South KoreaProfessional services2025 – ongoingSK Shieldus
Churn prediction and predictive control models for security customers
SK Shieldus raised its churn model accuracy from under 70% to over 90% and cut unnecessary dispatches 2%
Problem
SK Shieldus, a South Korean security company with more than 1 million customers, kept its data in separate business units and the quality was low. Its older models were not accurate enough to spot customers likely to cancel or to separate real security alarms from false ones.
Approach
The company built an integrated data analytics platform on Databricks in six months. On it, it built a cancellation prevention scoring model that flags at-risk customers for outreach. A predictive control model identifies abnormal security signals to guide dispatch decisions. Model experiments are tracked with MLflow, and the company also added generative AI guidance for customer consultations.
Results
Takeaway
Better data quality came first, and the model gains followed. The company reported a small saving in dispatches and did not say what the churn model did for cancellations.
Categories
- Size
- Large enterprise
- Industry
- Professional services
- Business function
- Customer serviceProduction and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Databricks reports that SK Shieldus built a data platform in six months and used it for a churn prediction model and a predictive control model for dispatches. It says model accuracy rose from under 70% to over 90% and that unnecessary dispatches fell 2%.
Full details are in the original sources.
