Enterprise AI · Banking · 0-to-1
PNC's First GenAI Platform
- Role
- End-to-end strategy lead, only non-engineering member of the deployment team
- Stakeholders
- ML Engineers, Data Scientists, Solution Architect, Business Stakeholders (a sales team within Corporate and Institutional Banking)
- Scope
- Thousands of documents across 3 knowledge repositories
- Outcome
- 90% weekly active users · 98% answer acceptance · 4.8/5.0 satisfaction
The problem
A sales team within Corporate and Institutional Banking helps businesses move, store, and optimize their cash, which means they need to know product, policy, and pricing detail cold before every client call. That information lived across three different knowledge systems with basic search. Finding a reliable answer could eat significant time, and still leave gaps.
What I did
I was the only non-engineering person on the build, working alongside a Technical Lead, Solution Architect, ML Engineers, and Data Scientists, without writing a line of code myself. My job was translation: turning a business problem into something engineers could build against, and turning technical results into something business stakeholders could trust. I built the business case, defined the success metrics that would matter, and got it funded. Then I scoped the document corpus by actual usage, so we weren't trying to solve everything at once. The most important thing I built was the Golden Set, a rubric of real questions and answers that became the shared standard both teams tested against. It turned "is this good enough" from a debate into a measurable threshold. I ran user testing the same way, designing the test cases and evaluating what came back, and every failure got sorted into what engineering needed to fix and what the business needed to fix in the source documents. That's what gave us the green light to launch: not a gut call, but a clear picture of what was left and who owned it. Then I ran change management myself: training, job aids, demos with real questions from their actual work, launch communications, and the surveys that told us whether the business case was holding. Across fourteen months, my job was to be the person who could see the whole thing at once: what the business needed, what the technology could actually do, and what had to be true for people to trust it enough to use it.
The result
90% weekly active users. 98% answer acceptance. A 4.8 out of 5 satisfaction score. And the process itself became a playbook, now the template being used to scale 5 more use cases across the bank, two of which I'm currently building.