A view into one payment network
On November 2, Forter published an analysis of agentic orders on its network through October. It looks at order values and differences between agents, among other things. After its stated adjustment, the provider reports an average agentic order value of $227.
Understanding the differences
Forter notes that individual shopping experiences affect the figures. For example, Google supported single item checkout in the observed flow. Differences in basket composition and available features matter before comparing agents directly.
What merchants can take from it
The more useful question is often how an agent moves through a particular store: where does it arrive, what information can it see and where does an order fail? Merchants can investigate those questions with their own tests and carefully attributed data. Forter’s observations describe its network and are not a representative measure of the whole market.
What an average can and cannot tell us
The reported order value is a starting point, not a forecast for an individual store. Forter describes differences between agents and notes that their checkout flows affect the figures. If one service supports only a single item per purchase, comparing it directly with a multi item basket needs care.
The mix of merchants, product categories and the number of observed orders also matter for interpretation. The published excerpt does not provide a complete representative sample of the market. That is a reason to put provider figures alongside a merchant’s own observations rather than apply one average to every store.
From a channel report to an economic question
The economics behind the agent asks whether a new channel creates additional value. An order attributed to an agent is not automatically an additional purchase. Intermediary costs, payment fees, fulfilment, returns and the work required from the team all belong in the calculation.
For a useful comparison, first decide how agent visits and orders can be identified. Then examine order journeys, abandoned steps and service cases in the store’s own data. Where attribution cannot be established, it should remain unknown in the analysis.
My article “When AI Meets Reality” also considered agents in the context of actual commerce operations. Forter’s figures offer a different view: observed orders inside one network.
Connecting decision and execution
The Visa benchmark considers whether models ask about requirements and make an appropriate decision. Forter looks at orders on its network. Those are different parts of shopping. Together they show why a useful test runs from customer intent through the offer to payment and service. From the offer to the service that follows develops that journey further.
Source and further reading
Read the original announcement ↗