Illustrative engagement across five service pillars and five industries, reflecting representative Meruksha engagement patterns.
Industry: Regional Banking
Client Profile
The client is a regional bank holding company with approximately $18 billion in assets, operating retail banking, commercial lending, and wealth management divisions across a multi-state footprint. The bank had adopted AI over the prior 18 months for two flagship use cases: real-time fraud detection on card transactions and a customer-facing chatbot handling account inquiries.
Challenge
Over the course of a single year, the bank’s AI-related vendor spend grew by 340%, driven largely by rising token consumption on its customer service chatbot and expanding usage of a premium-tier large language model across both the chatbot and fraud detection systems. The problem was not the growth in usage itself — adoption was genuinely increasing — but that no one in the organization had visibility into which specific use cases were generating that growth or whether the spend was proportionate to the value being delivered. This came to a head when the board was asked to approve renewal of a seven-figure enterprise LLM contract for the following year. The CFO’s office could report total spend but could not answer basic questions about cost per resolved customer inquiry, cost per fraud case caught, or whether a lower-cost model tier would perform comparably for the bank’s actual query mix. Without that visibility, the renewal decision was being made on vendor relationship and inertia rather than evidence.
Approach
Meruksha conducted a six-week AI total cost of ownership (TCO) assessment spanning every active AI workload in the bank, mapping token consumption, model tier assignment, and integration overhead against measurable business outcomes: fraud dollars caught, chatbot call deflection rate, and customer satisfaction scores by interaction type. The analysis segmented chatbot queries by complexity and found that approximately 60% of all customer interactions were routine account questions — balance checks, branch hours, simple transaction lookups — that were being routed through the same premium-tier model used for complex fraud-adjacent conversations, despite a smaller and substantially cheaper model performing identically on that query segment in blind testing. Meruksha designed a tiered routing architecture that matched model capability to query complexity and worked with the bank’s finance and technology teams to stand up an ongoing FinOps-for-AI dashboard, giving cost-per-outcome a permanent home in monthly financial reporting rather than being an annual point-in-time exercise.
Outcome
Model spend was reduced by 34% within a single quarter of implementing the tiered routing architecture, with no measurable degradation in fraud detection accuracy or customer satisfaction scores — the routine-query segment saw identical resolution quality at a fraction of the cost. Armed with this data, the board approved the vendor contract renewal at a lower usage tier reflecting the bank’s actual optimized consumption, rather than the inflated baseline. The FinOps dashboard Meruksha stood up is now a standing agenda item in the CFO’s monthly technology spend review, giving the bank continuous visibility rather than a one-time snapshot.