Retail's signed AI returns sit in service, content and checkout
The AI results retailers will put in writing come from narrow jobs: customer enquiries, marketing content and orders through assistants. The largest grocers describe AI programmes without an outcome figure, and pricing is where the rules arrived first.
What's genuinely in production
Customer service deflection is the best documented use. Ingka Group said its Billie chatbot resolved about 47% of the IKEA enquiries it received from 2021 to 2023, around 3.2 million interactions, saving nearly EUR 13 million, and reskilled 8,500 contact-centre co-workers for remote selling. Grupo Falabella's WhatsApp agent resolves 60% of inquiries on its own after a nine-week build, in a Salesforce customer story quoting named Falabella managers.
Content production has the newest signed figure. Zalando reported AI-generated content going from near zero to 90% of on-site marketing content in a year, with output up 70%. The company later narrowed that to banners, trend pages and teasers, not the product detail pages people buy from, and it built its own product-accuracy checks to catch details the models invent.
Retailers' own assistants, not checkout inside someone else's. Walmart says users of its Sparky assistant were up 70% year on year and spend 40% more per order than non-users (Q2 FY27 call, 20 August 2026). That compares people who chose the tool with people who did not, so it shows who uses it more clearly than what it causes.
What's still pilot theatre
AI programmes announced without an outcome figure. Tesco's 2025/26 results report about GBP 535m of Save to Invest savings and an AI assistant trial with about 280,000 colleagues, but attribute no saving to AI. Kroger's 22 September 2026 AI strategy covers forecasting, a shopping assistant and AI training for 400,000-plus associates, with no outcome metric. Not proof of failure; proof there is no public number to benchmark against.
Checkout inside a third-party chatbot. Walmart's in-chat conversion was three times lower than linking out, and it took checkout back into its own agent. The only large retailer to publish a figure published a bad one.
Camera systems that identify people for loss prevention. The FTC banned Rite Aid from facial recognition surveillance for five years in December 2023 after its 2012 to 2020 system falsely flagged customers as shoplifters. We have found no sourceable loss-reduction figure from any retailer for this category.
Vendors that keep appearing
| OpenAI / ChatGPT | the channel for Walmart's in-chat checkout test and its Sparky replacement |
| Shopify | AI sales channels on by default for eligible stores; the merchant stays merchant of record and can switch channels off |
| Salesforce | agent platform behind Grupo Falabella's 60% WhatsApp resolution figure |
| In-house data teams | Zalando built its content pipeline and checks itself; Tesco credits its dunnhumby scientists |
Regulation coming
New York's Algorithmic Pricing Disclosure Act (General Business Law section 349-a) requires the words THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA beside any personalised algorithmic price, with penalties of up to USD 1,000 per violation. Judge Jed Rakoff dismissed the National Retail Federation's First Amendment challenge on 8 October 2025, the state has enforced it since November 2025, and an appeal is at the Second Circuit. In the EU, Article 50 transparency duties of the AI Act have applied since 2 August 2026, and marketplaces are adding their own: Zalando requires invisible marking of AI-generated partner imagery by December 2026. The FTC's December 2023 Rite Aid order is the reference point for camera-based identification in stores. Corrected on 25 September 2026: this hub previously carried launch figures and company names we could not trace to any primary source; they have been removed.