PKO Bank Polski has rolled out its first in-house AI agent to support the handling of customer correspondence. The system analyses the content of messages, identifies the subject, assigns the enquiry to a category and suggests the next course of action. According to the bank, this could reduce response times by up to 70 per cent. The solution handles a portion of the tens of thousands of emails received each month, but decisions are still made by staff.
This implementation demonstrates that generative AI in banking is moving beyond the experimental stage and into operational processes. PKO BP had previously made knowledge-search and programming-assistance tools available to thousands of staff. By the end of 2026, the bank plans to launch a further 30 agents, including in the areas of complaints handling, cybersecurity, transaction security and after-sales service.
The business case for the project is clear. Automating repetitive tasks can reduce the cost of handling individual cases, minimise classification errors and free up staff to focus on more complex tasks. The European Banking Authority points out that banks are increasing their investment in AI precisely with a view to improving efficiency and reducing costs. At the same time, it warns against model errors, data quality issues, cyber-attacks and operational risks.
It is therefore not only speed that matters, but also control. The ‘human-in-the-loop’ model reduces the risk of automated, erroneous decisions, but does not eliminate it entirely. Implementation will take place shortly after most of the provisions of the EU AI Act come into force on 2 August 2026. For PKO BP, the most important challenge now will be to maintain quality, security and transparency whilst scaling the agents to cover further processes.
