Last reviewed
May 30, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v2 · 3,764 words
Add missing citations, update stale details, or suggest a clearer explanation.
AI in retail and e-commerce refers to the use of artificial intelligence techniques across the retail value chain, from how products are recommended and searched for to how they are priced, stocked, delivered, and sold. Retail was one of the earliest commercial proving grounds for machine learning: the product recommendation engines built by online stores in the late 1990s and early 2000s are among the most widely deployed AI systems ever created. Since the early 2020s the field has been reshaped a second time by generative AI and large language models, which power a new generation of conversational shopping assistants, automated content tools, and so called agentic commerce, in which software agents help complete purchases on a shopper's behalf.
This article surveys the main application areas, the underlying techniques, notable corporate deployments, and the principal benefits, risks, and controversies. Many of the adoption figures, revenue estimates, and effectiveness claims in this area originate with the companies selling the technology or with market-research firms, and they are presented here as attributed claims rather than independently verified facts.
The commercial history of AI in retail begins with recommender systems. In the 1990s researchers developed collaborative filtering, a family of methods that predict what a person will like by analysing the behaviour of many users with overlapping tastes. Amazon.com became an early and influential adopter. Engineers at the company developed an approach they called item-to-item collaborative filtering, which scales recommendations by precomputing similarities between products rather than between users, and described it in a 2003 paper in the journal IEEE Internet Computing.1 The familiar "customers who bought this also bought" feature popularised the idea that a store could be personalised for every visitor.
The Netflix Prize, an open competition that ran from 2006 to 2009, brought further attention and rigour to the field. Netflix offered one million US dollars to the first team that could improve the accuracy of its film-rating predictions by ten percent, measured by root mean squared error, and the grand prize was awarded in 2009 to a team called BellKor's Pragmatic Chaos.2 Although the contest concerned media rather than physical goods, it popularised benchmark datasets, matrix-factorisation techniques, and model ensembles that fed directly into retail recommendation work.
Over the following decade recommendation moved from explicit ratings to implicit signals such as clicks, dwell time, and purchase history, and from simple matrix methods to deep neural networks that learn embeddings for users and items. These systems spread from the product page to search ranking, marketing email, home-page merchandising, and advertising.
The second major shift arrived with generative AI. After the public release of ChatGPT in late 2022, retailers and platforms raced to build natural-language tools. An early bridge between consumer chatbots and shopping was the ChatGPT plugin program, in which OpenAI partnered with retail-adjacent companies including Klarna, Shopify, and Instacart starting in March 2023; that program and its shopping listings are covered in a separate article.3 Within roughly two years the major retail platforms had launched their own assistants, and by 2025 several companies were experimenting with agentic commerce, in which an AI agent can search a catalogue and, in some implementations, place an order. The structure, capabilities, and limits of these systems are described in the application sections below.
AI now touches nearly every retail function. The table below summarises the principal areas; the most significant are then discussed in turn.
| Application area | Typical AI techniques | Examples |
|---|---|---|
| Product recommendation and personalisation | Collaborative filtering, deep learning, embeddings | "Frequently bought together", personalised home pages |
| Search and visual search | Information retrieval, computer vision, CNNs | Amazon StyleSnap, Pinterest Lens, Google Lens |
| Conversational shopping assistants | Large language models, retrieval-augmented generation | Amazon Rufus, Shopify Sidekick, ChatGPT shopping |
| Demand forecasting and inventory | Time-series and neural forecasting | Walmart demand models |
| Dynamic and personalised pricing | Optimisation, reinforcement learning | Algorithmic repricing tools |
| Supply chain and logistics | Forecasting, optimisation, routing | Network planning and route optimisation |
| Fraud detection | Anomaly detection, classification | Real-time payment screening |
| Customer service | Chatbots, LLMs | Klarna assistant, merchant support bots |
| Marketing and ad targeting | Ranking, generative content | Retail media, generative creative |
| Virtual try-on | Generative image models, computer vision | Google try-on, Walmart Be Your Own Model |
| Cashierless checkout | Sensor fusion, computer vision | Amazon Just Walk Out |
| Warehouse robotics | Computer vision, motion planning | Amazon Proteus, Sparrow, Sequoia |
Recommendation remains the most pervasive use of AI in retail. Systems suggest products on the home page, the product detail page, in search results, and in marketing channels, drawing on a shopper's history and on patterns across the wider user base. Modern engines typically combine collaborative signals with content features such as category, brand, and price, and increasingly use deep-learning embeddings that map users and items into a shared vector space so that similarity can be computed efficiently at scale.
Text search ranking has long used machine learning, and a growing share of retail search is now visual. Visual search lets a shopper submit a photograph instead of keywords; a computer vision model, usually built on convolutional neural networks, extracts features such as shape, colour, and texture and matches them against an indexed catalogue. Pinterest introduced its Lens visual-search tool in 2017, and Amazon launched StyleSnap, a fashion-focused visual search feature, in 2019, later extending it to home furnishings.4 Google offers comparable capabilities through Google Lens. Generative AI is now also being applied to search itself, for example through conversational "AI mode" search experiences that summarise and compare products.
The most visible recent development is the conversational assistant embedded in a store or platform. These tools use large language models, frequently combined with retrieval over the retailer's own catalogue and reviews, to answer open-ended questions, compare products, and guide discovery.
Amazon announced Rufus, which it described as "a generative AI-powered expert shopping assistant trained on Amazon's extensive product catalog, customer reviews, community Q&As, and information from across the web," on 1 February 2024, rolling it out first to a subset of US mobile-app users.5 Amazon extended Rufus to the United Kingdom and several European markets later in 2024.6 On the company's third-quarter 2025 earnings call, chief executive Andy Jassy said Rufus was "expected to generate over $10 billion in annual incremental sales" and reported that 250 million shoppers had used it that year, with monthly active users up about 140 percent year over year; Amazon said customers who engage with Rufus are 60 percent more likely to complete a purchase, a figure measured with the company's own attribution model.7 In May 2026 Amazon rebranded Rufus as Alexa for Shopping.5
Shopify offers a merchant-facing assistant called Sidekick inside its administrative console. Rather than helping consumers shop, Sidekick lets store operators use natural language to query their own data, generate product descriptions and marketing content, build discounts, and edit themes; Shopify has described it as a commerce assistant designed to help merchants start, run, and grow a business.8 Shopify announced Sidekick in 2023 and broadened its availability and capabilities over the following years.8
OpenAI moved shopping directly into ChatGPT. In late September 2025 the company announced a feature it called Instant Checkout, built with the payments firm Stripe, that let US users buy from participating Etsy sellers within the chat interface, with Shopify merchants to follow; alongside it OpenAI published an open Agentic Commerce Protocol intended to let AI agents and merchants transact.9 In October 2025 PayPal said it would adopt the protocol to support payments and commerce in ChatGPT.10 Trade-press reporting in 2026 indicated that OpenAI subsequently de-emphasised native in-chat checkout in favour of a model in which the assistant handles product discovery while merchants retain control of the actual checkout.11 Chinese platforms pursued similar agentic features: Alibaba integrated its Qwen models with the Taobao and Tmall marketplaces, and at its 11.11 shopping festival deployed AI-enhanced listings, search, and a shopping assistant; the company describes these as large-scale applications across its catalogue, claims that originate with Alibaba.12
Forecasting how much of each product will sell, in each location and time period, is a classic machine-learning problem in retail. Accurate demand forecasting and inventory planning reduce both stockouts and overstock. Walmart has said it uses internally built neural forecasting models, including a multi-horizon recurrent neural network, to predict demand across its network and to plan inventory placement, incorporating signals such as location, weather, local events, and seasonality; a senior executive said the company can plan inventory levels "more accurately" with these models.13 Walmart has also described generative-AI merchant tools, including an assistant it calls Wally, used to combine sales and inventory data and surface underperforming or overperforming products.14 These descriptions and any associated benefits are Walmart's own claims.
Many retailers adjust prices algorithmically in response to demand, competitor prices, inventory, and time, a practice known as dynamic pricing. A more contentious variant is personalised or individualised pricing, in which prices vary by inferred characteristics of the shopper. In 2024 the US Federal Trade Commission used its information-gathering powers to study what it termed "surveillance pricing," examining intermediaries that retailers hire to target prices; in a January 2025 staff report the agency said that signals as granular as mouse movements and the contents of an abandoned cart could be used to tailor offers, and that the firms it examined collectively served hundreds of retail clients.15 The risks and policy debate around these practices are discussed below.
Beyond store-level inventory, AI is used to plan the wider supply chain: positioning stock across distribution centres, routing delivery vehicles, and anticipating disruptions. Walmart has described using neural forecasting for network planning, adaptive search algorithms to optimise delivery routes, and computer vision to flag quality issues such as damaged produce, and has said agentic tools give it unified visibility across stores and fulfilment centres; one executive characterised the operation as having "every segment" driven by "some form of intelligence."13 As with other vendor descriptions, these claims come from the company.
Machine learning is widely used to detect payment fraud, account takeover, and abuse of returns and refunds. Anomaly-detection and classification models score transactions in real time against large numbers of behavioural and contextual features, aiming to block fraudulent orders while minimising false declines of legitimate customers. Industry coverage reports that a majority of retailers now use AI in fraud prevention, although specific performance figures generally come from vendors.16
Retail customer service has been an early adopter of chatbots and, more recently, LLM-based assistants that handle returns, order status, and product questions. Vendor and operator claims about deflection rates and cost savings in this area should be treated as claims; they are frequently cited but rarely independently audited.
AI ranks and targets advertisements, the foundation of the fast-growing "retail media" businesses that platforms such as Amazon and Walmart operate. Generative AI is increasingly used to produce marketing copy and product imagery at scale. Reporting on Amazon's Rufus noted internal projections, attributed to a Business Insider account of company documents, that the assistant could contribute substantial operating profit partly through advertisements embedded in its responses; these are projections rather than reported results.7
Generative image models have made virtual try-on, in which a shopper sees how a garment might look on their own body, more realistic. Google rolled out an AI virtual try-on feature across Search, Shopping, and Images, powered by a custom image-generation model for fashion that it says understands how materials fold, stretch, and drape, and in 2025 introduced a standalone app called Doppl; Google has said the feature works with items from retailers including Macy's, Kohl's, Walmart, and Nordstrom.17 Walmart introduced a tool it calls Be Your Own Model, letting shoppers upload a full-body image to preview apparel, building on its 2021 acquisition of the virtual-fitting-room startup Zeekit.18
Amazon pioneered large-scale cashierless retail with its Just Walk Out technology, which uses ceiling cameras, shelf sensors, and sensor fusion to detect what shoppers take and charge them automatically. Amazon revealed the concept and opened the first Amazon Go store to employees on 5 December 2016, with a public opening on 22 January 2018 in Seattle; by March 2024 it operated dozens of Go and Amazon Fresh locations.19 In April 2024 Amazon said it would remove Just Walk Out from its US Amazon Fresh grocery stores in favour of Dash Cart smart trolleys, a change first reported by The Information, which also reported that the system relied on a behind-the-scenes workforce of more than a thousand workers in India to review and annotate video.2021 Amazon disputed characterisations of the technology as primarily human-operated, saying the India team mostly labelled data to train and validate the system after the fact.22 Amazon continued to offer Just Walk Out for other venues and to third parties. The controversy is discussed further below.
AI-driven robots increasingly handle storage, picking, and sorting in fulfilment centres. Amazon has deployed several systems, including Proteus, which it describes as its first fully autonomous mobile robot; Sparrow, a robotic arm that uses computer vision and suction to pick individual items; and Sequoia, a storage and retrieval system, all announced in 2023.23 In 2025 Amazon said it had surpassed one million robots deployed across its operations, a milestone reported in trade and general press.24
The applications above draw on a common toolkit. Recommender systems supply personalisation and merchandising. Computer vision underpins visual search, virtual try-on, cashierless checkout, and warehouse picking. Time-series and neural forecasting models drive demand prediction and inventory planning, while optimisation and, in some systems, reinforcement learning inform pricing and logistics. Anomaly detection and classification power fraud screening. The most recent wave rests on large language models and generative AI, usually paired with retrieval over a retailer's own product and review data so that answers stay grounded in the live catalogue; this retrieval-augmented pattern is central to conversational assistants and to agentic commerce.
The table below lists prominent systems referenced in this article. Capability and impact figures are the operators' own claims unless otherwise noted.
| Company | System | Role | First announced |
|---|---|---|---|
| Amazon | Item-to-item collaborative filtering | Product recommendation | Described 2003 |
| Amazon | StyleSnap | Visual search | 2019 |
| Amazon | Just Walk Out / Amazon Go | Cashierless checkout | 2016 (employee), 2018 (public) |
| Amazon | Proteus, Sparrow, Sequoia | Warehouse robotics | 2023 |
| Amazon | Rufus (later Alexa for Shopping) | Conversational shopping assistant | February 2024 |
| Walmart | Internal neural forecasting; Wally | Demand and merchandising | Ongoing |
| Walmart | Be Your Own Model | Virtual try-on | 2023 onward |
| Shopify | Sidekick | Merchant assistant | 2023 |
| Alibaba | Qwen with Taobao and Tmall | Agentic commerce | 2024-2025 |
| OpenAI | Instant Checkout and Agentic Commerce Protocol | Agentic commerce in ChatGPT | September 2025 |
Amazon, the largest Western online retailer, has applied AI across recommendation, search, logistics, robotics, cashierless retail, and a consumer shopping assistant. Walmart emphasises supply-chain forecasting, inventory, and merchant tooling, and has stated that its global supply-chain systems are being reengineered around AI. Alibaba's Taobao and Tmall marketplaces have rolled out generative tools for merchants and shoppers and integrated the company's Qwen models for agentic shopping. Shopify provides AI tooling to a large base of independent merchants through Sidekick and related features. Each company's specific performance claims should be read as such.
Proponents argue that AI improves the shopping experience and retail economics in several ways. Personalisation and recommendation can help shoppers find relevant products faster and can lift conversion and order value. Better demand forecasting reduces both stockouts and excess inventory, lowering waste and freeing working capital. Automation in warehouses and checkout can reduce labour costs and speed fulfilment. Fraud models can cut losses while limiting friction for legitimate buyers. Conversational assistants and visual search can make discovery more intuitive, particularly for complex or visually driven categories such as fashion and home furnishings. Many quantified versions of these benefits, however, come from vendors, operators, or market-research firms; for example McKinsey has estimated that generative AI could add several hundred billion dollars of annual value across retail and consumer goods, a projection rather than a measured outcome.25
Algorithmic and especially personalised pricing has drawn regulatory and public scrutiny over concerns that it can be opaque, can charge different people different prices for the same item, and can disadvantage some consumers. The FTC's 2024 surveillance-pricing study and its January 2025 staff findings highlighted how granular behavioural and personal data can feed pricing, and some US states have since opened their own inquiries.15 In a separate but related matter, in 2024 the US Department of Justice and several states sued the software firm RealPage over an algorithmic pricing product used by landlords, a case often cited in debates over algorithmic price-setting.15
The data that powers personalisation, targeting, and pricing also raises privacy concerns. Cashierless stores in particular rely on pervasive cameras and sensors, and the FTC's pricing work underscored how widely behavioural data is collected and shared among intermediaries.15 Critics argue that the granularity of tracking required for some AI retail features sits uneasily with consumer expectations of privacy.
Like other machine-learning applications, retail AI can reflect and amplify biases in its training data. Recommendation and search ranking can entrench popularity effects, and pricing or targeting models trained on historical data risk encoding demographic disparities, which is part of why personalised pricing in particular has attracted regulatory attention.
Automation in warehouses, checkout, and customer service raises questions about employment. Amazon's accumulation of more than a million robots and its construction of highly automated fulfilment centres have prompted debate, reflected in business and labour coverage, about how warehouse work will change.24 Cashierless checkout similarly bears on retail and cashier roles. Forecasts of net job effects vary widely and are contested.
Amazon's Just Walk Out became a focal point for skepticism about "AI" claims in retail. After The Information reported in 2024 that the system depended on a large team of reviewers in India, several outlets framed the technology as less autonomous than marketing had implied; Amazon countered that the India team chiefly annotated video after the fact to improve the model rather than approving each transaction in real time.2022 The episode is frequently cited as a cautionary example of the gap that can exist between automation marketing and operational reality, and of the human labour that often underlies systems presented as fully automated.
The near-term trajectory of AI in retail centres on generative and agentic systems. Major platforms are racing to embed conversational assistants and, in some cases, to let agents complete purchases, even as the precise division of labour between AI-driven discovery and merchant-controlled checkout is still being worked out, as OpenAI's 2025 to 2026 shifts illustrate.911 At the same time, regulators in the United States and elsewhere are scrutinising algorithmic and personalised pricing, the data practices behind targeting, and the labour and privacy implications of automation. The technology's commercial momentum is substantial, but many of its most striking claims remain vendor projections, and its societal effects, on prices, privacy, and jobs, are still being contested.
See also: Shopping ChatGPT Plugins
Linden, G., Smith, B., and York, J. "Amazon.com Recommendations: Item-to-Item Collaborative Filtering." IEEE Internet Computing, 2003. https://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf Accessed 2026-05-31. ↩
"Netflix Prize." Wikipedia. https://en.wikipedia.org/wiki/Netflix_Prize Accessed 2026-05-31. ↩
"ChatGPT plugins." OpenAI, 23 March 2023. https://openai.com/index/chatgpt-plugins/ Accessed 2026-05-31. ↩
"Visual Search (StyleSnap)." Amazon. https://www.amazon.com/visual-search/help/stylesnap Accessed 2026-05-31. ↩
"Amazon announces Rufus, a new generative AI-powered conversational shopping assistant." About Amazon, 1 February 2024. https://www.aboutamazon.com/news/retail/amazon-rufus Accessed 2026-05-31. ↩ ↩2
"Amazon announces the launch of Rufus ... in beta across Europe." About Amazon EU, 2024. https://www.aboutamazon.eu/news/retail/amazon-announces-the-launch-of-rufus-a-new-generative-ai-powered-conversational-shopping-assistant-in-beta-across-europe Accessed 2026-05-31. ↩
"Amazon says its AI shopping assistant Rufus is so effective it's on pace to pull in an extra $10 billion in sales." Fortune, 2 November 2025. https://fortune.com/2025/11/02/amazon-rufus-ai-shopping-assistant-chatbot-10-billion-sales-monetization/ Accessed 2026-05-31. ↩ ↩2
"Sidekick." Shopify. https://www.shopify.com/sidekick Accessed 2026-05-31. ↩ ↩2
"Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol." OpenAI, September 2025. https://openai.com/index/buy-it-in-chatgpt/ Accessed 2026-05-31. ↩ ↩2
"OpenAI and PayPal Team Up to Power Instant Checkout and Agentic Commerce in ChatGPT." PayPal Newsroom, 28 October 2025. https://newsroom.paypal-corp.com/2025-10-28-OpenAI-and-PayPal-Team-Up-to-Power-Instant-Checkout-and-Agentic-Commerce-in-ChatGPT Accessed 2026-05-31. ↩
"OpenAI shifts checkout plans in its agentic commerce strategy." Digital Commerce 360, 6 March 2026. https://www.digitalcommerce360.com/2026/03/06/openai-shifts-checkout-plans-agentic-commerce-strategy/ Accessed 2026-05-31. ↩ ↩2
"AI Powers Large-scale Applications at the World's Largest Shopping Festival." Alizila (Alibaba), 2024. https://www.alizila.com/ai-powers-1111/ Accessed 2026-05-31. ↩
"4 ways Walmart is scaling AI to unify its supply chain." Supply Chain Dive, 7 October 2025. https://www.supplychaindive.com/news/4-walmart-supply-chain-ai-uses/760891/ Accessed 2026-05-31. ↩ ↩2
"Walmart's U.S. Supply Chain Playbook Goes Global." Walmart Corporate, 17 July 2025. https://corporate.walmart.com/news/2025/07/17/walmarts-us-supply-chain-playbook-goes-global-and-its-reinventing-retail-at-scale Accessed 2026-05-31. ↩
"FTC Surveillance Pricing Study Indicates Wide Range of Personal Data Used to Set Individualized Consumer Prices." US Federal Trade Commission, January 2025. https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-surveillance-pricing-study-indicates-wide-range-personal-data-used-set-individualized Accessed 2026-05-31. ↩ ↩2 ↩3 ↩4
"More Than Half of Retailers Now Use AI to Catch Fraudsters." PYMNTS, 2025. https://www.pymnts.com/news/security-and-risk/2025/more-than-half-of-retailers-now-use-ai-to-catch-fraudsters Accessed 2026-05-31. ↩
"Shopping on Google: AI Mode and virtual try-on updates from I/O 2025." Google blog, 2025. https://blog.google/products-and-platforms/products/shopping/google-shopping-ai-mode-virtual-try-on-update/ Accessed 2026-05-31. ↩
"Google brings AI updates to its virtual try-on tool." Digital Commerce 360, 24 July 2025. https://www.digitalcommerce360.com/2025/07/24/google-brings-ai-updates-to-its-virtual-try-on-tool/ Accessed 2026-05-31. ↩
"Amazon Go." Wikipedia. https://en.wikipedia.org/wiki/Amazon_Go Accessed 2026-05-31. ↩
"Amazon says it is removing Just Walk Out technology from its Fresh grocery stores." NPR, 3 April 2024. https://www.npr.org/2024/04/03/1242448552/amazon-remove-walk-out-technology Accessed 2026-05-31. ↩ ↩2
"Amazon's Just Walk Out stores relied on '1,000 people in India watching,' not AI." The Washington Times, 4 April 2024. https://www.washingtontimes.com/news/2024/apr/4/amazons-just-walk-out-stores-relied-on-1000-people/ Accessed 2026-05-31. ↩
"Amazon to expand third-party Just Walk Out stores, rejects surveillance claims." Axios, 17 April 2024. https://www.axios.com/2024/04/17/amazon-walk-out-store-technology-grocery-expansion Accessed 2026-05-31. ↩ ↩2
"Amazon announces new fulfillment center robots, Sequoia and Digit." About Amazon, 2023. https://www.aboutamazon.com/news/operations/amazon-introduces-new-robotics-solutions Accessed 2026-05-31. ↩
"Amazon tops 1 million robots: Here's what they do." GeekWire, 2025. https://www.geekwire.com/2025/amazons-robot-workforce-hits-1-million-heres-what-they-all-do/ Accessed 2026-05-31. ↩ ↩2
"Generative AI in retail: LLM to ROI." McKinsey & Company. https://www.mckinsey.com/industries/retail/our-insights/llm-to-roi-how-to-scale-gen-ai-in-retail Accessed 2026-05-31. ↩