Ozon expands the use of AI in the moderation of marketplace content and evaluates the effect as a reduction in the cost of manual processes. Andrey Chupeikin, Ozon's Director of platform Solutions Development, said that the models automatically check product cards and moderate user blocks: reviews, questions, answers and comments.
The key financial thesis is related to the dynamics of the cost of manual moderation. According to Chupeikin, if the completely manual approach were maintained, expenses would increase by about 9 billion rubles in two years. The company also records a high level of automation by the end of 2025 and declares a target for 2026: the share of automated moderation has exceeded 96% and should reach 99.5% by the end of 2026.
The second circuit is language localization. Ozon translates content into foreign languages using ML models and indicates that 92% of translations meet internal quality criteria. This is an important signal for foreign economic activity: platforms are investing in cross-border communications and reducing barriers for foreign partners in personal accounts.
A separate operational area is the search and recommendations. The company attributes the increase in orders from the recommendations on the homepage to the introduction of AI, claiming an increase of 5.5 times per year. In 2026, Ozon plans to introduce an AI assistant for selecting and comparing products.
The application side of such automation is important for sellers and logistics operators. Card moderation affects the speed of SKU output, the correctness of attributes, and the display of mandatory labeling and certificate information. An error in an attribute creates a wave of cancellations, refunds, and claims, followed by a load on the warehouse and the last mile. When the platform enhances automatic checks, the business gets a more stable "content contour" and can plan supplies, packaging and bundling more accurately.
The recommendation to foreign trade participants is to synchronize catalogs with the requirements of the sites, store evidence of compliance with certificates and labeling in a structured form, and prepare description templates in the languages of target markets. Then AI moderation will work as a startup accelerator, not as a source of blockages and delays.
