When the Customer is a Machine

The page, the placement, and the promotion were all engineered for a human eye and an impulsive wallet, and now they are all landing on something that has neither. A machine agent filters and does not browse. It does not warm to a brand and instead reads a spec. The most expensive instincts in retail were honed for an audience that is quietly delegating the visit to machines.

Agentic Commerce inverts the discipline, moving retailers from channel optimization to machine-mediated selection. Adobe clocked a 4,700% jump in traffic to US retail sites from AI browsers and chat tools in a single month last summer; Bain expects agents to mediate $300bn–$500bn of American commerce by 2030. The money is real and arriving faster than most retailers’ strategies can adapt. The retailer’s job is changing from winning attention to being selected in comparisons it will never witness.

Two of retail’s most profitable inventions depend on the very thing the agent removes, a person looking at a screen. Loyalty and Retail Media were both built to monetize attention. Both now sit in the machine’s blind spot.

Loyalty, and How to Keep It

Retailers need to rethink the loyalty strategy they currently employ. A points balance, a tier, or the fact that a shopper is one order from a reward may not register with a system optimizing price, speed, and availability.

Deloitte surveyed over 100 major retailers. Eighty-one percent reported that they believed Generative AI would reduce customer loyalty in the United States by 2027. Although customer abandonment is unlikely, loyalty may shift to the AI that provided the recommendation, leaving the retailer to fulfill an order it no longer fully controls.

Two things, legibility and moat (or unique value), are key to solving this issue. Legibility refers to making certain elements of the loyalty program visible and readable via API. For example, tiers, thresholds, and rewards should be made visible so that an AI can recognize that purchasing an item from Company X instead of Company Y will earn a free delivery or help clear a reward threshold. Benefits that an AI cannot measure are effectively invisible.

Next, building a moat means creating unique benefits that an AI cannot arbitrate away, including exclusive products, early access, members-only drops, and services such as installation/extended warranties. Scarcity and Identity are harder to compare than prices. The retailers who retain their customers will be the ones that protect their value by providing offers and promotions that no generic AI model can provide.

Retail Media, and Where It Goes Next

Retail media, one of the most profitable revenue streams for large retailers, faces harder arithmetic. Its revenue comes from surfaces the agent skips, such as sponsored slots, banners, and promoted listings that lose value when a machine assembles the shopping cart. Forrester predicts agentic commerce will cut retail-media ad sales by a fifth in 2026. The pain compounds because the market is already lopsided, with Amazon and Walmart together commanding roughly 84% of US commerce-media spending, which leaves a long tail of networks built on on-site display with the most to lose and the least leverage.

The path forward is to move from selling placement to selling influence. What a retail media network ultimately owns is the first-party purchase data underneath the banner ad. As agents are searching for signals to rank products, this data becomes more valuable.

Those retail media networks that survive will license this data and compete for position inside the recommendation itself, trading click-through rate for a new form of currency, the rate at which an agent’s choice can be influenced.

The second step is to move back into areas that an agent cannot reach. Retailers should also protect the physical and in-store touchpoints that agents cannot easily access. No algorithm is intercepting a shopper halfway down aisle seven. In-store screens and shelf-level media remain, for now, out of the machine’s reach.

The Infrastructure Has an Order

None of this works without the right infrastructure foundation. Each layer needs to be built in sequence so that subsequent layers can be built upon it.

Structured Product Data. If an agent cannot understand structured product data, it cannot confidently select a product. Inventory, pricing, delivery, and returns need to be accessible through clean APIs (the new SEO), which are quickly becoming a new form of commerce visibility.

Interoperability. There are several interoperability protocols, such as Anthropic’s Model Context Protocol (MCP), the agentic commerce protocol developed by OpenAI and Stripe, and Visa’s Trusted Agent protocol. They are moving from lab testing to pilot testing. Retailers can start testing now or retrofit later.

Agent Identity and Authorization. Controls used for preventing fraud when using human interaction with button presses are completely inadequate for a machine that acts based on its own mandates. The signed authority required for each transaction, along with tokenized credentials and verification of the Agent, are essentially the tolls that must be paid for every transaction.

Agent-legible Loyalty. When a retailer becomes visible to an Agent, either because they were able to find them or become trusted, they also want to be preferred over others. All of the loyalty mechanisms listed above are readable by the Agent and defendable by the brand.

Fulfillment as a Ranking Factor. Agents rank on speed, reliability, and cost. Any delivery promise made by a retailer that it can actually keep is now a marketing asset rather than simply a logistics footnote.

Governance. Decisions must be auditable. Human escalation paths must be clearly defined. Compliance requirements, including obligations under the EU AI Act, need to be built into the operating model. This is the final piece needed to allow all of the elements below it to grow/expand at full speed without creating legal problems.

If any of these steps are skipped, the previous levels remain stranded, i.e., agents may not find the retailer, transact with it, trust it, repeat business, or scale.

This Belongs on the Board Agenda

One option available to the incumbent retailers is to pass responsibility for implementing this downstream and come back to it after the first year. But that would be a mistake. Leading retailers are already moving, for example, within one quarter, Walmart, Target, and Etsy wrote themselves into all the major assistants that exist today.

The retail stores that ultimately prevail will continue to maintain two front doors for their machines, one for the machines now arriving in massive quantities, and another for those humans who still wish to walk in. Customers are already beginning to adopt agents. The question for retailers is whether agents will adopt them.