The Receipt Is Moving to the Front of the Transaction: What Amex's Intent Framework Means for Merchants

Meta title: Agentic Payments: Why Verifiable Intent Matters at Checkout Meta description: Amex’s Intent Intelligence Framework moves authenticated purchase intent upstream. Here is what that could mean for merchant authorisation, fraud, disputes and recurring payments.

A receipt usually documents a purchase after it happens. In agentic commerce, the more important record may arrive first: evidence of what the customer actually authorised the AI agent to buy.

That is the strategic shift behind American Express’s Intent Intelligence Framework, announced on 13 October. Rather than relying on a customer’s approval at checkout alone, Amex wants authenticated purchase instructions to travel with the transaction. The idea is simple but consequential: consent at checkout is not enough if nobody can verify what the customer consented to.

For merchants, this makes agentic payments an authorisation-data issue, not just a new checkout interface.

Intent before payment

Amex’s framework builds on its ACE Developer Toolkit, launched in April 2026. The proposed flow starts with a Create Intent API, which records a customer’s instructions and constraints, such as the merchant, budget and product preferences. The customer authenticates and approves that intent. The system then creates an Intent ID and a Proof of Intent Token.

That token can be exchanged for a secure digital payment credential for use at checkout, rather than exposing the customer’s card number. The Intent ID and token can also help carry evidence of the approved intent into fraud prevention and dispute resolution.

Amex’s example shows why timing matters. If a customer asks an agent to buy white Nike sneakers in size 9, but the agent attempts to purchase Adidas shoes, the framework could block the mismatch before it becomes a completed transaction. The control is not simply “did the customer confirm a purchase?” It is “does this purchase fit the customer’s verified instructions?”

Amex says it has not yet seen an increase in disputes attributable to agentic commerce. It is building ahead of that risk as customer interest grows. In the Amex Trendex: Agentic Commerce Edition, 66% of surveyed US consumers said they were likely to use AI agents to research the best price, while 61% said they were likely to use them for product or service recommendations over the next 12 months. These are stated intentions to use AI for research and recommendations, not evidence that consumers are ready to delegate every purchase.

That distinction is important. Discovery can be delegated with relatively little risk; a purchase still needs an understandable, enforceable mandate. Amex hopes its framework will become an industry standard, but that outcome depends on adoption and interoperability across agents, merchants and payment providers.

The merchant problem is authorisation data

The payments industry is building several kinds of agent infrastructure, from Mastercard Agent Pay to tokenised-deposit experiments such as the HSBC and Ant Digital micropayment test. But agent identification, payment credentials and customer intent are distinct pieces of the puzzle.

As we noted in our analysis of Mastercard’s agent probability model, detecting that an agent initiated a transaction does not by itself prove that the customer authorised it. Intent evidence addresses a different question: what was the agent permitted to do?

This shifts the practical question for merchants. Can the checkout and payment stack recognise an agent credential and process the relevant intent data? If an agent cannot present verifiable authority in a supported format, a transaction may fail an issuer’s checks, or fall into a manual review path. At the same time, Amex’s framework is not yet a universal standard. Merchants should assess emerging requirements with their providers rather than assume that one token format or API will work everywhere.

The operational implication extends beyond checkout. A merchant’s systems may need to connect the Intent ID to the authorisation attempt, order, fulfilment record, refund and any later dispute. That gives teams a way to understand what was authorised, what the agent did and how the merchant fulfilled the order.

Fraud and disputes: build the evidence trail

An authenticated intent record could help distinguish an agent error from other causes of a disputed transaction. If the customer specified a product, budget or merchant limit, a transaction that sits outside those constraints may be easier to identify and investigate. An auditable record can also help teams examine whether an agent acted within its mandate or whether credentials were misused.

But intent is evidence, not a guarantee. Merchants still need to verify the payment credential, manage fraud controls and maintain their ordinary fulfilment and refund records. A token alone does not prove that the right goods were delivered, or settle who is responsible when an agent, merchant or other party makes an error.

This is where transaction-level data discipline matters. A useful record should be able to connect the intent reference with the agent or credential information available to the merchant, the payment attempt, the order and the final outcome. Access should be controlled and data minimised: retain what is necessary to operate, investigate and meet obligations, not an unrestricted copy of a customer’s whole conversation.

That data model may matter more than the eventual label attached to an agent payment. In our x402 analysis, we highlighted how raw protocol activity can be difficult to translate into useful merchant evidence. Intent-linked records offer a clearer way to relate machine activity to a customer instruction and a commercial outcome.

Recurring orders need rules, not a blank cheque

The framework could be especially relevant to agent-managed replenishment. Amex’s example is an autonomous reorder of household staples, where price and size might matter more than brand or merchant.

For a merchant, that is not the same as an open-ended instruction to “buy this whenever needed”. A workable mandate needs boundaries: which products or categories qualify, what price or quantity limits apply, and what should happen if the preferred item is unavailable or the price changes. These are merchant-side design considerations, not claims about features already provided by Amex’s framework.

A practical readiness checklist

Merchants preparing for intent-based agentic payments can start with six steps:

1. Map the data path. Identify how agent credentials, intent references, payment attempts, orders and refunds are recorded across your systems.

2. Ask providers about support. Confirm which agent credentials or intent data formats your gateway, acquirer and fraud tools can accept, and what happens when a transaction lacks them.

3. Define mismatch handling. Agree which cases should be declined, reviewed or escalated, such as a purchase exceeding a budget or failing a product constraint.

4. Link evidence to outcomes. Ensure operational teams can connect the authorisation to fulfilment, customer contact, refunds and dispute records.

5. Set rules for recurring orders. Specify acceptable product substitutions, price limits, quantities and when an agent should seek renewed approval.

6. Test the exceptions. Simulate expired or revoked permissions, duplicate attempts, out-of-scope products and partial fulfilment before enabling autonomous purchasing.

The receipt moves upstream

The emerging contest in agentic commerce is not only over who can initiate a payment. It is over whether a customer’s permission can be represented in a form that payment and merchant systems can verify before funds move.

That is the significance of Amex’s framework. If intent capture gains wider adoption, merchants may need to treat it as part of the authorisation and evidence chain, not merely as information gathered by an AI assistant. The “receipt” is moving to the front of the transaction because trustworthy automation requires a record of what was authorised before the agent acts.

For merchants managing online, in-person and recurring payments, the direction points towards joined-up commerce operations: flexible payment capabilities, connected transaction data and automation that extends from checkout through reconciliation. Quantum Payments’ unified commerce and modular platform positioning reflects the importance of managing those workflows together. As standards develop, the merchants best prepared will be those that can adapt their payment and operational systems without losing sight of the customer’s intent.