The arrival of agentic commerce is prompting nervous conversations across ecommerce and retail media teams. If AI agents shop on behalf of customers, surely sponsored search, the golden goose of retail media, is irrelevant?
We contend that the opposite is more likely and that agentic shopping creates big opportunities for retailers and big risks for brands.
Assuming that they are trusted sufficiently by customers to make good choices on their behalf (not a given in my view, not least because the agent is unlikely to know what products are precisely needed at any given point in time - but I digress), the first wave of agents will act in order to make the shopping trip faster, better, cheaper.
It’s critical, therefore, that retailers (and brands) and RMNs understand how agents will shop and how to influence the shopping trip.
Agents will make their product choices based on information within product detail pages (PDPs) and so, these pages, , take on even greater importance in an agentic future.
Testoimomething out and writting textt down to figure ouywhat happens with yyperlinks in Webflow CMS view. Testtestststststst linke goes above.
But, and here's the key insight - agents will not exhaustively evaluate every sku. Agents will be optimised for good enough certainty, not theoretical perfection. Agents will therefore identify a candidate set, evaluate candidates until confidence exceeds a certain threshold and then stop. They will typically operate within rules like:
“Accept the first option that is within X% of the best expected utility” or “compare until marginal improvement falls below Y” or “compare until confidence exceeds Z”
This means that sku #1 is evaluated, skus #2, #3, #4 are checked to see if an improved outcome is likely. The agent will stop when further search is unlikely to improve the outcome and confidence in meeting the task is sufficient.
In grocery shopping , there are typically hundreds of skus in a particular category. Many of them would be satisfactory to the particular task. The order in which they are presented to the agent is therefore critical and this is something that retailers will control and brands must influence.
In grocery categories where products are highly substitutable and differentiation is marginal, ranking position becomes the single most powerful determinant of agent selection.
Grocery retail has three structural properties that amplify ranking power:
In that environment, the marginal gain from comprehensive search collapses quickly.
Agents will not try to find theoretical perfection. They will find solutions that are “good enough” with high confidence. They will optimise for expected performance, under uncertainty, subject to constraints and with bounded search costs. This is what economists call "satisficing". When performance differences are marginal and if the first few evaluated skus cluster tightly, confidence thresholds will be reached quickly and so the need to keep searching ends and a product is selected.
Product ranking therefore defines where the agent stops and what skus it buys.
In agentic grocery shopping, being evaluated first, second or third will often be equivalent to being chosen. The order in which skus are presented to the agent is therefore critical and this is something that retailers will control and brands must influence.
Ranking determines the candidate set. Retailers determine ranking and, just as they do with sponsored products, can charge brands based on where they are ranked.
Just as now, the retailer will balance the need to make a sale with the need to generate retail media income but the retailer doesn't need to worry, as it does now, about the customer experience. Agents will be happy with any ranking that enables a sufficiently qualifying purchase. Aesthetics and ads will be ignored.
Just as now, the retailer will act on all the data that it has available to determine the product subset and their rank.
Retailers with loyalty programmes (whose rewards many agents will have been instructed to optimise) and the ability to identify non logged in customers/agents and understand their purchase history will have an advantage. The ability to infer, from analysis of past behaviour, the agent's (often conflicting) meta selection criteria (e.g buy organic, optimise my loyalty points, choose from what I normally buy, go for the cheapest etc etc) will also create competitive advantage.
Agentic commerce will increase retail media income because product ranking is even more important for agent decision making than it is for human decision making and product ranking is controlled by retailers. Brands will almost certainly pay significantly more for position than they do now given that branding, pack aesthetics etc will confer no benefit
There are many things that retailers, RMNs and Brands need to do to win in an agentic commerce world. Understanding search ranking and how this is impacted by the retailer's available purchase history and other data, customer (and therefore agent) prior behaviour, loyalty programme, identity management capability and PDP attributes are the key places to start.
If you need support with any of this, we are happy to help.
Agentic commerce won’t kill retail media. It will grow it. Retail media will continue to be a major growth driver for retailers and brands.
