Customer Churn Investigation

Discover what predicts customer churn and identify at-risk customers before they leave.

---

The Question

"What predicts customer churn?"

Or variations:

• "Why are customers churning?"
• "Which customers are likely to churn?"
• "What factors drive customer retention?"

---

What Scoop Investigates

Scoop uses machine learning to find churn predictors:

Investigation Plan:
├── Probe 1: Quantify churn rate
│   └── What's the current churn rate and trend?
├── Probe 2: Build predictive model
│   └── ML analysis: What factors predict churn?
├── Probe 3: Segment churned customers
│   └── What patterns exist in churned customers?
├── Probe 4: Time-based analysis
│   └── When in the lifecycle does churn happen?
├── Probe 5: Identify at-risk customers
│   └── Apply model to current customers
└── Synthesis: Actionable predictions

---

Example Output

Investigation Results: Customer Churn Prediction

MODEL ACCURACY: 84%

PREDICTIVE RULES DISCOVERED:

Rule 1 (Confidence: 89%):
IF last_login > 21 days
   AND support_tickets >= 3
   AND plan = "Basic"
THEN churn_risk = HIGH

Rule 2 (Confidence: 85%):
IF usage_trend = "Declining"
   AND contract_renewal < 60 days
   AND NPS_score < 7
THEN churn_risk = HIGH

Rule 3 (Confidence: 81%):
IF onboarding_incomplete = TRUE
   AND days_since_signup > 30
THEN churn_risk = MEDIUM

TOP CHURN PREDICTORS (by importance):
1. Days since last login (34% impact)
2. Usage trend last 30 days (28% impact)
3. Support ticket sentiment (18% impact)
4. Contract type (12% impact)
5. Feature adoption score (8% impact)

AT-RISK CUSTOMERS (applying model):
├── High Risk: 47 customers ($892K ARR)
├── Medium Risk: 123 customers ($1.4M ARR)
└── Low Risk: 1,847 customers ($18.2M ARR)

RECOMMENDED ACTIONS:
1. Immediate outreach to 47 high-risk accounts
2. Re-engagement campaign for users inactive >14 days
3. Onboarding completion program for incomplete users
4. Basic plan upgrade incentives before renewal

---

Sample Prompts

Prediction Focus

"What predicts customer churn?"

Root Cause Focus

"Why are customers churning?"

Segment Focus

"What predicts churn for enterprise customers?"

Action Focus

"Which customers should we reach out to before they churn?"

Comparison Focus

"What differentiates customers who churn from those who stay?"

---

Follow-Up Questions

---

Data Requirements

For best churn prediction results:

---

Understanding the ML Results

Predictive Rules

Scoop presents ML findings as human-readable IF-THEN rules:

• Easy to understand and act on
• Confidence percentage shows reliability
• Can be directly translated to alerts/triggers

Feature Importance

Shows which factors matter most:

• Higher percentage = stronger predictor
• Focus interventions on top factors
• Deprioritize factors with low impact

Model Accuracy

• 80%+ = Highly actionable
• 70-80% = Good for prioritization
• <70% = May need more/better data

---

Tips for Better Predictions

1. Include behavioral data - Login frequency, feature usage
2. Add engagement signals - Email opens, support interactions
3. Capture the outcome - Make sure you have churned/retained labels
4. Historical depth - 6+ months of data improves accuracy
5. Include negatives - Support issues, payment failures

---

Related Patterns

• Customer Segmentation - Group customers by behavior
• Win/Loss Analysis - Similar predictive approach for deals
• Period Comparison - How churn changed over time