According to Reuters, Denis Roux, chief information officer, investment bank at Deutsche Bank, said "We're seeing things that were two years that are now getting done in three to six months... we know the productivity is there." Reuters reports Roux added that backlogs that once took months are now being cleared in weeks and that he declined to quantify the impact. Reuters also reports the bank has around 9,000 employees in its technology function in India, accounting for about 45% of its global tech workforce. Per Reuters, Roux warned about rising computing costs as AI providers such as Anthropic and OpenAI shift toward token-based pricing, noting engineers are allocated token quotas and must demonstrate value to request additional capacity.
What happened
According to Reuters, Denis Roux, chief information officer, investment bank at Deutsche Bank, said "We're seeing things that were two years that are now getting done in three to six months... we know the productivity is there," speaking on the sidelines of the bank's Bank on Tech event in Bengaluru, India. Reuters reports Roux added that backlogs that once took months are now being cleared in weeks, and that he declined to quantify the impact. Reuters also reports the bank has around 9,000 employees in its technology function in India, accounting for about 45% of its global tech workforce. Reuters notes rising compute costs as AI providers such as Anthropic and OpenAI shift toward token-based pricing, and reports engineers at the bank are allocated token quotas and must demonstrate value to request additional capacity.
Technical details
Editorial analysis - technical context: Token-based pricing changes the unit economics of deploying large language models and generative services. Companies adopting cloud-hosted LLM APIs face two linked technical demands: cost-aware model selection and tighter usage governance. For engineering teams, that usually means more emphasis on prompt engineering, batching, caching, and mixing smaller local models with larger remote models to control per-token spend.
Companies adopting LLMs often implement three practical controls:
- •quota and approval workflows that gate high-cost experiments, mirroring the token quotas Reuters reports;
- •automated cost attribution and observability tied to CI/CD and MLOps pipelines, to correlate spend with measurable outcomes;
- •hybrid inference strategies (local distilled models for routine tasks, API calls for high-value generations) to reduce token volume.
Context and significance
Industry context: Enterprises across finance and other regulated sectors are pursuing productivity gains from generative AI while wrestling with opaque and usage-driven pricing. Reuters further observes that global firms are increasingly using Indian hubs for higher-value functions, a trend that frames the operational stakes for large outsourcers and captive engineering centers. The combination of accelerated delivery timelines and rising compute budgets raises practical trade-offs between speed, reproducibility, and cost control for large engineering organisations.
What to watch
For practitioners: monitor three indicators that will determine whether AI-driven acceleration is sustainable:
- •changes in total compute spend and cost per delivered feature
- •how organisations measure "value" when approving extra token allocations
- •whether observability and governance tooling keeps pace with rapid iteration. Observers should also watch whether teams invest in model-evaluation metrics tied to business outcomes, and whether hybrid deployment patterns (distilled local models plus API-first large models) become standard practice for cost-sensitive workflows
Scoring Rationale #
The report documents meaningful productivity improvement at a major bank and highlights a broader infrastructure challenge: token-priced models change operational economics. This matters to practitioners balancing speed, cost, and governance.
Practice with real FinTech & Trading data
90 SQL & Python problems · 15 industry datasets
[Active Verified Users by Income TierEasy](/problems/sql/active-verified-users-by-income)
[Technology Stocks with High BetaMedium](/problems/sql/technology-stocks-with-high-beta)
[Portfolio Performance ScorecardHard](/problems/sql/portfolio-performance-scorecard)
250 free problems · No credit card
See all FinTech & Trading problems