one.

Retailer E (the global footwear brand) maps their agentic strategy across three layers:

1. Their own site: Discoverability and shopper agents.

2. Third-party AI interfaces and social platforms: Being purchasable wherever shoppers are.

3. Internal operations: Agents that support merchandising, development, and quality assurance.

The part nobody fully understands yet is what happens inside the algorithm once your feed is in place. As Retailer D (the large US apparel retailer) put it:

"How do we know our products are being prioritized when someone asks an agent for a specific category?”

Sending your product feed to the AI surfaces is the prerequisite. What the algorithm does with it after that is still emerging. We're working directly with OpenAI, Google, and other platform partners to understand how products get surfaced and ranked, and we'll be sharing what we learn as the picture becomes clearer.

Article

Agentic commerce and product feeds: a guide for retailers

April 20th, 2026 · 3 minutes

The infrastructure underpinning agentic commerce is taking shape. Issuers and card networks are actively working through how agentic transactions fit within existing frameworks, and the broader payments ecosystem is catching up in real time. This dictates what's actually possible for brands right now, regardless of how prepared you are. In the meantime, several constraints stand out:

The checkout is no longer a linear journey

Traditional checkouts were built to handle a single, decisive moment: A customer clicks ‘buy’, the order is placed, and the transaction is processed. Agentic checkout works differently. Agents query inventory, pricing, tax, and availability throughout a session, recalculating in real time as the shopper's request evolves. Systems designed for a single checkout event weren't built to handle that kind of continuous, stateful querying without performance trade-offs.

Protocols only support one item per transaction

Current protocols support only one item per transaction, which goes directly against how many brands operate and how customers shop. For example, Retailer D's customers typically like to buy several items at once: a few t-shirts, a pair of jeans, maybe a pair of sneakers. Meanwhile, Retailer C (the global luxury fashion brand) operates an advanced returns model. They send multiple items to a customer, only charging for what they choose to keep. This points to a significant gap between where the technology is and how commerce actually operates, meaning that for now, brands are having to design around it.

Retailers need to decide who owns each layer

Architecture decisions add another layer of complexity. Retailer C, for example, sends their product feed through a separate aggregation tool rather than directly from their commerce platform. They do this because pulling large amounts of product data directly from the platform slows it down, which affects the shopping experience on their own site. The aggregation tool takes on that load instead, acting as a buffer. Longer term, they plan to build a direct API connection into their core product catalog.

Meanwhile, Retailer B is evaluating whether their commerce platform's MCP layer buys them speed, or whether building in-house retains more control long-term.

The question of who owns the customer relationship isn't new, but the stakes are now higher. When social commerce emerged, brands faced the same concern: what happens when the interface belongs to someone else? Most found a way to participate without losing meaningful ground. With agentic commerce, the stakes are higher. Lose visibility on a social platform and you lose a sale. Lose the customer relationship to an AI interface and you may not get it back. As Retailer A's CTO put it:

"More than a technical failure, we're worried we'll become a back-end commodity, piping product information into an interface the shopper never leaves. Meanwhile, the relationship we've built over years quietly disappears."

At the heart of this fear lies a practical problem: If a shopper discovers your brand through an AI interface, how do they log in? How does that interface recognize them as an existing customer, with their order history, their loyalty points, their saved preferences?

Maintaining a consistent relationship with your shopper across a third-party interface is one of the hardest problems to solve.

Retailer E's response is to be present everywhere a shopper might encounter the brand. By building an agentic strategy across their own site, third-party AI interfaces, and internal operations simultaneously, no part of the experience defaults to someone else.

Agentic commerce is still in its infancy. OpenAI is very transparent about the fact they are flying the plane and building it at the same time. So, for now, the retailers we speak to are focusing on building the foundations and gathering learnings. Here’s what that looks like in practice:

Retailer C is opening their site to LLM crawling, using a bot governance platform to control which agents can access what and under what conditions.

Retailer A is using their US operations as a regulatory testing ground. EU regulation makes autonomous AI payments significantly harder in Europe, so brands with a US presence are treating it as the primary proving ground to see what works.

Several brands are exploring the app flow while waiting for instant checkout as a way to accumulate learning before closed betas open up.

All five recognize that transaction volumes at this stage will be negligible in the near term. The goal is to be well-positioned when that changes.

Platform access, protocol maturity, and the underlying ecosystems are still assembling. Locking in decisions too early carries real risk. That’s why the brands that are setting themselves up for long-term success are moving deliberately, learning, testing, and staying adaptable.

From our perspective, agentic commerce is simply a new channel, not a new owner of the customer relationship.

We’re focused on building an infrastructure that keeps retailers in control regardless of which AI platform a shopper comes through. That means building integrations into ACP and UCP, and what comes next, so you connect once and reach multiple protocols.

We are also collaborating with industry partners to ensure retailers’ needs are represented as the foundations are laid. For example, we are a member of the Agentic AI Foundation, we’re collaborating on Google’s Agent Payments Protocol, and we’re a founding member of the x402 Foundation.

Want to stay ahead in the age of agentic commerce? Find our latest agentic updates and practical insights here >

Curious about what agentic commerce means for your business specifically?

Get in touch