Who Owns Agentic Commerce Inside a Retail Team
Six roles cover agentic commerce in a retail team. We map each to OpenAI and Google merchant requirements, with ownership matrices for small and large teams.
On this page
The internal team structure for agentic commerce comes down to six roles required by the work itself, and three of them only switch on once you build a cart or checkout integration. Every retailer needs a channel owner, a feed and catalog owner, and a policy owner. A checkout engineer joins for a cart or a checkout integration, since Google's cart path also runs on an endpoint you build. A payments lead and an order operations owner join when checkout happens inside the assistant. We found no prescribed org chart or job description in the OpenAI, Google or Shopify merchant pages we read on October 7, 2026. Those pages hold a task list, much of it written as "you must" sentences, and each task lands on a function a retailer already has.
What do the programs ask a retailer to do?
For retailers, the team requirements of an agentic commerce channel implementation come from two sets of merchant documentation: OpenAI's product feed program for ChatGPT, and Google's Universal Commerce Protocol (UCP) guide for AI Mode in Search and Gemini. Both are gated. OpenAI says "Onboarding product feeds in ChatGPT is currently available to approved partners." Google asks merchants to join a waitlist and adds: "Your integration must be approved by Google before you can go live on AI Mode in Google Search and Gemini."
The two programs ask for different amounts of work. On its merchant page, OpenAI says it is "moving away from a standalone Instant Checkout experience in ChatGPT" and that "purchases are completed on merchant-owned websites or apps." The developer site still publishes an Agentic Checkout production checklist from OpenAI that calls Instant Checkout "currently available to approved partners." Where the two pages disagree, we plan staffing around the merchant page. Google's guide offers both routes: a native checkout path for purchases on Google's surfaces, or a cart path that hands the cart to your website. Read as a work breakdown, the documentation yields eight recurring tasks.
| Task | What the documentation says | Function that carries it |
|---|---|---|
| Keep the feed current | OpenAI: "It is generally recommended to provide the entire feed once a day via file upload, and then send updates throughout the day via the API." | Catalog or merchandising, plus a delivery pipeline |
| Flag checkout-eligible products | Google: the native_commerce(checkout_eligibility) attribute defaults to FALSE, and the ineligible list includes subscriptions, personalized goods, pre-orders and final-sale items | Merchandising |
| Publish returns and support details | Google: "You must include return cost, return speed (window), and a link to the full policy," plus at least one customer service contact method | Policy owner with support |
| Attach legal notices | Google: the consumer_notice attribute is "mandatory for items with regulatory warning requirements" such as California Proposition 65 | Legal or compliance |
| Build the checkout or cart integration | Google: a native checkout path or a cart path, approved by Google before launch | Engineering |
| Accept the payment token | Google: "your Payment Service Provider (PSP) must be able to process our tokens" | Payments |
| Check stock and addresses live | Google: "As the Merchant of Record, it is your responsibility to perform real-time inventory checks." | Engineering with merchandising |
| Push order status and adjustments | Google: order created, shipped and delivered are mandatory events, and refunds, returns and cancellations go in as adjustments | Order operations with engineering |
The feed row comes from OpenAI's get-started guide and the integration row from Google's overview, both linked above. The eligibility, returns and notice rows come from Google's Merchant Center preparation page, the payment and stock rows from Google's UCP FAQ, and the order row from its order lifecycle guide, linked further down.
Merchant of record is the phrase that turns this list into a staffing question. Google's FAQ tells the retailer, "You are the Merchant of Record," and assigns stock checks and shipping address validation to the retailer on that basis. The assistant runs the conversation, and everything a customer can complain about afterward stays with you.
Which skills does each role need?
The agentic commerce skillset is spread across those six roles on the team, and most of it is familiar retail skill held to a stricter standard, since the reader of your data is software that acts on what it finds.
- Channel owner: usually the head of ecommerce. The skill is marketplace-style judgment: which programs to apply for, which parts of the catalog to opt in, and what return justifies the work.
- Feed and catalog owner: structured-data discipline, meaning stable identifiers, clean variants, scheduled exports and field-level checks. OpenAI's instruction to "Validate required fields for every record" describes a process habit more than a tool.
- Policy owner: knows which SKUs carry regulatory notices and what the return window is by category, and keeps those statements identical on the site, in the feed and in Merchant Center. A mismatch between the three is the kind of contradiction we cover in LLMs Cite the Brand That Agrees With Itself.
- Checkout engineer: API craft. Google's FAQ expects an
out_of_stockerror code when an item sells out mid-checkout, recoverable errors marked as recoverable, and endpoints that answer its health probes. - Payments lead: PSP token support, card configuration and reconciliation of orders that never touched your own checkout page. Per Google, you "must explicitly provide your Google Pay configuration within your UCP profile."
- Order operations owner: maps warehouse and carrier statuses to the events a checkout program expects, and owns refunds, disputes and contact from buyers who never saw the website.
We found no documented task that asks a retailer to operate a model or write prompts, which is why we would staff this channel from ecommerce operations before data science.
A responsibility matrix for a small team
The matrix below is a hypothetical example for an illustrative five-person team on a hosted commerce platform. A marks the one accountable name, R the person doing the work, and C someone consulted. The letters are our recommendation.
A platform absorbs most of the checkout roles for a team like this. Shopify says "Agentic storefronts is active by default for eligible stores," that ChatGPT users "complete their purchase on your online store checkout," and that orders from AI channels appear in the admin "with channel or referrer attribution." OpenAI's merchant page adds that Shopify and Etsy catalogs are already integrated, with no application required. What remains is data, policy, service and measurement.
| Task | Head of ecommerce | Merchandiser | Developer or agency | Support lead | Finance |
|---|---|---|---|---|---|
| Channel settings and program applications | A, R | C | C | - | - |
| Product data quality and eligibility | A | R | C | - | - |
| Store policies and legal notices | A | R | - | C | - |
| Checkout and payment settings | A | - | R | - | C |
| Order status, refunds and buyer contact | C | - | - | A, R | C |
| Channel reporting | A, R | C | - | - | C |
The head of ecommerce is accountable on five of six rows. That is honest for a team this size, and it is also the exposure: when that person is away, nobody decides. If budget is the constraint, our line-by-line budget for an AI visibility program works through the build versus buy break-even.
How the matrix changes in a large team
Across larger teams, the same agentic commerce skillsets and roles sit in separate departments, so the matrix gets wider. This second example is also hypothetical: a retailer with its own commerce stack, integrating directly with Google's native checkout. The letter I marks a team that is kept informed.
| Task | Ecommerce | Merchandising | Engineering | Payments | Legal | Support and fulfillment |
|---|---|---|---|---|---|---|
| Program approval and channel P&L | A, R | C | C | C | C | I |
| Feed content and eligibility flags | C | A, R | C | - | C | - |
| Feed pipeline and delivery | C | C | A, R | - | - | - |
| Checkout endpoints and approval testing | C | - | A, R | C | - | C |
| Token acceptance and reconciliation | I | - | R | A | C | - |
| Return policy, terms and consumer notices | C | R | - | - | A | C |
| Order events, refunds and disputes | I | - | R | C | - | A |
| Channel reporting | A | C | R | C | - | I |
Two things change at scale. Engineering becomes accountable for two rows outright, and the order row separates the doing from the answering: engineering sends the webhook while support and fulfillment answer for what the buyer sees. Google's order lifecycle guide, last updated October 2, 2026, is strict about it: "After checkout is completed and an order is placed, you must push status updates to Google using the order webhook." Three events are mandatory (order created, order shipped with a tracking number and tracking URL, and order delivered), and "A sequential progression from shipped to delivered is mandatory." A late carrier scan can become a stale status on the buyer's My orders page.
Larger merchants also face a ChatGPT decision the small team can skip. OpenAI's merchant page calls apps optional and "best suited for larger merchants who want more control over the shopping and purchase experience," which adds an engineering row only when ecommerce asks for one.
Where ownership gets contested
We expect three seams to produce most of the arguments. The eligibility flag looks like a merchandising lever, but Google's ineligible list makes it a compliance decision as well, so legal should set the rule and merchandising should apply it SKU by SKU. Inventory is the second seam. Because the retailer runs the real-time stock check during checkout, a number merchandising owns in the catalog becomes an API contract that engineering gets paged for, so the two teams need one source of truth before approval testing starts.
Measurement is the third seam and the easiest one to leave unassigned. Orders are countable, and what sits upstream of them is whether the assistant recommends your products at all. That question belongs to the channel owner alongside revenue. Track it as AI share of voice, and run a small prompt panel with its known failure modes in mind. Our AI Visibility Checker is one further resource for that side of the work, and our AI search optimization service is another for teams that want outside help.
None of this calls for a new department. The documentation hands a retailer a list of obligations, and the structure that works is one accountable name per obligation, written down before the application goes in. If a row has to stay unassigned for a while, keep it away from order updates, because that row is the one the customer watches in real time.
Sources
- OpenAI: Power product discovery in ChatGPT (merchant page and FAQ)
- OpenAI: Agentic Commerce, Get Started
- OpenAI: Agentic commerce in production
- Google: Universal Commerce Protocol Guide, Overview (last updated October 5, 2026)
- Google: Prepare your Merchant Center account (last updated October 2, 2026)
- Google: Order lifecycle (post-purchase) (last updated October 2, 2026)
- Google: UCP Frequently Asked Questions (last updated October 2, 2026)
- Shopify: Shopify agentic storefronts