Copilot Checkout: Stripe and Microsoft Enable Agentic Commerce with ACP and Shared Payment Tokens

When chat becomes checkout: how Stripe and Microsoft are building the rails for agentic commerce

Ask a digital assistant for a winter jacket, get product recommendations, and finish the purchase without leaving the chat window. That capability moved from prototype to product on January 8, 2026, when Stripe and Microsoft announced a Stripe-powered checkout embedded into Microsoft Copilot for U.S. users. Early partners include Etsy and Urban Outfitters, and the move follows prior pilots such as ChatGPT’s Instant Checkout that launched for U.S. users last September.

This is more than a UI tweak. It’s a rewire of the commerce stack for conversational shopping and AI agents. The key technical building blocks—Stripe’s Agentic Commerce Protocol (ACP) and a Shared Payment Token—are designed to connect assistants, merchants, and payments without exposing buyer credentials. Merchants remain the seller responsible for the order and payment (the merchant-of-record), while platforms like Copilot become the discovery and checkout interface.

Copilot Checkout: conversational commerce in practice

Copilot Checkout lets users discover inventory, ask product questions, and pay inside the same conversational thread. Microsoft is also working with PayPal to give customers branded and guest checkout options and card payments within Copilot. For Shopify merchants, Copilot can surface “Brand Agents” trained on a merchant’s catalog so recommendations and answers match brand voice and product context.

“AI is reshaping commerce and requires new infrastructure; Stripe is building that infrastructure while Microsoft applies it inside Copilot.”
— Kevin Miller, Head of Payments at Stripe (paraphrased)

How it works in 60 seconds

Quick sequence of a buyer flow:

  • User asks Copilot for a product.
  • Copilot queries merchant inventory via the Agentic Commerce Protocol (ACP).
  • The buyer confirms a purchase and a Shared Payment Token is issued—this hides the buyer’s card details.
  • Copilot passes the token to the merchant; the merchant (merchant-of-record) charges it and fulfills the order.
  • Payments, fraud signals, and order metadata move through Stripe’s Agentic Commerce Suite so merchants can reconcile and protect transactions.

Agentic Commerce Protocol (ACP) and Shared Payment Tokens explained

ACP is an open standard Stripe is promoting to let AI agents find inventory, ask product questions, and initiate purchases across many sellers. Think of ACP like a universal translator between conversational assistants and merchant systems—inventory, price, availability, and product metadata move in a common format so agents can reason about options.

The Shared Payment Token is the payment primitive that makes conversational checkout practical and safer. It’s a cryptographic token that represents a buyer’s payment method without exposing the underlying card or account credentials to the assistant. The token is passed to the merchant, who is still responsible for charging, refunds, and fulfillment.

Those two pieces together solve two core problems: discovery (how does an assistant find the right product?) and payments (how can you buy safely inside a chat?). But they raise their own operational questions: how long does a token remain valid, how are refunds routed, and how do fraud systems incorporate agent-origin signals?

Why this matters for merchants and payments teams

Conversational shopping rewrites who controls discovery and customer experience. Historically merchants optimized product pages, paid search, and marketplace listings. Now, AI agents will surface and recommend products inside a conversation, and many buyers may never see a merchant’s storefront or marketing pages.

That shift creates opportunities and risks.

  • New revenue channels: Being discoverable to AI agents opens incremental demand—think of it as adding a marketplace channel that’s powered by conversational intent.
  • Brand control challenges: Brand Agents help, but merchants must ensure their product descriptions, returns policies, and voice are encoded clearly so assistants represent the brand accurately.
  • Operational changes: Orders will arrive with different metadata and intent signals. Fulfillment, customer support, and attribution systems must be updated to handle agent-originated orders.
  • Payments and fraud: Tokenized payments reduce credential exposure, but fraud models need to use new signals (agent origin, confirmation steps, token lifecycle) to detect abuse.

Operational headaches to plan for

Practical issues that teams will encounter:

  • Refunds and chargebacks: If a tokenized charge is disputed 120 days after purchase, who starts the refund and how is reconciliation handled between the agent, platform, and merchant?
  • Attribution and analytics: How do you measure conversion, AOV, and LTV when purchases start inside an assistant rather than via a tracked landing page?
  • Inventory accuracy: Agents surface products in real time; inaccurate stock data can cause cancellations and customer dissatisfaction.
  • Customer identity and consent: What data does Copilot pass to merchants? Are buyers consenting to data sharing and marketing opt-ins at the point of purchase?

Governance, transparency, and regulatory risk

Agent-driven commerce introduces governance questions that regulators and legal teams will watch closely. Key areas to address:

  • Disclosure of sponsored results: If an agent recommends a product due to a commercial relationship, how is that disclosed to the buyer?
  • Liability and accuracy: Who is liable if a Brand Agent gives incorrect product advice that leads to harm—merchant, platform, or agent developer?
  • Auditability: You’ll need logs proving what the agent recommended, what the customer agreed to, and what data was exchanged—useful for both disputes and compliance.

“Microsoft and Stripe are working to provide reliable, rapidly evolving infrastructure so Copilot can support smooth product discovery and purchasing.”
— Nayna Sheth, Microsoft (paraphrased)

Metrics merchants should track from day one

Make these KPIs part of your tracking plan for agentic channels:

  • Conversion rate from chat session to purchase
  • Average order value (AOV) versus other channels
  • Dispute and chargeback rate for agent-originated orders
  • Time-to-fulfill and on-time-delivery percentage
  • Customer satisfaction or NPS specifically for agent-assisted purchases
  • Percentage of orders using Brand Agents versus generic agent discovery

Questions to ask your payments vendor and AI partner

  • How are tokens revoked and what is their lifecycle?

    Ask for token expiration policies, renewal flows, and how refunds are handled when tokens expire.

  • How are disputes routed for agent-originated transactions?

    Clarify whether disputes are handled the same as web-checkout disputes and what metadata accompanies the claim.

  • What metadata is propagated to our systems?

    Request fields for agent ID, user confirmation timestamps, and product intent signals to preserve attribution and auditing.

  • How are Brand Agents trained and audited?

    Get the training data scope, update cadence, and a process for correcting misleading or inaccurate brand responses.

Short tactical checklist for merchants and payments teams

  • Make product data agent-friendly: use structured SKUs, clear titles, rich descriptions, and availability metadata so ACP-powered agents can understand your catalog.
  • Map refund and dispute workflows: simulate chargebacks and refunds for tokenized payments and document reconciliation steps.
  • Train Brand Agents: create conversational scripts for common buyer intents and test them against edge cases (size questions, returns, cross-sell suggestions).
  • Update fraud models: incorporate agent-origin, token presence, and confirmation steps into scoring and rules.
  • Audit privacy flows: map data passed to agents, obtain explicit consent where required, and ensure marketing opt-in is handled correctly.
  • Instrument analytics: tag events that flag agent-origin, purchase step completion, and fulfillment outcomes.

90-day plan: a lean rollout for teams that want to move fast

30 days — Quick wins

  • Audit product metadata and add missing SKUs, images, and availability flags.
  • Run a kickoff with payments, legal, and operations to map token lifecycle and refunds.
  • Instrument analytics events for agent-origin orders.

60 days — Pilot and validate

  • Enable discovery via ACP for a subset of SKUs and run a controlled pilot.
  • Simulate dispute scenarios and refine reconciliation playbooks.
  • Test a Brand Agent script for high-intent product lines (e.g., high AOV, complex fit).

90 days — Scale and optimize

  • Roll out ACP discovery across broader catalog and measure conversion lift.
  • Update fraud thresholds based on initial pilot signals.
  • Iterate Brand Agent voice and product guidance based on customer feedback and returns data.

“Combining Copilot’s intelligent shopping with PayPal’s agentic commerce capabilities will create seamless, dependable transactions for customers and merchants.”
— Michelle Gill, GM at PayPal (paraphrased)

What’s still unclear — and what leaders should demand

Availability is currently U.S.-focused and early-stage. Broader adoption of ACP across other platforms (OpenAI, Google, Amazon) will determine whether it becomes the standard or one of many competing protocols. Leaders should push for:

  • Clear SLAs and documentation for token lifecycle and dispute handling.
  • Audit trails for Brand Agent recommendations.
  • Transparent commercial terms: fees, revenue share, and onboarding costs compared with existing channels.
  • Regulatory clarity on consumer disclosures when an AI agent recommends or promotes a product.

Bottom line for business leaders

Agentic commerce is a structural shift in how products are discovered and purchased. Platforms like Microsoft are turning assistants into storefronts, and payments providers like Stripe are building the economic rails to make those stores operate at scale. For merchants, the immediate task is practical: make inventory discoverable, lock down operational workflows for tokenized payments, update fraud models, and govern Brand Agents so your voice and legal obligations travel with the sale.

Testing these channels early gives a competitive edge. The teams that move fast will learn how conversational shopping changes conversion dynamics, fulfillment expectations, and long-term customer relationships—while also shaping the standards that will make agentic commerce reliable and fair.

Want a quick readiness checklist or a 90-day playbook tailored to your business? Use the tactical items above as a baseline and run small, measurable pilots before wider rollout.

5