B2B SaaS · AI Chat Agent
Sample writeup70% of support tickets resolved before a human sees them.
A Series B B2B SaaS had a support backlog growing faster than headcount and a knowledge base nobody read. We embedded with their support team, deployed a RAG agent grounded in their real docs and tickets, and deflected 70% of inbound while raising CSAT.
- Client
- Anonymized · Series B B2B SaaS
- Industry
- B2B SaaS
- Region
- United States
- Duration
- 8 weeks
The challenge
Support volume up 3x year-over-year, headcount frozen. Median first-response time had crept past 4 hours; CSAT was sliding. The team had tried two off-the-shelf chatbots; both hallucinated policies that didn't exist and got pulled within weeks.
The approach
Identify. Build. Adopt.
Identify
We read the last 6 months of real tickets, segmented by intent, and mapped which 12 question types covered 78% of volume. Built the eval harness against 400 real tickets before writing a line of agent code.
Build
RAG agent grounded in their actual docs + ticket history + product changelog. Reranker tuned on real queries. Hard guardrails on anything pricing- or policy-adjacent; those route straight to humans. Embedded on the marketing site, in-app, and Slack-Connect for enterprise customers.
Adopt
Two-week shadow mode: agent drafts, humans send. Then full deploy with a single 'send to human' button always one tap away. We sat in their Slack daily for the first month, reading every escalation and tuning the retrieval.
Outcomes
What changed, measurably.
Other studios pitched us a slide deck. Orelyn pitched us a working prototype on day three. That's the difference.
Got a similar leak?
30 minutes with the founders. Bring us the workflow that hurts most. We'll cost the leak live and tell you whether the same playbook deploys here.
More work