Simon Hunter, Director, Agentic Commerce & Payment Optimisation at Signifyd
For most of the internet’s history, the mechanics of buying something online have been relatively straightforward. A customer will go onto the internet to search, compare, decide, and then pay. Retailers can track and see the journey as it unfolds, observing the page a shopper lands on to the moment they abandon their basket or complete checkout.
This model is changing. Consumers are already turning to AI to do more than surface options. With 23% of Americans having used AI to buy something in the past month, it’s an early sign that AI-assisted purchasing is moving beyond experimentation. In the next phase, it’s a natural evolution that agents will be able to act on instructions with differing levels of human oversight.
This is what makes agentic commerce more consequential than another shopping interface. By having software acting on behalf of a human consumer, it introduces a new participant into transactions. For merchants and payment providers, the question should have moved beyond whether an AI agent can complete a purchase. The technology has already moved in that direction.
Wider considerations should be around whether everyone involved can establish information to inform its decision and who is responsible when something goes wrong. This is the central commercial challenge.
AI is moving from adviser to buyer
Not all AI-led commerce will look the same. The most immediate change is product discovery. Consumers are using AI to compare alternatives and receive recommendations before even setting eyes on a retailer’s site. In these cases, the customer still makes the final decision and completes the purchase themselves.
The next stage is more material. A buyer might ask an agent to find and book the most suitable hotel within a set budget, reorder household essentials when prices fall below a threshold, or select a gift that meets a particular brief. The agent is no longer only presenting information. It is acting for a customer, within a mandate set by them.
Over time, those mandates will become more sophisticated as AI learns about buying decisions and tastes. Some purchases will still require explicit approval but others, such as repeat orders or routine bookings, may be completed with little or no intervention. The point is not that every consumer will hand over every buying decision to an AI agent, but that the line between recommendation and transaction is increasingly blurred.
For merchants, this means preparing for a future in which their products are assessed not only by humans but also by systems weighing information such as price, availability, delivery terms and returns policies.
Fraud decisions cannot rely on old signals
The evolution of this technology has implications for fraud and risk. Most e-commerce fraud controls have been built around the behaviour of a human shopper; for instance, the device they use, how they move through a site, the time they take to complete an order, and the signals they might leave behind at checkout.
An AI agent may not produce any of those behaviours, which means traditional fraud controls only apply to part of the customer base. The risk is twofold: if retailers treat every unfamiliar agent interaction as suspicious, they risk turning away genuine customers at the very moment a new channel is emerging. If they treat it as legitimate by default, they create an opening for fraudsters to exploit the same technology to act at speed and scale.
The answer is not to add another layer of authentication. Retailers must build a clearer picture of delegated authority: who the agent represents, what it has been permitted to do, the boundaries of that permission and a record of the decisions it makes. This evidence will be essential when a customer disputes a transaction or if a retailer needs to establish whether an agent acted within its mandate.
The next battleground is trust
When e-commerce first began to take hold, convenience alone was not enough. Consumers had to be persuaded that it was safe to enter card details online and that an unfamiliar retailer would deliver what it promised. The businesses that earned that trust helped make online shopping routine.
Agentic commerce is likely to follow a similar path. Consumers may be happy to delegate routine purchases and early-stage research before handing over bigger decisions that are expensive, personal or difficult to reverse. But the more an agent is permitted to do, the more clearly the customer will need to understand the limits of that permission and how they are protected if the outcome is not what they intended.
The more consequential shift may be in where commercial power sits. When agents become the primary route through which customers discover and compare products, retailers will compete less for attention on a webpage and more to be selected by systems applying a customer’s preferences. Price will matter, but so will the reliability and clarity of the information an agent receives and the confidence that a transaction can be resolved fairly if it fails.
That makes trust a commercial asset, rather than a back-office concern.
About the author: Simon Hunter is Director, Agentic Commerce & Payment Optimisation at Signifyd. His work focuses on agentic commerce and its implications for merchants, payment providers and digital risk.