Nornickel saves 2000 hours monthly: How AI Agents Transformed Budget Control

Each month, Nornickel employees spent up to 2000 working hours on routine financial document checks. This massive volume of work had to be performed manually to ensure the accuracy of budget and project control. After implementing AI agents, these 2000 hours were freed up, and the accuracy of risk detection tripled.

In a large corporation like Nornickel, budget and project control is not just a formality but a critically important element of financial stability. But when hundreds of parameters have to be checked manually across thousands of documents, errors, delays, and missed risks are inevitable. These are not just lost hours; they are potential millions in losses. This used to be considered a necessary evil; today, it's a task for an AI agent.

The Complexity of Budget Control at Nornickel's Scale

Nornickel, one of the world's largest metal producers, operates with vast budgets and numerous projects. Controlling expenditures is a multi-layered process that requires checking every document against dozens, if not hundreds, of parameters. Employees had to manually cross-reference data from contracts, purchase orders, repair orders, memos, minutes, and corporate governance decisions.

Such work not only consumed enormous amounts of time but also carried a high probability of human error. With thousands of documents, even an experienced specialist could miss a critical detail, leading to financial risks, project delays, and reduced overall efficiency. Each check took an average of 15-20 minutes, accumulating into thousands of hours monthly.

Why Existing Systems Were Insufficient

Nornickel already utilized various systems for document management and financial accounting. However, these were designed for recording and storing information, not for intelligent analysis. These systems could show where a document was located but could not independently analyze its content for compliance with budget rules, identify discrepancies, or potential risks. Manual checking remained the only way to ensure an adequate level of control.

The need was not for another database, but for a tool that could "read" and "understand" documents, compare them against regulations, and identify anomalies. This is why the company turned to the concept of AI agents, capable of performing cognitive tasks.

How AI Agents for Financial Control Were Designed

To solve the problem, it was decided to use several AI agents, each specializing in a specific aspect of control. The main idea was to create "virtual auditors" capable of analyzing textual information and numerical data from various sources.

• Budget Compliance Agent. Its task was to check each document (contract, order) for compliance with the approved budget for a project or expenditure item. It compared amounts, deadlines, and nomenclature against planned indicators, identifying excesses or discrepancies.
• Project Control Agent. This agent focused on adhering to the regulations and conditions of specific projects. It analyzed memos, minutes, and decisions to ensure that all actions complied with the approved plan and did not create risks for timelines and costs.
• Risk Detection Agent. The most complex agent, which not only recorded deviations but also looked for indirect signs of potential risks: unusual wording in contracts, deviations from standard procurement procedures, untimely submission of documents. It aggregated information from the other two agents and used it to build a comprehensive picture.

All agents worked in conjunction, exchanging data and forming a unified report for the economic department employees. Human intervention was only required for decision-making on identified anomalies.

Implementation and Adaptation

Implementation proceeded in stages, starting with the least critical but most labor-intensive processes. Budget compliance checks, where rules were clearest, were automated first. This quickly demonstrated the value of AI agents and garnered support from employees. Gradually, functionality expanded, agents were trained on real data, and adapted to the specifics of Nornickel's internal regulations.

An important aspect was employee training: they were not just taught how to use the new system but also explained how AI agents helped them in their work, freeing them from routine and allowing them to focus on more complex analytical tasks. This contributed to the rapid adoption of the new tools.

Achieved Results

Freeing up 2000 working hours per month is equivalent to the work of over 12 full-time employees. These resources were redirected to more strategic tasks requiring deep analysis and management decision-making. The significant increase in risk detection accuracy allowed Nornickel to prevent potential financial losses and optimize spending.

How to Implement This in Your Company

Nornickel's case demonstrates that even in the most complex and regulated processes, such as budget control, AI agents can bring immense benefits. If you have extensive manual checks and approvals, start by:

• Identifying the most labor-intensive routine checks. Gather data on how much time employees spend on routine comparisons and document checks.
• Formalizing control rules and parameters. Clearly define the criteria for checks; this is the basis for training the AI agent.
• Phased implementation. Start with one, simpler, and high-volume process to quickly achieve initial results and scale the solution based on them.
• Training and engaging the team. Show employees how AI agents free them from routine, making their work more interesting and productive.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager