The challenge
A leading commercial bank wanted to automate credit limit increase decisions for its credit card holders. It lacked a model to estimate each customer's natural spending propensity, what they would spend without a limit constraint, and to determine which customers are eligible for an increase and what the optimal amount is.
At a glance
StackApache Spark, R, XGBoost, TensorFlow
Duration3 months
DataAnonymised transaction and revolving-balance history
What we did
SWICON delivered a two-phase machine-learning solution on Spark, R, XGBoost and TensorFlow. Phase one, a Natural Spending Propensity Estimate model, predicts limit-free spending capacity per account, separately for transactor and revolver segments. Phase two, an Eligibility Assessment business model, converts those estimates into targeting decisions and offer amounts, incorporating provisioning and business rules. The output is a custom regression result per customer, not a segment.
The result
The bank received a model that determines, for each customer, the exact amount up to which raising the limit stays at the optimal intersection of profitability and risk cover, rather than applying uniform limit-increase bands. Feature-engineering scripts, analytical reports and knowledge transfer to the CRM team were part of the delivery.
