Banks have seen it all before. When digital communications took hold in the early 2010s, institutions raced to bolt on point solutions: one tool for email, another for SMS, a third for web chat, a fourth for mobile push. Each solved a narrow problem in isolation. A decade later, many banks are still untangling the resulting mess of disconnected systems, inconsistent customer data and integration debt that eats budget faster than it delivers value.

AI is on track to repeat this pattern, and the warning signs are already visible. A typical bank could soon be running as many as 30 AI agents, each operating in silos, with no shared rules for how they interact. Picture a document intake agent, an origination agent and a servicing agent quoting the same borrower different terms, or an unsupervised underwriting agent that's been making poor recommendations for months. This is the direction institutions are heading when AI is procured one point problem at a time.

The financial cost of this approach is already showing up in the numbers. A recent MIT study found 95% of organizations see no measurable return on their generative AI investments, despite banks pouring billions into the technology. Institutions are treating AI as a series of IT projects bolted onto existing workflows rather than as infrastructure that needs to support many workflows over time.

Off the shelf tools solving one narrow use case force banks into an uncomfortable choice: deploy dozens of disconnected point solutions and accept the resulting "agent sprawl," or build custom frameworks demanding constant technical maintenance and rarely translating cleanly from one department to another.

Source link