← All work

Automation · AI agents

Account health and anti-churn engine

Every account scored each week on eight signals, with the right next step chosen before anyone cancels.

Delivered
  • Node.js
  • REST and SignalR
  • monday.com GraphQL
  • Latenode
  • Cloudflare Workers
  • Headless Chrome PDF

What was hard

There was no single view of how 250+ client accounts were doing. Unhappy clients were only noticed once they cancelled.

Who it affected

A service team responsible for hundreds of client accounts on WhatsApp.

What we built

  • A weekly pull of every account's messaging data, scored out of 100 on eight signals
  • Green, amber and red bands on the team's CRM board, with a one-click branded PDF report
  • An anti-churn engine that sorts accounts into Nurture, Watch or Save and picks the next message from approved templates
  • Guardrails: one automated touch per account per week, and a check that each message was actually delivered

How it works

  1. 01Collect each account's activity
  2. 02Score eight signals
  3. 03Band and explain the score
  4. 04Choose and verify the next step

What changed

  • Risk across 252 accounts is visible every week, not at cancellation
  • Runs on about 9% of the monthly automation budget
  • A hidden pagination cap was found and fixed: one test account went from 200 contacts reported to 11,569