Scope
A leading MVNO operating in Europe wanted to optimize wholesale purchase costs by predicting usage pattern of subscribers and then selecting the lowest costing bundles
Solution
- Aggregation of high volume CDR data, subscriber data and product data available in discrete sources
- Developed statistical models to:
- Predict usage patterns for the subsequent months
- Prescribe wholesale bundles which need to be purchased
- Compare performance of models and self-correct as the model matures
- Developed IT solution to automate execution of statistical models and present reports to end users
- Created of reports that show accuracy of forecasts. This data will be used for:
- Refining models
- Billing the client
Impact
- The solution saves the client around €725,000 per month
- Accuracy of Sasken models is very close to the perfect foresight (PF), i.e., reactive estimates of cost after the usage cycle has completed