Data Power has developed an AI agent for marketplace sellers that automatically searches for unnoticed profit losses. According to the company, a large seller can lose from 1.5 million rubles per week due to calculation errors and analytical 'blind spots'.
The system connects to the personal account via API and analyzes sales, advertising, and inventory. For example, it identifies campaigns where promotion costs have already exceeded the margin, records a drop in card conversion, and searches for products on which advertising budgets continue to be spent without noticeable returns.
Interestingly, the AI itself is hardly involved in the calculations here. Mathematics and anomaly detection are performed by ordinary program code according to predefined rules, and the neural network receives ready-made results and translates them into understandable conclusions and recommendations for the manager. This way, the developers try to avoid situations where the model 'makes up' financial indicators itself.
According to Data Power, a manual search for such deviations can take an analyst about eight hours. The new agent reduces this process to approximately 15 minutes, but the final decision still rests with a human.
The solution is primarily designed for medium and large sellers with a turnover of 20 million rubles per month or more. Data from marketplace accounts is anonymized and consolidated into a single database on Russian servers.