PKO Bank Polski built its own AI voice assistants instead of outsourcing them

What happened

PKO Bank Polski built a fleet of voicebots and a voice assistant for its IKO mobile app using models developed and controlled internally, rather than relying on third-party AI vendors for a regulated, customer-facing function. The bots now handle call-center inquiries, soft debt collection, and customer surveys.

The bank has reported over 50 million customer conversations handled by its AI systems, across 17 conversational bots, with continued expansion planned. Keeping model development in-house gave the bank direct control over what data the models see and how they behave - a materially different risk profile than routing customer conversations through an external AI vendor.

The business problem

The bank needed to scale customer service without a proportional increase in headcount, in a sector where data control and regulatory scrutiny are unusually high.

Why it worked

• In a regulated industry, building and controlling your own models can be worth the extra investment over renting a vendor's
• Scaling AI customer service works when it's rolled out function by function, with room to test before expanding
• Internal AI competency is itself a governance asset, not just an engineering choice
• Measurable, published usage numbers make the case for further investment easier to make internally

Where this points