B2B SaaS · Customer support
First-response dropped from hours to minutes — without a chatbot that made things worse
A product-led SaaS team was drowning in ‘how do I…’ tickets. We built a retrieval layer over their docs and past tickets, with a clean hand-off when the answer wasn’t in the corpus.
Minutes
First response
was hours on after-hours tickets
Weeks
First build
one workflow, one channel
Their cloud
Where it ran
as the engagement required
Before
- Support volume growing faster than headcount; after-hours tickets sat until morning.
- An off-the-shelf widget deflected on paper but sent customers in circles.
- Agents re-asked the same context on every escalation.
After
- Routine how-to and status questions resolve in the same conversation.
- Agents start from context instead of a blank ticket.
- The team measures deflection quality, not just deflection rate.
Constraints
- Answers had to cite product docs — no hallucinated settings.
- Build had to run on their cloud, in their region.
- Shipping window: one focused workflow in weeks, not a platform rewrite.
What we built
- Indexed help centre, changelog, and a curated set of resolved tickets.
- Defined what the layer can answer, what it must confirm, and what it must never attempt.
- Escalation carries transcript, detected intent, and the customer record already loaded.
- Weekly evaluation against real transcripts — not a happy-path demo set.
“We didn’t want another bot. We wanted answers that hold up — and a hand-off that doesn’t make the customer repeat themselves.”
What this replaced
- Generic chat widget
- Manual after-hours triage
- Copy-paste between helpdesk and CRM
Get this outcome scoped for your workflow
A 30-minute discovery call. You leave with a written plan — scope, approach, and trade-offs. No obligation to continue.
We’ll tell you if a platform already fits better.
