COMETAL's Experience: How Neural Networks Replaced Routine in Operations Directorate

COMETAL, a digital platform for industrial cooperation and a subsidiary of Severstal, has implemented several AI tools to automate routine processes. The first was a document comparison service: uploading two versions of a contract, the neural network highlights differences with 95% accuracy, saving lawyers and office managers up to 40 hours per month per department. Then came a tool for cleaning drawings, removing confidential data from frames in seconds, whereas previously it took 15–30 minutes per drawing. Call recording with summaries has been introduced in the CRM, and analytics for script adherence is under development, but progress is slow due to the uniqueness of the products. The neuro-lawyer project (assessing contract risks) faced challenges: the first version did not account for transaction context or the mirroring of terms; the second version is now being configured by an employee who describes risks in plain language. The chatbot 'Arkady' acts as a virtual business analyst: it helps formulate ideas into technical specifications and write regulations, saving analysts' time. The internal platform 'Da Vinci' provides access to several large language models (LLMs) in a secure environment, used by 500 users for 200 solutions. Plans include recognizing primary accounting documents and generating universal transfer documents (UTD), a bot for linking photos to requests, generating specifications, and monitoring counterparties with AI. The company emphasizes a pragmatic approach: if a problem can be solved by an algorithm, they use an algorithm; AI is applied for indirect logic, and human oversight is mandatory.