Agentic commerce means AI agents that search, compare and buy on a shopper’s behalf, and UCP, the Universal Commerce Protocol, is an open standard Google announced so agents and merchants can complete those purchases in a common way. For a store, the work is a set of workflows: clean product data, an agent-ready checkout, clear policies and new measurement. Each can be prepared step by step before volume arrives.
Several efforts are shaping this space at the time of writing. Google introduced UCP with retail and payments partners for checkout through its AI surfaces. OpenAI and Stripe published the Agentic Commerce Protocol for purchases inside ChatGPT. Card networks have announced agent payment programmes, such as Visa Intelligent Commerce and Mastercard Agent Pay. Specifications and partner lists are still moving, so treat details as provisional and follow official documentation.
What changes when an agent is the shopper?
A human browses pages, reads reviews and clicks through checkout. An agent reads structured data, compares facts and calls an interface to place the order. Visual design matters less at the moment of decision; accuracy and completeness of data matter more.
- Product discovery happens inside an assistant, not on your category page
- Comparison relies on attributes such as size, material, price, delivery time and returns
- Checkout may run through a protocol, with payment tokens instead of a typed card number
- The customer still owns the order, so service, returns and email follow-up stay with you
Workflow 1: How do you prepare product data for agents?
Start here, because every protocol depends on it. Agents cannot recommend what they cannot read.
- Step 1: Audit your product feed in Google Merchant Center for errors and missing attributes
- Step 2: Fill core attributes for every product: GTIN, brand, size, colour, material, condition
- Step 3: Write titles and descriptions that state facts plainly, without marketing filler
- Step 4: Add Product structured data on each product page, matching the feed exactly
- Step 5: Publish shipping times, costs and return windows as data, not only in a policy page
- Step 6: Set a schedule so price and stock changes reach the feed the same day
Many stores already do part of this for Google Shopping. Agentic commerce raises the bar because a mismatch between feed and page can cause a failed or disputed order, not just a disapproved ad.
Workflow 2: How do you make checkout agent-ready?
The checkout workflow depends on your platform. Shopify, Salesforce Commerce Cloud, commercetools and other platforms have announced or are building support for agent protocols, sometimes through payment partners such as Stripe, Adyen or PayPal. Check your platform’s current documentation before custom work.
- Step 1: Confirm which protocols your platform and payment provider support
- Step 2: Enable the integration in a test environment and place test orders through it
- Step 3: Map discount codes, shipping options and taxes so totals match the normal checkout
- Step 4: Decide which products are excluded, such as age-restricted or made-to-order items
- Step 5: Set fraud rules suited to agent orders, working with your payment provider
- Step 6: Launch on a limited catalogue and watch orders daily
Workflow 3: Orders, service and returns
An order placed by an agent is still your order. Confirmation emails, tracking updates and returns must work exactly as for any other purchase. The difference is that the customer may not remember your store’s name, because they never visited it.
- Tag agent orders in your order system by channel or protocol
- Make confirmation emails clearly state your store name and contact options
- Train support staff to recognise agent orders and their typical questions
- Track return rates for agent orders separately, since a mismatch in data shows up here first
How do you measure agentic commerce?
Classic analytics sees little of the journey. There is no session on your site before the purchase, so GA4 may only record the order if your platform sends it. Plan measurement from the order system outward.
- Orders and revenue by agent channel, from platform order tags
- Average order value and return rate compared with web orders
- Feed health: error counts and attribute completeness in Merchant Center
- Visibility: how often your products appear when you test shopping prompts in assistants
- Customer lifetime value of agent-acquired buyers over the following months
How Alien Road applies agentic commerce and UCP
In a typical ecommerce engagement we start with a data readiness audit: feed errors, attribute coverage, structured data accuracy and how often price or stock data goes stale. Most stores have gaps here, and closing them improves Google Shopping performance even before any agent traffic appears.
Next we review the client’s platform and payment provider for protocol support, enable integrations in a test store, and run test orders that cover discounts, shipping zones and returns. After a limited launch, we report weekly on agent orders, return rates and feed health, and run monthly prompt tests in the main assistants to see whether products appear. For example, if test prompts surface competitors with clearer delivery data, the fix is in the feed, not in ads.
What risks should a store plan for?
The main risks are wrong orders from bad data, fraud patterns that differ from human shopping and a weaker customer relationship. Fewer site visits mean fewer chances to show related products or collect newsletter sign-ups. Plan post-purchase emails carefully, since they may be the first real contact a buyer has with the brand.
Who owns the customer data?
Check what each protocol shares with you and with the agent platform. You need enough data to fulfil the order and handle service, and the shopper must have agreed to that sharing. Ask how marketing consent is captured, if at all, since an agent purchase does not automatically mean the buyer agreed to your newsletter. Under GDPR and KVKK, that consent has to be explicit.
Write these answers into your privacy notice and order flow before launch. Fixing consent after thousands of orders is much harder than designing it in.