The Challenge
Self-service gaps and escalation overload
A top-tier U.S. credit union with millions of members had inherited a chatbot that couldn’t carry its own weight. Built by a prior team and never released to production, the system had no integration with core banking, limited self-service capability, and ADA compliance gaps, which meant routine inquiries like balances, transactions, and FAQ-level questions kept flowing to live agents. Operational costs climbed, member experience suffered, and the institution remained heavily dependent on a third-party live-chat platform to fill the gap. Compounding all of this, an aging codebase and strict compliance reviews slowed every change, and the kinds of AI capabilities the team wanted to introduce had to clear regulatory approval before they could go anywhere near member data.
The approach
Production-grade NLU, secure integration, and iterative deployment
We rebuilt the assistant as a production-grade conversational AI system across web and mobile, designed to do the work the prior version couldn’t.
The foundation was the natural language layer. We tightened intent recognition and expanded dialogue coverage using conversation language understanding and custom question-answering models, so the assistant could resolve the categories of questions that had been escalating by default. With the NLU layer working reliably, we added secure authentication, which opened the door to authenticated data flows like account balances, recent transactions, and account-level guidance that the previous version had never been cleared to handle. From there, the system grew: FAQ handling through the question-answering module, multi-step task guidance through intent-based dialogue flows, and a clean handoff to a live representative reserved for cases where the member’s need actually required it.
Connecting a conversational AI system to core banking in a regulated institution meant coordinating across multiple internal teams, working within the compliance perimeter, and engineering each capability so it could withstand audit before it ever reached a member. We worked in tight iteration cycles with the client’s engineering and member-services leaders, deploying capability incrementally as each piece cleared review.
The outcome
Live across channels with growing self-service adoption
The assistant is live in production across web and mobile, a milestone the prior version never reached. Members now get authenticated self-service for the questions they were previously waiting on agents to answer, which means live representatives are increasingly free to focus on the complex cases where human judgment actually matters, and the institution’s reliance on its third-party live-chat platform has steadily decreased as the assistant’s dialogue coverage has grown. The platform is structured for continued capability expansion, with new conversational AI features rolling out as they clear regulatory review.
About the Company
Top Credit Union
Large U.S. financial institution on a mission to modernize their technology and update the customer experience.
