Scoring & Inference Engine

Health scoring and churn prediction system

This engine evaluates the health of every customer account and predicts the likelihood of churn using behavioral signals from the product, support activity, customer engagement, and renewal timelines.

The scoring process runs daily and helps Customer Success teams identify at-risk accounts early and take proactive action.

Daily Scoring Pipeline
The system runs a multi-step scoring pipeline for all customer accounts.
1

Health Score (0 – 100)

Each account receives a Health Score between 0 and 100. The score represents how healthy the account is based on product usage, adoption, support activity, engagement with the success team, and renewal timeline.

Health Score Interpretation

80 – 100 → Healthy
60 – 79 → Stable
40 – 59 → At Risk
Below 40 → Critical

How the Score is Calculated

Each factor contributes a percentage (%) to the total score. The percentage represents the importance of that signal in determining account health.

Example: If Product Usage has a weight of 35%, it contributes up to 35 points out of the total 100 score. More important signals have higher weights.

Health Score Signals

Product Usage
35%

Measures how actively the customer is using the product.

  • • Daily / Weekly / Monthly active users
  • • Product activity trends
  • • Feature usage frequency

Since product usage is the strongest indicator of value realization, it has the highest weight.

Feature Adoption
20%

Measures whether customers are using the product's most important features.

  • • Key feature adoption
  • • License or seat utilization

Higher adoption usually means customers are receiving more value from the product.

Support Experience
20%

Measures customer support activity and product friction.

  • • Number of support tickets
  • • Open or unresolved tickets
  • • Support sentiment

Frequent issues may indicate problems with product usability or stability.

Customer Engagement
15%

Measures how actively the customer interacts with the Customer Success team.

  • • QBR meetings
  • • Customer success calls
  • • Email interactions

Higher engagement generally correlates with stronger customer relationships.

Renewal Timeline
10%

Considers how close the account is to its renewal date.

Accounts approaching renewal with low engagement or adoption require attention.

2

Machine Learning Churn Prediction

In addition to the health score, the system predicts the probability that a customer may churn. A machine learning model analyzes historical behavioral patterns across customer accounts to estimate churn risk.

The model looks for patterns such as:

  • • declining product usage
  • • low feature adoption
  • • increasing support issues
  • • reduced engagement
  • • approaching renewal dates

Using these signals, the model calculates a churn probability score.

Model Output

Each account receives a Churn Probability score between 0% and 100%.

0 – 25% → Low
25 – 50% → Medium
50 – 70% → High
Above 70% → Critical

The higher the probability, the more likely the customer may churn without intervention.

Why Machine Learning is Used

Rule-based scoring provides a structured health score. Machine learning improves this by:

  • • identifying patterns across large datasets
  • • detecting early signals of churn
  • • improving prediction accuracy over time

As more historical data becomes available, the model continuously improves.

3

Automatic Alert Generation

The system automatically generates alerts when risk signals are detected. Alerts help Customer Success teams take proactive action.

Examples include:

churn probability rises above 70%
health score drops significantly within a short period
a high-risk account is approaching renewal
sudden drop in product activity
spike in support tickets

Data Flow

Customer Activity Data
Feature Analysis
Health Score Calculation
Machine Learning Prediction
Risk Alerts
Run Scoring Pipeline
Runs the full scoring process for all accounts and updates: Health Scores, Churn Probability Predictions, and Risk Alerts