Churn Analysis Prompt
Prompt
You are a customer success manager analyzing churn patterns. Churned customer data (last 12 months): [PASTE: Account | ARR | Churn date | Stated reason | Actual reason (if different) | Industry | Company size | Product(s) used | Tenure at churn | Health score at 90 days before churn | Any escalations in last 6 months] Analyze: 1. Churn rate by segment — which industries, sizes, or product tiers churn most? 2. Churn by tenure — are customers churning early (onboarding failure), mid-term (value not realized), or late (competitive displacement)? 3. Leading indicators — what health score, usage, or behavior patterns were present 90 days before churn? 4. Stated vs. actual reasons — is "budget" the real reason or is it masking product or service issues? 5. Preventable vs. unpreventable — what % of churn could have been avoided with different actions? Output: Churn analysis report. Leading indicators for early detection. Preventable churn amount. Recommendations to reduce churn rate.
Why it works
The stated reason versus actual reason distinction is the most valuable element of churn analysis — customers who say they churned for price often actually churned because of product gaps or relationship failures, and diagnosing stated reasons as actual reasons produces the wrong interventions. Segmenting churn by tenure identifies whether churn is primarily an onboarding failure (early churn), a value delivery failure (mid-tenure churn), or a competitive failure (later churn). Health score at 90 days before churn validates or challenges the health score model's predictive power.
Watch out for
Churn analysis based on rep-entered loss reasons is almost always biased toward externalising reasons (price, competition) and underrepresenting internal reasons (product gaps, CS execution, relationship failures). Build a separate win/loss interview programme to get unbiased reasons from churned customers, and treat CRM loss reason data as a first-pass signal rather than ground truth for strategic decisions.
Used by