How are different AI agents reshaping the retail shopper journey?

Bain & Company's report on agentic AI in retail identifies three types of AI shopping agents now emerging, each reshaping retail in different ways: "Third-party 'objective' agents" such as Perplexity, ChatGPT, and Gemini that "crawl vendor sites, aggregate listings, compare prices, read reviews, and recommend products"; "on-site retailer agents" that retailers build themselves, such as Amazon's Rufus, which Bain says helped generate "nearly $12 billion in incremental annualized sales last year"; and "off-site retailer agents" that extend a retailer's own agent beyond its own inventory, such as Amazon's "Buy for Me" feature, which "can purchase on other brands' sites, even as the retailer restricts other agents from scraping its own listings."

Bain frames the three types as carrying different stakes. Third-party agents pose "the greatest immediate threat to the traditional retailer-shopper and retailer-advertiser relationships" because they disintermediate the top of the funnel — Bain cites estimates from Similarweb, a digital marketing intelligence company, that "shopping referrals from ChatGPT have more than doubled in the US, France, UK, and Germany over the past year" — yet AI still accounts for "less than 1%" of total retailer traffic even where it reaches "up to a quarter of referral traffic" for some retailers. On-site agents let retailers keep "proprietary retailer data, such as return rates and reasons, expert reviews, and longer-term purchase history," and Bain reports that Rufus's monthly active users were "up 115%" and "were 60% more likely to complete a purchase." Off-site agents like "Buy for Me" let a retailer extend its own agent's reach while still gatekeeping its own listings — a hedge against disintermediation that doesn't require ceding the loyalty relationship.

The taxonomy sharpens Will brands succeed by building their own AI shopping agents?: Bain's on-site and off-site categories are two concrete strategies for becoming the agent of choice rather than being chosen by a third-party one. It also supplies retailer-side context for Do consumers want to approve AI shopping purchases before they happen? — Bain reports that shoppers "trust retailers' on-site agents three times more than third-party agents," a trust gap that helps explain why retailers are racing to build their own agents rather than cede the relationship to third parties. Could AI agents reassemble shopping baskets across multiple retailers? describes the more consequential shift Bain sees past this taxonomy.

Bain's figures (the 115% user growth, the 60% higher purchase likelihood, the $12 billion in incremental sales) are the company's own retailer-reported estimates, not independently audited numbers, and the report does not say how its referral-traffic or trust-gap figures were measured. The taxonomy is also a near-term snapshot — Bain itself says "timing and depth of adoption are uncertain" — so it describes the current landscape of agent types rather than a stable or final structure for how agentic retail will organize.