Agentic commerce is shopping in which an AI agent, acting for a customer, finds products, compares options and completes checkout, often inside a chat assistant. Google’s Universal Commerce Protocol (UCP) is one of the open standards meant to let merchants expose catalogue, cart and checkout to such agents. Implementation comes down to clean product data, a machine-readable checkout, trustworthy payments and clear policies, tested in stages.
What is agentic commerce, and where does UCP fit?
For years, a shopping assistant could only recommend and send a link. Agentic commerce goes further: the agent can add items to a cart and pay, with the shopper’s approval. Several efforts are shaping how this works. OpenAI and Stripe introduced the Agentic Commerce Protocol for checkout inside ChatGPT. Google announced the Universal Commerce Protocol for agents across its surfaces, with partners from retail and payments. Payment networks such as Visa and Mastercard have announced agent-related payment programmes, and Google has proposed an Agent Payments Protocol.
These standards are young and still changing. The details of each specification, eligibility rules and supported countries should be checked in the official documentation at the time you start. The underlying requirements, however, are stable enough to prepare for now.
Phase 1: readiness checklist
Before connecting to any protocol, make sure your store can be understood by machines:
- Product data is complete: titles, descriptions, GTINs, brand, variants, sizes, colours, materials and images.
- Prices and stock are accurate and updated frequently, ideally in near real time.
- Shipping costs, delivery times and return policies are structured, not buried in paragraphs.
- Your Google Merchant Center feed is free of major errors and disapprovals.
- Structured data (Product, Offer, MerchantReturnPolicy, ShippingDetails) is valid on product pages.
- Your platform, such as Shopify, WooCommerce, BigCommerce or Salesforce Commerce Cloud, has announced or supports agent checkout integrations, or you have developer capacity for an API integration.
An agent that sees a wrong price or a missing size will either make a bad purchase or skip your store. Data quality is the whole game at this stage.
Phase 2: pilot design
A pilot should be small and observable. Pick a limited product range, ideally items with simple variants and low return rates. Decide which surfaces you will test on and which customers or markets the pilot covers.
Pilot checklist
- Choose 50 to 200 products with clean data and steady stock.
- Define success measures: completed agent orders, error rate, returns, support contacts.
- Set up order tagging so agent-originated orders are identifiable in your back end.
- Prepare customer service with a short guide on agent orders.
- Agree a rollback plan: how to switch off agent checkout quickly.
Phase 3: checkout and payments
The checkout is where trust is won or lost. Agents need a predictable sequence: create cart, add items, apply shipping, show the full price including taxes, receive explicit user confirmation, then pay. Each protocol defines its own way of doing this, but all depend on your systems returning accurate totals and clear error messages.
Payments typically flow through tokenised credentials from a payment provider, so your store never sees raw card numbers from the agent. Work with your payment service provider to confirm support, fraud screening and chargeback handling for agent-initiated orders. Strong customer authentication rules in the EU still apply where relevant.
Phase 4: policies and customer protection
Agents act on behalf of people who did not read your site directly. That raises the bar on clarity:
- Return and refund terms must be machine-readable and consistent everywhere.
- Order confirmations should go to the customer directly, with full details.
- Age-restricted or regulated products may need to be excluded.
- Your privacy notice should cover data received from agent platforms.
Phase 5: moving to production
Production readiness means the pilot ran long enough to show steady error rates, customer service handled issues without escalation, and finance can reconcile agent orders. Expand product coverage in steps, watch return rates by product, and keep monitoring feed health daily.
How Alien Road applies agentic commerce and UCP
In a typical engagement, discovery is a data audit: feed health in Merchant Center, structured data on product pages, policy clarity and how fast price and stock changes reach the feed. These issues usually need fixing regardless of agents, because they affect Shopping ads and organic product results too.
Setup works with the client’s platform and developers. We prepare the feed and structured data, help define the pilot range, set up order tagging and analytics so agent orders are visible, and document the rollback steps.
Measurement tracks agent-originated sessions and orders where platforms report them, error rates, returns and support contacts versus normal orders. Reporting to the client covers pilot results, data fixes made and a recommendation on scope for the next phase.
Who needs to be involved?
Agentic commerce touches more teams than a normal marketing project. E-commerce owns the catalogue and platform settings. Developers handle integration and monitoring. Finance needs to reconcile orders and fees. Customer service handles questions about orders customers did not place on your site directly. Legal reviews terms and privacy wording.
Name an owner for the whole pilot and hold a short weekly check-in while it runs. Most pilot problems are not technical; they are a feed update that nobody scheduled or a returns question nobody could answer.
What can go wrong during the pilot?
Typical issues are stock mismatches, where an agent buys an item that sold out minutes earlier, prices that differ between feed and checkout, shipping estimates that the warehouse cannot meet, and orders that analytics fail to attribute. Each one should have a planned response, such as an automatic cancellation with a clear message, before launch.
Where to start this week
Run a full diagnostics check in Google Merchant Center and export every error and warning. Pick your top 100 products and fix their data first: GTINs, variant attributes, images and shipping details.
Then validate the structured data on those product pages with Google’s Rich Results Test. That groundwork pays off in Shopping and organic search today and prepares you for whichever agent protocol your platform supports first.