Build One Connected Marketing System That Can Cater to Agents
Companies prepare for agentic commerce by making their product data easy for AI agents to read, finding new signals to replace lost clicks, testing how each AI model chooses, deciding offers in real time, writing rules for which agents they serve, and measuring what agents recommend and buy. All six work best as one loop, inside the limits of customer consent.
Six Ways to Prepare for Agentic Commerce
Each step answers a change that happens when the buyer is an agent.
Consent is implied and should be carried through all 6 steps. Every agent request carries a person's permission for specific data, a specific purpose and a specific time. Your system has to honor those limits every time.
AI Agents Are Already Completing Purchases
New protocols, shared rules that let an agent read a store and pay, are now switched on by default across major commerce platforms.
Six Questions to Answer to Prepare for Agentic Commerce
Each step answers a change that happens when the buyer is an agent.
Is our product data complete, accurate and the same everywhere an agent might read it?
Do we know what each major AI assistant says about our brand today?
Which agent requests and orders can we capture now, and where will that data live?
Could we answer a price request in seconds, inside our margin and brand rules?
Who has the authority to decide which agents we serve and what they can be offered?
Which metrics will tell us whether a sale made by an agent was good for the business?
FAQ
Your agentic commerce questions answered.
Agentic commerce is shopping done by artificial intelligence (AI) agents on behalf of people. An AI assistant, such as ChatGPT, Gemini or Amazon's shopping assistant, finds products, compares them and completes the purchase. The shopper sets the goal and gives permission. The agent does the searching, comparing and checkout.
New protocols make this possible. They let an agent read a store's catalog, fill a cart and pay. The person often never visits the store.
Three lead today. The Agentic Commerce Protocol (ACP) comes from OpenAI and Stripe and launched in ChatGPT. The Universal Commerce Protocol (UCP) comes from Google and Shopify and launched in Google AI Mode and Gemini. The Agent Payments Protocol (AP2) lets an agent pay for a person using a digitally signed permission called a mandate.
In e-commerce, a person visits the store, sees ads and pages, and clicks through to buy. In agentic commerce, an agent does that work inside the assistant. It reads structured data, reviews and price rather than creative, and the brand often sees only the final order.
No. The protocols handle the transaction. They don't decide which agents you serve, what you offer, at what price or under which rules, and they don't tell you whether a sale was good for the business. Those are still marketing decisions.
Path-based attribution needs a logged view or click, and an agent makes neither. Mix models and holdout tests still work, but they need new inputs. Useful new metrics include share of agent recommendation, agent conversion rate and offer acceptance rate. Because brands, retailers, retail media networks and assistants each see only part of a sale, brands and retailers will need to measure together.
Every agent request carries a person's consent, and that consent covers specific data, a specific purpose and a specific time. Your system needs to read those limits on every request, apply them to signals, audiences, offers and measurement, and keep a record of how it applied them.
A senior leader with authority across marketing, merchandising, finance and legal. That person owns the rules for which agents you serve, what each can be offered and which data flows each way.
Keep Reading
More resources on the marketing operating layer.
