AI support copilot for a payments platform
We built a retrieval-augmented support copilot on the client's knowledge base, with human-in-the-loop escalation and full audit logging for compliance.
Where things stood
NorthArc's support team was drowning: ticket volume had tripled in two years, first-response times had slipped past a day, and hiring more agents was eating the margin the product was supposed to deliver. Two-thirds of tickets were variations of questions already answered in their documentation — but agents still handled each one by hand, and a payments company can't let an unsupervised chatbot guess about money.
What we did
Built a retrieval-augmented (RAG) copilot over NorthArc's help center, internal runbooks, and resolved-ticket history, with citations on every answer
Kept humans in the loop: the copilot drafts, agents approve — with one-click escalation for anything touching account changes or funds
Added evaluation before launch: a 400-case golden set scored weekly, so quality regressions surface before customers see them
Shipped full audit logging of every retrieval and response to satisfy the compliance team from day one
What the numbers say
42% lower support cost per ticket within one quarter of launch
First-response time down from 26 hours to under 2
68% of tickets resolved without human escalation, with CSAT unchanged
Zero compliance incidents — every answer traceable to a source document
Built with
- Claude
- LangGraph
- Qdrant
- FastAPI
- Next.js
- PostgreSQL
- Azure
MIVIRI didn't sell us AI hype — they shipped a copilot that our support team actually uses every day. Ticket costs dropped 42% in the first quarter after launch.
Priya Raghavan
VP of Operations, NorthArc Payments
Start the conversation
Want a result like NorthArc Payments's?
Tell us what's slow, expensive, or broken today — and what the number should look like a year from now.