Agentic shopping promises to compress hours of product research into seconds and turn routine purchases into automated tasks. But for retailers, the shift is not simply a matter of adding an AI tool to an online shop. It requires product information, stock and pricing to be accessible to software, alongside clear controls over what an agent may do.
That is the view of Saurabh Pitkar, Dell’s Director of Product Management for Agentic Commerce and Agentic AI, who spoke to CX To...
Continue Reading This Article
Enjoy this article as well as all of our content, including reports, news, tips and more.
By registering or signing into your SRM Today account, you agree to SRM Today's Terms of Use and consent to the processing of your personal information as described in our Privacy Policy.
Pitkar sees two especially useful consumer applications. An agent can help people make sense of a complex purchase by searching and comparing products against an unclear set of preferences. It can also take the friction out of repeat buying, where a customer already knows what they want and simply needs the same item again.
Those different tasks place different demands on retailers. For considered purchases, agents need accurate product details that can be compared consistently. For repeat orders, they need dependable access to current prices, stock and a straightforward way to complete a transaction.
OpenAI’s commerce guidance describes a similar journey: a shopper states a need in natural language, and ChatGPT can draw on public information, retail product feeds and, where available, private integrations to surface options. A merchant may allow an agent to follow its existing website journey or provide structured actions through an integration. The latter can give the business more control over how requests are checked and processed.
Pitkar’s first priority for retailers is to connect product information with live inventory and pricing. A catalogue that lists an item without reflecting whether it is actually available, or what it currently costs, is of limited use to an agent making a recommendation. Linking product data to stock held across warehouses, factories and distributors, and to pricing decisions, could also help businesses manage the commercial terms on external channels.
A second priority is to make commerce functions available without requiring an agent to imitate a person clicking through a website. In a headless system, actions such as checking availability, creating a basket or placing an order are exposed through interfaces that software can use directly.
Microsoft’s Dynamics 365 Commerce documentation offers one example of this approach. Its Model Context Protocol server connects agents to existing commerce functions, including product search, inventory checks, promotions, carts, checkout and order information. Because those actions use the retailer’s existing engine, pricing, discounts, tax and order records can remain consistent with other sales channels. Access can also be controlled through authentication and permissions.
The broader technical landscape is still taking shape. Google’s Universal Commerce Protocol is presented as an open-source means of connecting consumer interfaces, retailers and payment providers, while supporting existing retail infrastructure. Google says merchants can retain their own business logic and remain the merchant of record. The protocol can work with APIs, Agent2Agent and the Model Context Protocol, and is designed to support retailers’ existing checkout arrangements.
Pitkar argues that businesses will need to monitor activity across channels as agent-led journeys become more common. A conventional website visit provides signals such as browsing behaviour and time spent on a page. An agent may not reveal the shopper’s original intent in the same way, leaving retailers to develop other measures of how products are discovered and whether those interactions lead to sales.
HUMAN Security’s practical guide to agentic commerce recommends separating AI-related activity from human traffic in measurement systems and tracking how agents behave, including their speed and whether activity leads to referrals or purchases. It also advises retailers to make product, stock, pricing, shipping and returns information machine-readable, and to provide documented, authenticated transaction endpoints. For businesses that still receive browser-driven agent traffic, the guide recommends simpler pages and checkout journeys, with fewer disruptive pop-ups and unpredictable changes.
Marketing may also need to adapt. Pitkar expects human shoppers and agents to coexist, but says the two respond to different kinds of information. People may be swayed by presentation or urgency prompts; an agent is more likely to compare concrete details such as price, specifications and availability. Retailers may therefore have to divide spending between campaigns aimed at people and efforts to make their products legible and competitive in agent-led discovery. Pitkar’s view is that revenue and conversion remain important measures, even as traditional signals about how a customer first arrived or what they intended to buy become harder to obtain.
For business-to-business commerce, the stakes may be higher. Pitkar sees potential in automating sales processes and recurring purchases, but says organisations will need to verify the identity of both buyer and seller agents and set limits on what they can agree. A purchasing agent negotiating a bulk order, for example, should not be able to exceed an authorised discount or spending threshold.
Federal Reserve Governor Christopher Waller has drawn a distinction between agent-assisted shopping, in which a person remains responsible for the decision and payment, and agent-delegated commerce, in which the user authorises an agent to act within defined limits. Waller’s speech notes that recurring, rule-based B2B purchases may be well suited to automation, while the larger sums involved make errors or unauthorised actions more consequential. Authentication, liability and fraud remain central concerns.
Visa’s Trusted Agent Protocol is one proposed response to the identity problem. Its specifications describe signed credentials that can help merchants recognise an agent’s commercial purpose and verify that its requests have not been tampered with or replayed. Such signals could help distinguish authorised shopping activity from malicious automation, although adoption remains optional and no single approach has yet prevailed.
That lack of agreement is itself a challenge. Pitkar points to the emergence of multiple protocols from technology and payments companies, arguing that greater convergence would make it easier for businesses to choose where to invest and how to establish secure transactions across platforms. The Federal Reserve speech likewise notes uncertainty over whether interoperable standards or platform-specific systems will dominate.
Trust is not the only risk. If agents can make rapid, repeated requests, fraud or technical weaknesses could be exploited at a scale that is difficult to manage. Automated calls to customer service or bursts of searches may also create unexpected demand. Pitkar says retailers will need close monitoring and effective anomaly detection, alongside clearly defined responsibility for incorrect purchases, returns or other losses.
The transition is therefore about more than enabling an agent to find a product. Retailers must decide what information it can access, which actions it can take, how those actions are authenticated and where responsibility lies when something goes wrong. For businesses that can establish those foundations, agent-led shopping may open new routes to customers; for those that cannot, the convenience offered to shoppers could expose gaps in their commerce systems.
Source: Noah Wire Services



