Churn Analysis PDCA Framework
A structured human-AI collaboration process for monthly SaaS customer churn analysis. Cycle frequency: Monthly (first week of month, covering prior month's cohort data).
How to Use This Skill
Work through phases in order. Each phase has a STOP condition before proceeding.
PLAN — Confirm cohort definitions and define the analysis approach before executing.
See references/phase-prompts.md → PLAN phase.
DO — Compute rates, surface patterns, and draft the deck with discipline. Human
intervention is mandatory at defined gates.
See references/phase-prompts.md → DO phase.
CHECK — Verify quality against the agreed criteria before declaring done.
See references/phase-prompts.md → CHECK phase.
ACT — Retrospect. Propose refinements to this skill. Human approves changes.
See references/phase-prompts.md → ACT phase.
See references/working-agreements.md for the current version and process discipline rules.
STOP Triggers — Intervene Immediately
If any of these occur, stop the AI and restate the relevant phase prompt:
- AI makes any causal claim about why customers churned (any "because", "due to", or "driven by" framing not directly quoting the analyst's own words)
- AI redefines cohort boundaries or segmentation without the analyst's explicit confirmation
- AI proceeds to draft the presentation deck before the analyst has confirmed the pattern review is complete
- AI adds a recommendation slide without the analyst authoring or explicitly approving the recommendations
- AI generates "key takeaway" or "headline finding" framing without the analyst's input
Human Ownership — Non-Negotiable
The human owns these decisions. AI must not proceed past them without explicit approval:
- Causal attribution: any statement of why customers churned is owned by the analyst, not AI
- Signal vs. noise decisions: which patterns are meaningful given business context AI does not have
- Leadership narrative framing: what the churn data means for business strategy
- Recommendations and action priorities: which segments to focus on, what interventions to pursue
- Sign-off on the final presentation before it is sent to leadership
- Cohort definition confirmation: how cohorts are segmented before AI computes any rates