Understanding the challenge
The credit union was protecting members reactively, one incident at a time. Fraud was usually caught after the money moved, at-risk loans surfaced only when a payment was already missed, and the signals that could have warned staff earlier were scattered across the core banking system, the card processor, and spreadsheets. "We found out about problems from our members instead of before them," the credit union's COO said. Leadership wanted AI but had tried a couple of tools that never connected to how the branch and back-office teams actually worked, and in a regulated institution they were wary of anything they could not explain or oversee. So the credit union stayed reactive, and every prevented loss stayed a might-have-been.
Our approach
The credit union did not need another alert nobody had time to read. It needed AI that watches the member signals and acts, within limits staff set. We combined three of our services:
• AI Operations Design at the core: an operating layer that continuously monitors transaction, card, and account signals, flags likely fraud and at-risk loans early, and drafts the next step for a staff member to approve.
• A governed model with clear boundaries and human sign-off, built for a regulated institution, so every action is explainable and a person owns each decision.
• A Dynamics 365 integration so the layer works from one live member record and the whole team sees the same, current picture.
• A paid search program so the credit union could grow membership in its community, not just protect the members it had.
Technical innovation
At the core is an AI layer that operates rather than just answers: it monitors the member signals across systems, detects the patterns that precede fraud or default, and acts within defined boundaries, freezing a suspicious transaction for review or flagging a loan for outreach, always with a staff member's sign-off. Self-calibrating models learn from every confirmed case and every false alarm, so the alerts get sharper each month rather than noisier, and because everything is logged and explainable, the layer fits the oversight a regulated credit union has to maintain.
Outcome
Preventable fraud losses fell by 31% because the layer caught the patterns before the money moved, and at-risk loans surfaced early enough for staff to reach out and help. The team worked from one current member picture in Dynamics 365 instead of chasing scattered data, the paid program grew membership in the community, and every action stayed explainable and staff-approved. "We get ahead of problems now, and we can show exactly why the system did what it did," the COO said.
