Embed

Northgate is a mid-sized retail bank with about 1.4 million customers and a compliance team of sixty. Every new account, and every account that trips a periodic review, passes through a Know Your Customer check. In 2025 that check took a median of 4.2 days, a backlog of eleven thousand cases sat open at any time, and the team was rejecting roughly a fifth of applications on incomplete documents rather than on risk.

On May 24 the head of Operations, the Chief Compliance Officer and the CTO met with two Brainbase engineers and a solutions lead in Northgate's offices. The first day was not about agents. It was about the flow: who touches a case, in what order, which systems they open, and where the time actually goes. By the end of it the team had walked the floor with three analysts, sat through nine live reviews and mapped forty-one distinct steps between an application landing and a decision being recorded.

The second and third days turned that map into an outcome. Northgate did not want a chatbot for analysts. It wanted the median case closed in under a day, with analysts spending their time on the cases that carried real risk and the regulator seeing a clearer audit trail than before, not a murkier one. Those became the KPIs. The constraints came from Compliance and Security: nothing leaves the bank's own data centre, every decision is explainable to a human reviewer, and the agent never issues a final rejection on its own. Northgate had no platform team to run an agent, so the brief also said that Brainbase would deploy and operate it with them, on their hardware, rather than hand over a repo and leave.

The week closed with a one-page brief signed by all three executives. It named the outcome, the numbers, the constraints, the systems in scope, and a date for the first working version. Everything after this point was measured against that page.