14% of US retailers use AI. Shoppers move faster.

Retail has a speed problem, and it runs the wrong way. Inside the firm, AI adoption is modest: about 14% of US retail firms use AI, below the national average of 19.8% and far below the information sector at 39.7% (US Census Bureau, May 2026). About 17% expect to use it within six months. Outside the firm, customers have already moved. Between 30% and 45% of US consumers use generative AI for product research and comparison, according to Bain & Company's November 2025 study. Shoppers are adopting AI faster than the stores they buy from.

14% of US retail firms use AI, against 19.8% of all firms (US Census Bureau, May 2026)

Discovery moves to the assistant

The traffic numbers are still small, and they are rising. AI accounts for up to 25% of referral traffic for some retailers, though still less than 1% of total traffic (Bain). The shift changes where a purchase starts. A customer who asks an assistant for the best running shoe under a budget may never see a category page, a banner or a sponsored listing. Retail media built for keyword search has to be rethought for natural-language questions.

Agents are the next step. Worldwide AI spending is forecast at US$2.67 trillion in 2026, and Gartner expects spending on AI agents and assistants to grow from US$29.2 billion in 2026 to US$65.5 billion in 2027. Yet half of consumers are not comfortable letting AI complete a purchase on its own (Bain). And Gartner expects more than 40% of agentic AI projects cancelled by the end of 2027. Agentic commerce is coming, but slower and more carefully than the demos suggest.

Trust is the asset retailers own

One number stands out. Consumers trust a retailer's own on-site agent three times more than a third-party agent (Bain), though Bain expects the gap could narrow. That makes trust the asset a retailer can still own. Bain advises retailers to "dig the customer moat" with exclusive products and loyalty benefits, and to keep ownership of first-party data and checkout wherever possible.

3× more trust in a retailer's own on-site agent than in a third-party agent (Bain & Company, Nov 2025)

The practical question for 2027 is which AI channels to own and which to serve. An on-site assistant, a loyalty app and checkout are owned channels. Third-party agents are channels the retailer serves on someone else's terms. Each needs a decision on product data, pricing data and customer data.

13 offerings face the shopper, 6 work on stock and stores

The Good AI Prism maps 27 AI offerings and two provider records (Bloomreach, Blue Yonder) in Retail & Consumer, each recorded with a dated source from the provider's own pages. They cover the whole shop. Shopping assistants and agents come from Constructor, epoq, Retresco and Zoovu, alongside the assistants retailers and platforms run themselves, such as Amazon's Rufus, Klarna's AI Assistant, Shopify Sidekick and the Zalando Assistant. Product discovery and search come from Algolia, Doofinder, FactFinder and Syte. Price optimisation from 7Learnings, Competera and Quicklizard. Demand forecasting and replenishment from Delicious Data, paretos and RELEX. Product content from Salsify and Lily AI. Personalisation from Hello Retail and Nosto. In the store, Everseen and Trax use computer vision for checkout loss and shelf recognition, and Heuritech forecasts fashion trends.

Nine of the 29 entries come from providers headquartered in Germany, including epoq, FactFinder, paretos, Delicious Data, Retresco, Shopware and Spryker, and 16 from providers headquartered in Europe. The Global Market Study sets retail against ten other sectors.

Behind the shelf: forecasting and replenishment

Not every retail AI faces the customer. Some of the most measurable uses sit behind the shelf. Demand forecasting and replenishment tools predict sales by article and store, set order quantities and plan markdowns for perishable goods. They work on data a retailer already owns: sales history, stock, promotions, weather and local events. Results show up in numbers a category manager tracks every week: forecast accuracy, stock-outs, write-offs and working capital. Product content tools sit in the same quiet zone. They enrich attributes, descriptions and metadata so that search on the own site, and the answers of external assistants, find the right article. For a retailer that wants AI-referred traffic, clean product data is the entry ticket.

Seven questions before you buy

Settle seven questions in writing before a demo. Which outcome should move: conversion, margin, forecast accuracy or cost to serve? What is today's baseline? Where is customer and product data hosted, and does the provider train on it? What is the billing unit: per seat, per resolution, per order or on quote? Which integrations does it need with the shop system, the ERP and the loyalty platform? Who can stop an agent that changes a price or completes a payment? And which result ends the pilot? A provider that answers these on paper is worth a test.

The rules: disclosure now, credit later

Retail AI mostly sits in the minimal-risk zone of the EU AI Act, with three exceptions that carry dates under the amended timetable of Regulation (EU) 2026/1744:

• Since 2 August 2026, people must be told when they interact with an AI system, such as a shopping assistant (Art. 50).
• From 2 December 2026, generative systems already on the market must mark synthetic outputs, including product images, in a machine-readable way.
• From 2 December 2027, creditworthiness assessment of natural persons, for example instalment payments at checkout, is a high-risk use under Annex III.

In the US, Colorado's bill SB 26-189 on AI and California's rules on automated decision-making bring duties from 1 January 2027. Personalised prices are also a consumer-protection topic in several markets; check the disclosure duties where you sell before launch. The article on synthetic faces and voices covers the marking rules in detail.

Measure margin, not only sales

Measured gains exist where AI supports people. In customer support, a generative AI assistant raised issues resolved per hour by 14% on average and by 34% for less experienced agents (NBER Working Paper 31161). Pricing and personalisation need the same discipline. Run AI pricing or personalisation against a holdout of stores, products or customers, so that seasonality and promotions do not inflate the result. Track conversion, gross margin, markdown rate and return rate together. Measure AI-referred traffic and its conversion separately. Watch complaint rate, price-related service contacts and opt-outs as trust signals. And agree in advance which result scales the pilot and which ends it. The retail pricing and personalisation article sets out the method step by step.

In practice

Use case: owned assistants. Amazon's Rufus, Klarna's AI Assistant, Shopify Sidekick and the Zalando Assistant are assistants that retailers and platforms run themselves, and all four sit on the Good AI Prism map. They keep product data, customer context and checkout with the operator, in the channel consumers trust three times more than a third-party agent, according to Bain.

Scenario: a pricing pilot with a holdout. A fashion retailer tests AI markdown pricing in part of its stores and keeps a matched group on the old rules. After one season it compares gross margin, markdown rate, return rate and price-related complaints between both groups. Only a margin gain without a rise in complaints scales the pilot.

What to do Monday

1. Pick one use with a control group. Pricing or personalisation, measured on margin and complaint rate, not only on sales.
2. Decide your channel map. Which AI channels you own, which third-party agents you serve, and on what data terms.
3. Map every customer-facing AI system to its duty: Art. 50 now, machine-readable marking from 2 December 2026, checkout credit from 2 December 2027.
4. Scope agent permissions tightly. Anything that changes a price or completes a payment needs a limit and a log; an AI Governance Setup defines both.
5. Ask for hosting, training use and billing unit in writing before any demo.

This article describes market and regulation. It is not legal advice.