Agentic commerce means AI agents find, compare and buy products on behalf of a shopper, and the Universal Commerce Protocol (UCP) is an open standard that Google announced in early 2026 with retail and payments partners to let those agents talk to merchant systems in a common way. Before connecting a store, a business should set governance rules on what agents may do, which data they see, how payments are authorised and who handles disputes.
What problem does UCP try to solve?
Without a shared protocol, every AI assistant would need a custom integration with every store’s catalogue, cart and checkout. UCP describes a common language for those steps: discovering products, checking availability and price, building a cart, applying offers and completing checkout with a payment credential the shopper has approved.
It is not the only effort. OpenAI and Stripe published the Agentic Commerce Protocol for purchases inside ChatGPT, and payments networks have announced their own agent-payment frameworks. Specifications are still developing, so treat details as subject to change and read the current documentation before building.
What are the main risks for merchants?
Agentic buying moves part of the shopping journey out of your website and into a third-party assistant. That shift creates risks that ordinary e-commerce teams have not had to manage before.
- Wrong orders: an agent misreads size, colour or quantity and the customer receives something they did not mean to buy.
- Price and stock mismatch: the agent quotes from a cached feed while your live price or inventory has changed.
- Fraud: agents can be used to automate card testing or abuse promotions at a speed a person could not.
- Disputes and chargebacks: when a customer says “my assistant bought that, not me”, it is unclear who is responsible.
- Loss of the relationship: the shopper may never see your site, brand story or cross-sell offers.
- Data exposure: agents may request more customer or product data than the transaction needs.
Which limits should you set before going live?
Limits make the first months manageable. They can be relaxed once you have evidence that agent orders behave like normal ones.
- Product scope: start with a subset of the catalogue, such as replenishable items with simple variants, and exclude made-to-order, regulated or very high-value products.
- Order value caps: set a maximum basket value for agent-placed orders.
- Quantity caps per order and per customer, to limit reseller abuse.
- Promotion rules: decide whether discount codes and loyalty offers apply to agent orders.
- Geography: limit shipping destinations at first to markets where your returns process is reliable.
What governance rules belong in writing?
Governance is the set of decisions people agree on before the technology forces them. Write them down and assign an owner to each.
Authorisation and consent
Define what counts as the shopper’s approval. The protocols generally rely on tokenised payment credentials and explicit confirmation steps, but your own terms should state that agent orders are binding only when confirmed through an approved flow. Record the confirmation event with each order.
Data minimisation
List exactly which fields an agent receives: product data, price, stock, shipping options. Customer data should flow only as needed to fulfil the order, and in line with the GDPR or Turkey’s KVKK where they apply. Review the list with whoever handles data protection.
Source of truth for price and stock
Agents must read live price and availability at checkout, not a stale feed. Decide which system is authoritative and what happens if the price changes between cart and payment, for example cancel and re-confirm.
Disputes and returns
Agree how customer service tags agent orders, what evidence is kept, and how returns work when the buyer never visited your site. Your payment provider’s rules for chargebacks still apply, so check them with the provider.
Monitoring and kill switch
Have a way to pause agent ordering within minutes if something goes wrong. Name the person who can trigger it.
How do you monitor agent orders?
Treat agent traffic as its own channel in your analytics and order system. Tag orders by source so you can compare them.
- Conversion and cancellation rates for agent orders versus web orders.
- Return rate and return reasons, especially “wrong item” or “not what I expected”.
- Chargeback and fraud flags per channel.
- Average order value and repeat purchase rate.
- Errors in the integration logs: failed price checks, stock conflicts, declined payments.
If any of these diverge sharply from your web channel, narrow the scope until you understand why.
What are the limits of the technology itself?
Agents are only as accurate as the data they receive. Thin product descriptions, missing dimensions or unclear variant names lead to wrong choices. Agents also cannot read your brand voice the way a person browsing does, so differentiation shifts to structured attributes, reviews and price. And because standards are new, not every assistant, payment method or platform supports the same features yet.
How Alien Road applies agentic commerce planning for clients
In a typical engagement, Alien Road starts with a readiness review: product data completeness, feed accuracy against the live store, checkout and payment provider capabilities, and current fraud and returns processes. The output is a gap list ranked by risk.
Next comes a governance workshop with the client’s e-commerce, finance, legal and customer service leads to agree the rules above and assign owners. The pilot is then scoped to a small catalogue segment with caps, separate order tagging and a dashboard comparing agent and web orders. Monthly reports cover those metrics and recommend whether to widen scope, hold or pause.
Treat the first written version as a draft. Review it after the first month of real orders and update it whenever a protocol, payment provider or platform changes its terms.
Where to start this week
Pick twenty of your best-selling, simplest products and check their data as an agent would see it: title, variants, dimensions, price, stock and shipping times. Fix anything missing or ambiguous.
Then book one hour with finance and customer service to draft answers to three questions: who approves an agent order, what the maximum order value is, and who handles disputes. Those answers are your first governance document.