Sber Radically Reduced Routine Tasks: How AI Agents Save 80% of Employee Time and Accelerate Reporting by 70%

The implementation of AI agents at Sber has shown impressive results: the time spent on information analysis has been reduced by 30-50%, document processing speed has increased by 80%, and reporting preparation has accelerated by 70%. The GigaCowork platform, designed for managing these agents, allows delegating routine business process steps to AI agents, defining their logic using business regulations without developer involvement. This not only saves up to 81.5% of employee working time but also significantly reduces operational costs.

In any large company, especially a bank, routine operations invisibly consume enormous resources. Thousands of hours are spent on searching, verifying, preparing repetitive documents and reports that do not create value but merely keep current processes afloat. These hours, multiplied by the cost of highly qualified specialists, result in colossal losses. This problem has long had a solution, and the only question is who in your niche will implement it first.

The Problem of Routine at Sber's Scale

Sber, like any large financial holding, faces an enormous volume of routine operations. Document processing, responses to internal requests, compliance checks, data collection for reports—each of these tasks, though small individually, collectively consumes a vast amount of time from thousands of employees. These specialists, hired for analytics and client interaction, spend a significant part of their day performing operator functions: searching, copying, and verifying information.

This approach not only slows down processes and exhausts personnel but is also a direct source of errors. In the banking sector, the cost of an error is significantly higher than in most other industries, making the problem of routine critically important to address.

Why Previous Automation Was Insufficient

Sber already had automation systems, but they operated on rigid scripts and scenarios, effectively handling only strictly formalized processes. As soon as a request deviated from the template, for example, a non-standard document or a live employee question, the task would fall back to a human. The need was not just for another automated system, but for a tool capable of understanding natural language, extracting information from various corporate systems, and autonomously performing routine actions without constant intervention.

This is why Sber turned to creating a platform for AI agents: not just another "button," but an intelligent assistant integrated into daily work, taking over routine tasks and allowing employees to focus on tasks requiring human judgment and expertise.

How the GigaCowork Platform Was Designed

The GigaCowork platform was conceived as a unified environment for managing AI agents, capable of delegating routine business process steps to them. It required several key functionalities:

• Understanding requests. Agents needed to understand natural language so employees could assign tasks in free form.
• Integration with corporate systems. The platform had to provide seamless connection to CRM, ERP, email, file storage, and other internal systems via connectors (using the Model Context Protocol, MCP).
• Skill creation without programming. The logic of agent operations had to be described in simple terms in a chat, in the format of regulations, instructions, or checklists, without developer involvement.
• Autonomous operation. Agents were to independently perform tasks such as preparing analytical reports, project calculations, risk assessments, drafting documents, and interacting with each other.
• Human control. The final result is always passed to a human for ultimate decision-making, maintaining expert oversight.
• Scalability. Created skills needed to be available for use by other employees and departments, scaling the effect across the entire company.

The platform includes workspaces for different departments, "skills"—instructions for agents, "connectors" for linking with internal systems, and strict access control to ensure security and transparency.

Implementation and Application of AI Agents at Sber

The implementation of GigaCowork began with testing, allowing the platform to be gradually integrated into workflows. AI agents were deployed to solve specific tasks in various departments:

• Back-office. A task to change payment terms for a contract, which previously took 2-3 hours (contract search, credit policy verification, risk assessment, drafting an addendum), is now completed by an agent in 10 minutes. The employee merely reviews and sends the prepared document.
• Legal department. Agents check contracts for compliance with templates, identify deviations, and prepare summaries of comments. Initial document review, which previously took 1-2 hours, now takes a few minutes. Lawyers get involved at the stage of substantive work with the document.
• Accounting. On schedule, agents collect data from 1C and CRM, reconcile primary documents with contracts, identify discrepancies, and by the beginning of the week, provide a table of problematic points. The financier receives a complete report without spending time on manual information gathering.

This approach allowed employees to quickly see the real value of the platform, which contributed to its widespread adoption within the company.

Implementation Results

The results of GigaCowork's implementation were impressive and far exceeded expectations:

These figures demonstrate that AI agents not only automate individual operations but also fundamentally change work approaches, freeing up significant human resources. Employees can now focus on strategic tasks requiring creative thinking and expert judgment, instead of routine data collection and processing.

How to Implement This in Your Company

Sber's case shows that AI agents can bring tremendous benefits to almost any large company with a high volume of routine operations. To replicate this success, start small:

• Identify the most common routine tasks. These could be information retrieval, data verification, preparing standard responses or documents that dozens of employees perform daily.
• Describe agent logic without programming. Use business regulations, instructions, or checklists to simplify agent creation and configuration as much as possible.
• Integrate agents into existing systems. The less employees have to change their habits and learn new interfaces, the faster the adaptation will be, and the higher the return.
• Maintain human oversight. Agents should handle preparation and preliminary work, but final decisions and verification should remain with a human. This ensures accuracy and security.

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