3 minutes
Agentic commerce: what it means for your business
Get a clear view of how AI-driven purchasing is changing checkout, fraud and liability, and what merchants should do now to prepare.
Key points
- Agent-merchant interactions fall into two categories: closed loop, where a merchant's own agent controls the full experience, and open loop, where general-purpose agents such as ChatGPT or Gemini direct users toward a purchase.
- Merchants preparing for agentic commerce should address three layers: discovery (clean product feeds and structured data), checkout (API-accessible cart logic) and payments (tokenised credentials rather than raw card data).
- Worldpay's modular credential management system and Ravelin's agent detection capability let merchants adapt to new protocols such as ACP and UCP without rebuilding their existing stack.
Picture a customer who never opens your website. An AI agent compares products across the web, builds a cart and completes checkout on their behalf. For some merchants, that is not a future scenario. It is already happening in narrow, human-in-the-loop form, and the shape of it is changing fast.
Worldpay, now Global Payments, recently hosted a webinar on agentic commerce to unpack where things stand and what merchants can do now. Reshmi Suresh, head of agentic commerce, and Harsh Mehta, head of AI and agentic commerce partnerships, covered how agents are showing up in commerce today, what merchants need to prepare and where Worldpay is investing to reduce the load on merchant teams.
What agentic commerce means right now
Mehta defined agentic commerce broadly: any purchase journey where an AI agent plays a role, from product discovery through checkout and after the sale. Most agent activity today is reactive – a person asks a question and the agent responds. The more advanced version, where an agent acts on a person's behalf without being asked, is not yet widespread despite the volume of commentary suggesting otherwise. Mehta framed that move from responding to acting as the underlying promise of agentic commerce.
"Agentic commerce is... any purchase journey where an AI agent plays a role, from product discovery through checkout and after the sale."
Closed loop versus open loop
Mehta split merchant-agent interactions into two buckets. Closed loop covers agents a merchant builds and controls directly, such as Walmart's Sparky or Ralph Lauren's Ask Ralph, where the merchant owns the full experience, the data and the checkout. Open loop covers general-purpose agents outside merchant control.
Within open loop, Mehta identified three patterns:
- Advisory agents that recommend products and link out to a merchant site
- Human-in-the-loop agents (ChatGPT, Gemini, Copilot) that handle discovery while a person confirms the purchase
- Browser agents such as OpenAI's Operator or Amazon's Buy for Me that attempt fully autonomous shopping
Fully autonomous flows remain limited today, largely because merchant fraud tools and CDNs are built to block bot-like traffic, which raises latency and lowers transaction success rates for these agents.
Three layers merchants need to prepare
Mehta broke the general-purpose agent journey into discovery, checkout and payments.
Discovery still depends on fundamentals merchants already know: clean product feeds, structured data, accurate descriptions, availability, pricing and reviews. The difference is that content now needs to serve an agent interpreting intent, not just a search ranking or a product page.
Checkout requires agents to interact with live commerce logic – cart creation, tax, delivery options, address validation and promotions. Suresh noted that checkout has historically been a human interface; agentic flows need parts of it to be reachable through an API instead. She pointed to two live models: Google's UCP, where checkout happens within Gemini or AI Mode, and OpenAI's ACP, where the cart is built in ChatGPT but checkout completes on the merchant's site.
Payments should rely on tokenised credentials rather than raw card details – PSP tokens, network tokens, wallets or emerging agentic tokens with fraud, authentication and liability controls attached.
Suresh added two low-effort first steps: Sign up for ACP and UCP waitlists now, and confirm whether a commerce platform already integrates with Google Merchant Center or OpenAI's product feed requirements.
Fraud, identity and liability
Suresh outlined four areas Worldpay is building toward:
- Know-your-agent identification through Ravelin's agent detection, which uses signals such as web bot auth to flag and score agent traffic
- Intent verification, relevant as purchases move toward semi-autonomous or autonomous flows, where frameworks from Visa, Mastercard and Amex are forming around agentic tokens
- Credential management, an acquirer-agnostic system that supports scheme tokens, major wallets, and OpenAI's delegated payment token
- Dispute readiness, with additional intent data being collected proactively so merchants have a defense record if a dispute occurs
On liability, Mehta noted the industry has not settled how responsibility shifts when an agent completes a purchase a customer did not want. Existing card-not-present frameworks apply for now. Mastercard's verifiable intent framework, which uses selective disclosure so each party sees only the intent information it needs, is one early approach. Amex's ACE program has introduced customer protection language for approved agents, though full details are not yet published.
Suresh said Worldpay is not currently seeing evidence that agentic transactions carry different friendly-fraud characteristics than existing channels, and will update merchants as scheme guidance develops.
Who should act now
Mehta offered a way to prioritise: businesses with a younger, AI-native customer base, high-deliberation categories such as electronics or skin care, standardised products with simple checkout and large catalogs where customers face decision fatigue are seeing agent adoption sooner. Merchants in those categories should start building a plan and consider signing up for ACP and UCP. Categories involving regulation, identity verification, high customisation or complex liability will take longer to mature, and Mehta recommended those merchants focus on discoverability while keeping their stack ready.
Four steps for merchants to take today
- Sign up for ACP and UCP waitlists to signal intent early.
- Treat the product feed as the storefront – accurate, current data determines whether an agent finds and recommends a product.
- Continue investing in SEO fundamentals alongside newer techniques, since both affect visibility to agents.
- Assess the current stack: whether checkout is API-accessible, whether fraud tools block legitimate agent traffic and whether a PSP supports ACP and UCP.
Suresh and Mehta closed by pointing merchants to their relationship managers for specific questions, and noted Worldpay plans to share updates as scheme and protocol guidance – including AP2's move to the FIDO Alliance – continues to develop.
Related insights
