fintech · Series B fintech
A support copilot that shipped with real evals
A retrieval-grounded support copilot taken from demo to production behind an evaluation set, cutting resolution time.
- -43%
- First-response time
- 94%
- Answer accuracy
- 1 in 3
- Tickets auto-resolved
measured on eval set
The challenge
Support volume was outpacing the team. An earlier AI prototype gave confident wrong answers and nobody trusted it enough to ship.
What we did
We built an eval set from real tickets first, then a RAG pipeline grounded in the product docs with citations, guardrails, and a clean handoff to humans. Every change was measured against the evals before release.
Stack
Our support copilot went from demo to production with real evals behind it. Resolution time dropped and customers noticed.
More work
Related case studies
Odoo rollout for a stalled manufacturing ERP
A year-stalled Odoo project taken to production in one quarter, with custom modules that matched real shop-floor operations.
Headless re-platform for a high-traffic store
A storefront re-platformed to a headless architecture that held up under launch-day traffic and lifted conversion.
let's build it
Have something to build?
Tell us the problem. We'll come back with a plan, a price, and who'd actually build it.
- Free scoping call
- Reply within 1 business day
- No lock-in