Kestrel
AI customer support that automates 80% of tickets.
Deployed a domain-trained AI agent and smart escalation flow that saved $400K annually on support headcount.
- 80%Tickets automated
- $400KAnnual savings
- <30sResponse time
- 4.7CSAT (5)
Where they were when we started.
Kestrel was a fast-growing B2B SaaS receiving 2,000+ daily support tickets. Response times had stretched to 36 hours; the support team had grown 3× in 18 months and was still losing ground. Their first AI experiment — a generic chatbot bolted onto Intercom — had churned itself out with single-digit deflection rates.
How we shipped it.
- 01
Built a domain-trained agent on Claude with a retrieval layer over their actual product docs, ticket history, and changelog — not a generic FAQ scrape.
- 02
Designed a confidence-aware escalation flow: the agent routes to a human the moment it can't cite a source for its answer, with the relevant context already attached.
- 03
Wired full ticket-level analytics into Looker — every conversation evaluated for resolution, satisfaction, and escalation reason.
- 04
Shipped a continuous-improvement loop: humans rate every escalated ticket, those signals retrain the retrieval ranking weekly.
What changed for the business.
- 80% of tickets resolved without human involvement within 30 seconds, validated against a hand-labelled audit set.
- $400K annualized savings on support team scaling that no longer needed to happen.
- CSAT for AI-handled conversations: 4.7/5 — equal to their best human agents.
- Escalated tickets carry full context, so human handle-time on the harder ones dropped 40%.
“Big Bold's AI support agent handles 80% of our customer queries. Our human team went from drowning in tickets to focusing on the complex stuff that actually needs them.”
Have a project in mind?
Let’s build it.
Tell us what you’re working on. We’ll come back within 48 hours with a sharp first take, a rough scope, and a path to v1.