Wildberries: AI in marketplace operations — warehouse machine vision, neural search, and review processing

• Корпоративный блог Wildberries: масштаб платформы, Техношкола, ML Meetup — Habr (блог Wildberries), 2024
• Warehouse robots at Wildberries — TAdviser, 2025-03
• Wildberries начал тестировать роборуки для сортировки товаров на складах — Habr (новости), 2025-03-11
• Автоматизация складов — опыт Wildberries — RoboTrends, 2025
• Wildberries объявила о запуске в Красном Бору нового роботизированного склада — Ведомости. Северо-Запад, 2025-12-25
• Wildberries теперь имеет право самостоятельно распределять товары продавцов по разным складам и регионам — Oborot.ru, 2026
• Wildberries тестирует нейросети, которые помогут покупателю искать и выбирать товары прямо в мобильном приложении — Oborot.ru, n/a
• Wildberries создала ИИ-помощника (ответы продавцов на отзывы) — CNews, 2026-02-20
• Возвраты на Wildberries: как сократить потери (отраслевые оценки доли возвратов) — Uniseller, n/a

Background

Wildberries is Russia's largest marketplace, operating since 2024 as part of the merged RVB company (Wildberries & Russ). According to the company's corporate blog on Habr, the platform processes tens of billions of user events and around 15 million orders per day. To train ML talent, the company opened its own Tech School (data science, analytics) and held its first ML Meetup in September 2024.

In warehouse robotics, however, Wildberries is a latecomer: per industry outlet RoboTrends, the company only got serious about warehouse automation and robotization in 2024, later than many industry leaders. That gives the case a particular dynamic: this is not a twenty-year automation program like Amazon's, but a forced march to close the gap within two-three years — with in-house software development rather than off-the-shelf purchases.

An important framing: Wildberries discloses noticeably less about its AI than banks or Ozon do. The company publishes workstreams, robot specifications, and goals — but almost never the business metrics of impact. So this case includes only what is documented: technical posts in the corporate blog, launch news, and public statements by unit leaders. Anything beyond that we explicitly mark as absent from the public record.

Problem

Two operational challenges at this scale are visible from public materials. First, logistics' dependence on manual labor: sorting millions of variously sized parcels in warehouses. Fifteen million orders a day is a continuous stream of items of every shape and size, each of which must be recognized, gripped, and routed to the right cell; historically hands did this. The company felt the model's fragility literally: in January 2024 a fire destroyed the Shushary warehouse — and with it a substantial share of the Northwestern hub's capacity.

The second challenge is inventory placement: sellers struggle to predict which regional warehouses will see demand for their goods. A placement error means either long, expensive delivery across half the country or stock frozen in an unwanted warehouse.

Industry context raises the stakes: per third-party seller-service estimates, around 30% of marketplace orders are returned on average, and in apparel and footwear returns reach 70%. Every return is another warehouse handling cycle — the first and third pains amplify each other.

Finally, Andrey Ulyanov, head of Wildberries' warehouse automation department, publicly drew the boundary of the possible: full warehouse automation, in his assessment, remains a utopia — the realistic goal is a hybrid in which automation covers about half of the processes. That is unusually sober goal-setting for a market accustomed to promises of 'lights-out warehouses'.

Solution

Four workstreams are publicly confirmed — and for warehousing the technical details are now known.

Warehouses. Since 2024 Wildberries has been building an automation portfolio: goods-to-person shelf robots (the item travels to the person, not vice versa), robotic forklifts for pallet movement, a vertical cross-belt sorter (per RoboTrends, paid back in about 6 months thanks to shipment volumes), and industrial machine-vision robots transferring goods between conveyor lines. In March 2025 the company began testing six-axis robot arms with proprietary AI software: machine vision recognizes each item's boundaries, ranks items by 'best grip' probability, determines the position and number of suction cups, and picks an efficient collision-free trajectory. Per TAdviser, the manipulator's productivity is at least 950 units per hour with a grip success rate above 97%; industrial operation began on an active sorting line at Koledino. 'Our robots are a step toward smarter and more efficient logistics,' says Andrey Fomin, head of the WB Automation robotics department. The next announced step is an autonomous mobile robot for cargo transport.

The Krasny Bor warehouse launch showed the approach scaling (first phase — December 26, 2025, 95,000 sq. m of a planned 154,000): three vertical automatic sorting lines, conveyor systems, and robots for autonomous pallet movement; the complex replaces the burned-down Shushary facility and becomes the hub for the entire Northwestern Federal District.

Placement. From 2026 (an offer-terms change) the company itself redistributes sellers' goods across warehouses and regions based on demand — placement forecasting is lifted off the seller and handled by platform algorithms which, per RoboTrends, factor in sales history and seasonality.

Shopper experience. The company is testing neural search and product selection inside the app and uses AI summarization and filtering of reviews.

Sellers. In February 2026 an AI assistant for replying to reviews launched: the seller configures parameters once (for example, auto-publishing replies only to 4–5-star reviews), then the neural network works on its own — analyzing the product card, the review text, and the seller's past replies, optionally adding recommendations of complementary products with article numbers. If the product card lacks data, the review is left unanswered — a deliberate safeguard against hallucinations.

Result

Wildberries still discloses almost no quantitative business results for its AI systems — the case's key honest limitation. What is published are technical indicators: robot-arm productivity of at least 950 units per hour, a grip success rate above 97% (TAdviser), and roughly 6-month payback for the cross-belt sorter (RoboTrends). The public warehouse-robotics goal voiced by Elena Obraztsova, Wildberries' marketplace automation director, is to radically reduce dependence on warehouse staff within one to three years; her colleague Andrey Ulyanov honestly caps the ambition at a hybrid with about half of the processes automated. The platform's scale (about 15 million orders per day, tens of billions of events per day) is confirmed by the company's corporate blog.

The trajectory, meanwhile, is visible to the naked eye through infrastructure launches: from the first industrial robot tests in 2024 to industrial operation of manipulators at Koledino in 2025 and the launch of an entire robotized hub at Krasny Bor by the end of 2025. Product AI launches (neural search, AI review replies, automatic goods redistribution) run in parallel with the warehouse work.

In our view, the Wildberries case is interesting precisely as a model of 'AI without impact press releases': the company changes its operating model through contract terms and infrastructure rather than marketing numbers. Automatic goods redistribution is the clearest example: the AI system changes the division of responsibility between platform and seller, and its 'metric' is a legal document, not a percentage on a slide. For market analysts it is a reminder: the absence of public metrics does not equal the absence of deployment — but attributing specific improvement percentages to the company without its own publications is off-limits.

Frame the technical numbers correctly too: 950 units per hour and 97% grips are the spec-sheet indicators of one manipulator type on sorting, not 'the efficiency of Wildberries' AI' overall; industry return estimates (30%, up to 70% in apparel) are market context, not company metrics. In our view the case's most transferable element is Ulyanov's sober goal-setting: 'a hybrid with ~50% of processes automated' is a more honest and achievable frame for warehouse robotics than a 'lights-out warehouse' — and it applies to most logistics operators.

Lessons learned

1. Not all market leaders publish AI metrics: for Wildberries, workstreams, robot specs, and goals are documented, but not business effects. Any 'precise numbers' about WB's AI in third-party case studies should be checked against primary sources.
2. WB's most measurable bet is operational: warehouse robotics with spec-sheet indicators (950 units/hour, >97% grip) and an explicit public goal to reduce dependence on staff within 1–3 years.
3. Sober goal-setting beats slogans: 'full warehouse automation is a utopia; a hybrid with ~50% of processes is realistic' (Andrey Ulyanov) is a more honest frame than a 'lights-out warehouse'.
4. A latecomer can move fast: from first robot tests (2024) to industrial manipulator operation and an entire robotized hub in under two years — thanks to in-house AI software.
5. Shifting demand forecasting from the seller to the platform (auto-redistribution across warehouses) shows an AI product changing the rules via contract terms, not UI.
6. Marketplace AI serves three audiences at once: the warehouse (robots), the shopper (neural search, reviews), and the seller (reply assistant) — distinct projects with distinct economics.
7. Safeguards matter more than coverage: WB's AI assistant leaves a review unanswered when the product card lacks data — a deliberate 'better silent than hallucinating' choice.
8. Industry pains (about 30% returns, up to 70% in apparel — third-party estimates) are legitimate context, but attributing specific improvements to the company without its own publications is off-limits.

Frequently asked questions