Case Studies
Improved ROI from sales efforts
The issue
How to recommend strategic and operational changes with the aim of increasing the value of credit reports
The approach
- Identification of levers for improving ROI, seasonal factors, opportunities and file quality
- Development of an analytical profitability model for sales prospecting
- Prioritization of prospecting by type of customer and demand
The results
- One-third reduction in costs with no decrease in revenue
- Management of the sales team made more flexible
Prediction of loan non-payment
The issue
How to dynamically assess the latent risk of home loans to individuals by creating a credit score based on customer behaviour
The approach
- Creation of explanatory variables with high added value from data available internally and in open data
- Development of a machine learning model based on decision trees
- Risk assessment by a credit score defined as the probability of non-payment within six months
The results
- Effectiveness of the credit score: multiplied by 2 with respect to the partner’s existing indicators and by 3.2 with respect to the “Basle” regulatory credit rating
- Concentration of resources on cases requiring preventive management of defects and adaptation of the provisioning system
- Mapping of underwriting risks by agency and by region
Prediction of bank customer churn
The issue
How to build predictive indicators of customers using several banks or churn
The approach
- Creation of explanatory variables with high added value, based on available data (bank flows, volume of savings, use of means of payment)
- Development of a self-learning gradient descent model
The results
- Construction of a monthly score for each client indicating the risk of churn within six months
- Learning automatically and regularly updated with new data and performance monitoring of the machine learning model implemented
- Very good feedback from customers contacted, doubling the number of products purchased and identification of life events based on the analysis carried out
