A deployment write up, anonymised. The company is a tax services business in the United States. The name is withheld because publishing it requires the client's approval.
The situation
Two problems that look separate and are the same one.
An existing customer base that nobody was working, because the team was busy with current filings. And outbound activity that nobody could trace, so nobody could say which spending to keep.
What was built
The CRM became the trigger. A function inside the CRM fires a webhook on each stage change. Nothing runs from a spreadsheet somebody has to remember to update.
Seven flows, one job each. Intake, the enrichment branch, the sheet to CRM upload, the call request, the after call write back and the bad contact route. Splitting by job means a failure in enrichment does not stop the calls.
Enrichment before contact. Every lead leaves the CRM, gets waterfall enriched for contact details and company signals, is scored, and returns before a person opens it.
Channels connected, not parallel. Email sequences written by level of awareness, AI calls for the volume the team could not reach, and SMS designed and held pending carrier registration. Replies and outcomes return to the CRM as leads with a stage rather than as noise in an inbox.
A stamp on every record. The campaign, the list, the handoff, the call, the outcome. The least impressive part of the build and the one that mattered most.
The result
The first signed contract after launch was traced back to the reactivation system, through the handoff, the conversation and the list it came from.
That sentence is the whole point. Pipeline gets a project attention. Traceability is what gets it renewed, because it answers the only question a finance conversation actually asks: what happens if we stop paying for this.
The hardest part was not technical
Agreeing what each CRM stage meant.
Several automations branch on stage. When a stage means different things to different people, identical records get treated differently and nobody can explain why. Writing the definitions down, and documenting that changing one cascades into three flows, took longer than building any of them.
The rule that caught the silent failures
A green execution is not proof. Every flow was built so verification happens on the record: open the lead, and the business result is either there or it is not.
That single change in the definition of done surfaced failures that would otherwise have run quietly for weeks, which is the normal failure mode for automation that nobody is checking.
What would be done differently
Start the carrier registration for messaging on day one rather than when the SMS follow up was ready. It was the only part of the plan that waited on someone else's clock.