The Problem
A 32-person B2B SaaS company had recently raised a Series A and was acquiring customers faster than its 4-person customer success team could onboard them. New customers were waiting 2-3 weeks for a kickoff call. Support tickets were piling up in Linear with no prioritisation; complex customer questions sat for days while simple ones got answered fast (loud-customer bias). Renewal conversations were happening 30 days before contract end. too late to influence outcomes.
The CFO had modelled out hiring three more CSMs. The CEO asked: can we do this differently?
What I Built
A four-part stack:
1. Self-serve onboarding sequence. New customers receive an interactive onboarding flow tailored to their declared use case (extracted from sign-up data via Claude). Key steps are tracked; if a customer stalls, a CSM is alerted with the exact step + suggested talking points.
2. Support ticket triage. Every inbound ticket is classified by Claude:
- Severity (blocker / standard / nice-to-have)
- Category (auth, integration, billing, feature request, bug)
- Customer tier (enterprise / SMB)
- Suggested first response (where the model has high confidence)
Tickets are routed to the right engineer/CSM with context attached. Triage time per ticket: 4 minutes → 20 seconds.
3. Customer health scoring. A nightly job analyses product usage, support volume, payment history, and contract terms; produces a per-account health score with a one-paragraph reasoning. CSMs start their morning by reviewing the bottom 10%. not chasing the loudest customers.
4. Renewal forecasting + early intervention. 90 days before each renewal, the system flags accounts at risk and drafts a tailored CSM intervention plan (specific value moments to highlight, expansion opportunities, decision-makers to engage).
- 01TriggerNew customer signs up
Sign-up data flows into Claude, which extracts the declared use case and primary success metric.
- 02StepTailored onboarding sequence
Interactive onboarding adapted to use case. Stalled steps page the assigned CSM with talking points.
- 03StepInbound tickets triaged
Every ticket auto-classified by severity, category, tier and routed to the right engineer or CSM with context.
4 min → 20s - 04StepNightly health scoring
Usage, support volume, payments and contract terms feed a per-account health score with reasoning.
- 05DecisionRenewal risk flagged at 90d
Bottom-decile accounts surface with a drafted CSM intervention plan: value moments, expansion plays, key contacts.
- 06OutputCSM acts before churn
Human review + sign-off on every customer-facing message. AI proposes, CSM approves.
+$340K renewal saves
Outcome
The company avoided three CSM hires (~$390K annualised). Onboarding became a competitive advantage. they started referencing the 36-hour time-to-first-value in sales pitches. Annual churn dropped from 18% to 10%. for a SaaS company at their scale, that's roughly $1.4M of recurring revenue retained.