The Maze: AI commerce finally has a planning number. AI platform-driven US retail ecommerce sales could reach $144.45 billion in 2029, equal to 8.8% of online sales. That is big enough to matter in annual planning, but narrow enough to handle carefully. The printed 2026-2028 labels are redacted, so the visible bars give a rough ramp, not permission to invent precise yearly forecasts.
The endpoint is the signal. The ramp is a warning label. The 2029 value turns AI platforms from "interesting traffic source" into a distribution question. Retailers will need product data that assistants can parse, content that answers comparison questions, and attribution rules that survive sessions starting outside owned search. The earlier bars imply roughly ~$20.3 billion, ~$44.0 billion, and ~$84.4 billion, but those are visual estimates. Useful? Yes. Official labels? No.
The definition is where the strategy gets sharp. The forecast counts third-party AI platforms, same-session attributable orders, AI-powered browsers, retailer apps inside AI platforms, and Google AI Mode. It does not count retailer-native assistants, specialized shopping assistants, AI search summaries, or browsing that does not immediately convert. Translation: this is not "all AI in retail." It is the piece where the demand interface moves away from the retailer.
The shopper workflow is moving before the purchase button. Hundreds of millions of people already use assistants to compare products, refine options, check prices, and read up-to-date product details. That matters because the retailer may lose the first sorting moment before the shopper ever lands on a product page. Product feeds, review coverage, specs, availability, and answer-ready merchandising become upstream growth work, not back-office hygiene.
Checkout is becoming part of the interface layer. A shopper can already buy inside chat from supported merchants, while AI Mode is being built to narrow products through Gemini and the Shopping Graph before purchase. That does not mean every assistant session becomes revenue. It does mean the buying surface is creeping closer to the recommendation surface. The commercial question is no longer just "did AI send traffic?" It is "did AI decide who made the shortlist?"
The consumer signal supports the forecast, but does not prove every step. Among AI-tool users, 44% have used these tools for product recommendations or comparisons, and 48% say AI assistance has influenced a final purchase decision. Those numbers explain why a $144 billion endpoint is plausible. They do not turn a redacted annual path into disclosed data. That distinction matters. Strategy needs ambition. Forecasting needs humility.
Why it matters: Retailers should treat AI platforms as a new demand layer, not a novelty referral source. The work is boring in the best possible way: cleaner feeds, stronger comparison content, structured product data, better availability signals, and tests that separate incremental demand from attribution theft. The winners will not be the teams saying "agentic commerce" the loudest. They will be the teams that know where assistants actually move shoppers, where they only rearrange credit, and where the forecast does not apply.
Sources: Forecast | Shopping research | Instant checkout | AI Mode shopping
