Your next million customers might never walk into a store, never scroll a product page, and never click “Add to Cart.” They are AI agents, software that shops on behalf of humans, and they are already on your site. Retailers are moving from competing for human attention to competing for algorithmic inclusion.

Every decade or so, retail faces disruption: self-service, the rise of the mall, e-commerce, and one-click payments. Agentic AI is the next inflection point. Bain estimates that AI agents could be responsible for up to 25% of U.S. e-commerce sales by 2030.¹ We already see early evidence of this. Over 300 million Amazon customers used the shopping assistant Rufus in 2025, generating nearly $12 billion in additional annual sales.^[2] Customers who use Rufus are 60% more likely to complete a purchase.^[3]

Startups like Phia position AI as a conversational shopping companion. Perplexity is becoming a first stop for product research. These early AI experiments have become the infrastructure of a new commercial channel.

Three Channels, One Customer

The traditional debate between physical and digital commerce misses what has begun emerging: a tri-channel model in which experience, exploration, and execution each serve distinct commercial functions. These are not stacked on top of each other. They run in parallel, and customers move between them fluidly. A shopper might form a preference in-store, research online, and delegate the purchase to an agent. Or an agent might surface a product that the customer then wants to touch before committing. The channels coexist and reinforce each other.

Experience

The experience channel is physical retail. Stores remain suited for what machines cannot replicate, such as sensory evaluation and emotional connection.

Apple’s retail stores illustrate this. Customers visit not just to buy devices but to interact with products, attend “Today at Apple” sessions, and build confidence through hands-on experience. Luxury brands like Hermès treat boutiques as immersive spaces for craftsmanship and storytelling rather than transaction volume. Among Gen Z brands, Rare Beauty has shown how retail can extend community and identity, turning in-store activations into participatory cultural moments.

Even when the final purchase is digital or delegated, physical experience anchors customer preference. In a tri-channel world, physical retail is not legacy infrastructure; rather, it is a strategic asset whose value grows as routine transactions migrate to agents.

Exploration

The exploration channel is digital retail. Customers form preferences through browsing, comparison, reviews, and inspiration. But not all your customers are human. Agents are becoming voracious consumers of the exploration channel, reading every review, interpreting every product description, and comparing every price signal. The information you expose through your digital infrastructure, such as product data, pricing, inventory signals, and reviews, is the raw material from which agents form recommendations and execute decisions.

Execution

The execution channel is new. This is where AI agents handle the mechanics of commerce: routine replenishment, constraint matching, price-performance comparison, and the purchase itself. What is genuinely novel is not the supply side (inventory, pricing, and forecasting have long been machine-driven) but the demand side. For the first time, AI systems can truly bring to life the idea of personalisation, moving beyond reordering the same product to interpreting signals about style, occasion, budget, and evolving taste.

Amazon’s Buy for Me is an early operational example: the AI completes a multi-step purchase on a brand’s external website while the consumer browses and shops in the Amazon app.^[4] Consumers do not want to shop less. They want a higher return on the time they invest. Delegation becomes the default for low-emotion, repeatable purchases because it frees bandwidth for the ones that matter.

The Interface Posture Decision

The tri-channel model creates tension, forcing retailers to thoughtfully answer the question: Do you make your commerce infrastructure fully open to AI agents, selectively open, or closed? The answer likely depends on your category and where preference is formed.

In consumables and household essentials, demand is functional—availability, reliability, and fulfilment consistency outweigh storytelling. The platform that guarantees these earns the agent’s default preference. Openness is almost always the right posture. Agents will gravitate toward whichever source delivers the most reliable inventory signal and the fastest path to the doorstep.

In fashion, beauty, and lifestyle, preferences are emotional and formed upstream of any platform, through social media, editorial content, and physical experiences. Here the question is sharper: What information do you share, with whom, and in what format? A mid-market fashion brand might feed agents its full size and availability matrix while keeping lookbooks and editorial styling on its own channels, where context converts browsing into brand affinity. The goal is not to hide from agents but to choose which layer of the value proposition they see.

In electronics, specification complexity increases the value of AI-mediated comparison, and structured data, not marketing, determines visibility. The retailer whose product feed includes detailed compatibility matrices, benchmark results, and standardised feature taxonomies will consistently rank higher in agent evaluations than one relying on persuasive copy alone. Across all categories, one pattern is consistent: Agentic commerce rewards data integrity, fulfilment reliability, and coordination capability. It penalises those who have optimised only for human attention.

What to Do in the Next 90 Days

Capturing agent-mediated demand comes down to execution in four areas.

First, rewire product information architecture. AI agents need more than basic SKU data. They need rich, semantic product descriptions with detailed attribute taxonomies, relationship mappings between complementary products, usage context, and comparative advantages. The starting point is a canonical product truth: one authoritative source per product, enriched for machine consumption. Assign clear data ownership and treat product data as critical infrastructure, not a merchandising side task.

Second, expose real-time decision data. Agents making split-second choices need millisecond-accurate availability, predictive inventory based on in-transit goods, dynamic pricing and promotion eligibility, and location-aware fulfilment options. If data is stale by the time an agent reads it, a competitor’s agent gets the sale.

Third, decide the interface posture: open, selective, or closed. Define what data to expose to agents, in what format, and under what conditions. Treat machine-facing interfaces with the same design rigour applied to consumer-facing websites.

Fourth, establish organisational readiness. This is where most retailers stall. Assign clear ownership for the AI commerce channel. Stand up a cross-functional team that bridges merchandising, pricing, data, and platform engineering with the autonomy to experiment and iterate without routing every decision through existing hierarchies. Define new KPIs: agent conversion rate, share of agent-mediated baskets, and agent-driven revenue as a percentage of total. Without clear ownership, cross-functional alignment, and metrics that matter, this remains a technology experiment.

Beyond Retail: The Hybrid Future

The retail future is hybrid: Humans inspire but agents execute. The organisations that move first will not just capture agent-mediated demand, they will shape the criteria by which agents evaluate everyone else.

But this is not only about competitive advantage. As agent-mediated commerce scales, retailers who are invisible to agents become invisible to customers. Retailers need to ensure that their commerce infrastructure is readable, reliable, and relevant to the systems that increasingly decide where demand flows.

This tri-channel shift is not unique to retail. Any industry where customers make repeated, information-intensive decisions faces the same dynamic. In travel, agents will compare itineraries, loyalty redemption value, and cancellation flexibility faster than any human can. In financial services, they will screen mortgage products, insurance policies, and investment options against personal constraints at a level of precision that no adviser call can match. In procurement, they will evaluate supplier reliability, compliance credentials, and total cost of ownership across dozens of vendors simultaneously. In each case the same two questions apply: What in your model is challenged by agents becoming a primary customer interface? What early signals suggest that how customers engage with you is already changing?

Leaders need to check whether their product or service data is machine-readable. Most is not. Identify your equivalent of “routine repurchase,” the low-emotion, high-frequency decisions customers would willingly delegate. Decide your interface posture before a competitor or a platform decides it for you. And appoint an owner, because without one, this remains a topic for the next offsite rather than a line on someone’s objectives.

References

[1] Bain & Company. 2030 Forecast: How Agentic AI Will Reshape US Retail. 2025.

[2] CNBC. Amazon surpasses Walmart in annual revenue for first time, as both chase AI-fueled growth. February 2026.

[3] Amazon News. Amazon’s next-gen AI assistant for shopping is now even smarter, more capable, and more helpful. November 2025.

[4] Amazon News. Amazon’s new ‘Buy for Me’ feature helps customers find and buy products from other brands’ sites. April 2025.