The bank’s interest comes as usage charges become a key concern for AI adoption. Providers charge based on tokens, the small units of information their models handle. Ramamoorthy said powerful international models can be unnecessarily expensive for narrowly defined Indian banking tasks.

He pointed to Sarvam and BharatGen as local alternatives for work such as extracting information from documents and turning Indian speech into text. In his assessment, these models are better suited to such requirements and can cost a tenth as much. However, their capabilities are still developing, and adoption would depend on demonstrated value and privacy protections.

Ramamoorthy also cautioned against expecting AI to quickly resolve the burden of ageing technology. Banks operate under tight regulation, making system changes a gradual exercise involving testing and safety checks. He estimated that meaningful AI-led modernisation of the bank’s technology environment would need another 12–18 months.

AI is increasingly part of board-level planning, with management expected to show business benefits. Across banking, financial services and insurance, current applications largely support risk controls and internal operations. Customer-facing applications could follow, initially with human assistance, while trust and the RBI’s expectations around model governance remain important.

YES Bank recently organised an AI Fest to help product managers and business teams explore possible applications. The aim was to encourage departments to develop ideas together around actual business needs.