A subscription business was losing customers without understanding why. Predictive modeling identified at-risk accounts and the factors driving churn.

The Challenge

A B2B SaaS company with 500+ customers was experiencing higher-than-expected churn. They knew they were losing accounts, but couldn’t predict which customers were at risk or intervene before it was too late.

Their customer success team was stretched thin and needed to prioritize their time. They wanted to:

Identify at-risk accounts before they churned
Understand the leading indicators of churn
Create a systematic approach to customer health

The Approach

Phase 1: Data Exploration

Built a historical dataset of churned vs. retained customers going back 18 months, with features capturing behavior in the 90 days before churn/renewal.

Phase 2: Feature Engineering

Created features across usage patterns, engagement signals, and business context.

Phase 3: Model Development

Tested several approaches and chose a gradient boosting model for its balance of accuracy and interpretability. Key focus was on practical utility, not just accuracy.

Phase 4: Deployment and Integration

Delivered weekly risk scores for all accounts, top 3 factors contributing to each account’s risk, and integration with their CRM.

The Outcome

Model performance:

Identified 73% of eventual churners in the “high risk” category
60+ day lead time for most flagged accounts

Business impact:

22% reduction in churn rate over 6 months
CS team focused on accounts that actually needed attention
Identified two product issues that were driving churn for a specific customer segment

Key insight: Champion user departure was the strongest single predictor of churn.