Finance firms struggle to prove multiagent AI returns, lean on model routing to manage costs

Financial services organizations are racing to adopt multiagent AI systems, but results remain elusive. According to research from the Capgemini Research Institute based on interviews with 1,100 C-suite executives at financial firms, roughly half of those building multiagent systems for core banking and insurance operations struggle to demonstrate consistent returns on their investments.

The gap between adoption and execution is stark. Nearly 60% of surveyed financial institutions have piloted or deployed agentic AI systems, yet only 1 in 10 has managed to orchestrate agents running autonomously across their environments, the report shows.

Cloud foundations and orchestration

Successful multiagent deployments hinge on establishing a robust cloud foundation that acts as the execution layer controlling agent data access, decision-making and actions. According to Ravi Khokhar, global head of cloud for financial services at Capgemini, "As agentic AI scales, those with the strongest foundation will see the true value."

For CIOs navigating this landscape, the priority is clear: establish the right infrastructure through cloud systems and cost controls before scaling multiagent workflows.

Model routing emerges as cost control strategy

Nearly all surveyed financial institutions are either using or planning to implement automated model routing to manage token expenses. This approach directs queries to the most economical AI models capable of delivering adequate output quality. Khokhar noted that financial institutions expect cloud providers to develop capabilities enabling model selection.

Bank of America has deployed Orchestra, an AI routing layer that uses a small classifier model to direct queries to the most cost-effective option. Travelers Insurance has taken a different path, building TravelersLLM, its own large language model designed to handle queries at lower cost than frontier models.

According to Khokhar, "The cloud foundation is seamlessly able to offer that routing without losing the context and without losing the access to the right data. That is what is unlocking what we call return on intelligence."

Hybrid cloud becomes the dominant model

Financial institutions are unlikely to move entirely to cloud-based AI. The survey indicates that more than two-thirds of financial organizations will adopt a hybrid approach combining cloud and on-premises agentic AI capabilities within three years.

Khokhar described this strategy as composable: "A hybrid strategy is a composable ecosystem where you use fit-for-purpose language models, fit-for-purpose compute environments, and fit-for-purpose cloud environments."

Shifting investment priorities

Beyond infrastructure decisions, the rationale for AI spending is undergoing a fundamental shift. Financial institutions are moving away from justifying AI investments based on operational efficiency and toward business outcomes. Capgemini projects that the share of executives prioritizing business outcomes as justification for AI spending will increase by 141% by 2028.