Background
DataVision provides analytics software to mid-market companies. Their product had strong adoption among new accounts, but retention lagged industry benchmarks despite a capable customer success team.
The challenge
DataVision's CS team reacted to churn too late—often after cancellation requests. They needed predictive signals and automated interventions before accounts disengaged irreversibly.
Our approach
NexisBlue partnered with DataVision's product and CS leaders to build a custom machine learning pipeline:
- Feature engineering from product usage, support tickets, billing events, and NPS responses
- 30-day churn prediction model trained on historical cohorts
- Real-time inference integrated into Salesforce with risk score fields
- MLOps infrastructure for retraining and drift monitoring
- Playbooks triggered by risk tier—high-touch outreach, training offers, executive escalation
Change management
CS reps received training on interpreting scores and when to override model recommendations. Weekly reviews validated false positives and refined features.
The results
- 94% accuracy on 30-day churn predictions at launch
- 27% reduction in overall churn within two quarters
- CS team prioritized outreach with data-backed risk scores
- Average time-to-intervention dropped from 18 days to 4 days for at-risk accounts
Key learnings
Predictive retention only works when models connect to action workflows. DataVision's win came from integration and playbooks, not algorithm sophistication alone.
