Preparing the catalog for a shopper that isn't a person
Buried inside Best Buy's fiscal fourth-quarter results discussion was a strategic claim that deserves more attention than the headline revenue figures: the company says it is "among the early retailers enabling its catalog to appear in these experiences," referring to AI-driven shopping interfaces, through a partnership with OpenAI that lets ChatGPT users browse Best Buy's products directly. Executives also said they are exploring OpenAI advertising and shopping integrations beyond basic catalog visibility, suggesting the relationship is expected to deepen rather than stay a one-time integration.
On the other side of the same bet, Best Buy also supports Google's Universal Commerce Protocol, which the company describes as "a cross-industry standard designed to enable agent-based purchasing across the web." That support means Best Buy customers will be able to buy products directly inside Google Search's AI mode and the Gemini app, not just browse them. Running both integrations in parallel, rather than choosing one AI platform as a primary partner, is a deliberate hedge against exactly the kind of platform volatility that pushed OpenAI itself to overhaul its own checkout model within the same year.
What 'agent-ready infrastructure' actually requires
Best Buy's own language for this work, preparing its systems so automated agents can browse the catalog and interact with it during shopping, sounds abstract until you map it to specific engineering work: product data clean and structured enough for a non-human agent to parse reliably, pricing and inventory feeds accurate in real time rather than batch-updated overnight, and APIs exposed securely enough that an external AI platform can query them without becoming a new attack surface.
That is a materially different technical bar than the one most retailers built their e-commerce stack to meet, which assumed a human clicking through a web page or app, tolerant of minor inconsistencies and able to recognize when something looks wrong. An AI agent making a purchase decision on a customer's behalf has none of that human forgiveness built in, and a data feed that is 95 percent accurate, which was fine for a human browsing, becomes a steady stream of bad agent-driven purchases or abandoned carts at scale. Retailers who have not yet audited their catalog data quality against that higher bar are not actually agent-ready, regardless of which platforms they have nominally integrated with.
The marketplace numbers hiding in plain sight
While the agentic commerce integrations get the forward-looking attention, Best Buy's own marketplace business is already a meaningful part of its strategy today: the U.S. marketplace generated about 300 million dollars in gross merchandise value in the fourth quarter alone, built on a seller base that has grown past 1,100 sellers, more than 90 percent of whom recorded weekly sales. Store employees can now help customers order marketplace items that aren't stocked in the physical store, blurring the line between Best Buy's own inventory and its third-party sellers' inventory at the point of in-store service.
That marketplace infrastructure is not incidental to the AI agent strategy; it is the same underlying capability. A retailer that has already built the systems to ingest, price, and fulfill a third-party seller's catalog alongside its own has done most of the structural work an AI shopping agent integration also requires: normalized product data from multiple sources, real-time inventory truth, and a checkout flow that can route an order to whichever party actually holds the stock. Best Buy's marketplace build appears to be functioning as unplanned but useful infrastructure for its agentic commerce ambitions.
Operational speed as the quiet prerequisite
Seventy percent of Best Buy's online orders were fulfilled within two days last holiday quarter, which the company says is the fastest holiday-season delivery speed it has recorded. That figure matters for the AI agent story specifically because an agent completing a purchase on a shopper's behalf, with no human in the loop to manage expectations about delivery timing, makes fulfillment speed and reliability more load-bearing, not less. A shopper who picks a slow delivery option themselves has accepted the tradeoff; a shopper whose AI agent picked it for them has not, and will blame the retailer when it disappoints.
Best Buy Business revenue crossed 1.1 billion dollars in the fiscal year, with mid-single-digit growth expected into the next fiscal year, a reminder that the B2B and marketplace sides of the business are scaling in parallel with the consumer-facing AI integrations, not separately from them. The operational discipline required to hit that delivery speed consistently is the less visible half of being agent-ready; the flashy part is the ChatGPT integration, but the part that actually determines whether an AI-driven purchase turns into a repeat customer is whether the order shows up when promised.
What this means for the ecommerce platform roadmap
The strategic choice worth copying here is not which AI platform Best Buy picked, since it picked both, but the decision to treat agentic commerce protocols as plural and unsettled rather than betting the integration budget on a single winner. OpenAI's own retreat from its original Instant Checkout model within a year of launching it is the clearest evidence yet that picking one horse early carries real re-architecture risk, and Best Buy's parallel integrations with OpenAI and Google read as a hedge against exactly that outcome.
For a retail CTO setting next year's roadmap, the actionable piece is the data and fulfillment groundwork, not the platform partnerships themselves: clean, real-time, agent-legible product data; fulfillment speed that can absorb a purchase decision made without a human's situational judgment; and a marketplace or multi-source inventory architecture flexible enough to plug into whichever agentic commerce standard actually consolidates the market. Build that foundation once, and it serves ChatGPT, Gemini, and whatever comes after both of them, which is the only version of this bet that doesn't need to be re-placed every time a platform changes its mind.
