Artificial intelligence is significantly boosting productivity at Deutsche Bank, with work that previously required years now being completed in just a few months, according to a senior executive. Denis Roux, Chief Information Officer for the investment bank at Deutsche Bank, said on the sidelines of the company’s Bank on Tech event in Bengaluru that AI is helping speed up technology initiatives and reduce long-standing internal backlogs. However, he noted that the bank is also closely monitoring the rising cost of computing resources.

Roux explained that timelines have shortened dramatically, with projects that earlier took around two years now being delivered in three to six months. He added that internal backlogs that once took months are now being cleared within weeks, emphasizing the goal of using AI tools to improve efficiency, though he did not provide specific figures on the impact.

Deutsche Bank has around 9,000 technology employees in India, which makes up about 45% of its global tech workforce. The bank, like many multinational firms, is increasingly relying on its Indian operations for higher-value functions such as software development, finance, and research work.

At the same time, Roux highlighted that managing AI-related costs is becoming important as providers shift toward usage-based pricing models. Companies like Anthropic and OpenAI are moving to token-based billing, where customers are charged based on how much they use the service. At Deutsche Bank, engineers are given token usage limits but can request more capacity if they can justify the business value, with insights shared across teams. The bank tracks usage patterns carefully to balance efficiency and cost control.

Deutsche Bank is also building AI systems to automate tasks like financial data extraction and analysis, as well as tools that connect external events—such as geopolitical or market changes—to portfolio exposure. Roux added that the bank is still cautious, using simpler AI models for routine tasks and evaluating where traditional methods may still be more effective.