Drove 25% increase in credit card applications through Machine Learning based response models

Financial Services, Marketing

We used machine learning models to drive 25% increase in credit card applications


  • Client was a US credit card issuer, acquiring customers through Direct Mail and Digital channels
  • We optimized DM strategy by first developing a “Cold Start” response prediction model (for the initial campaign) and then a refined model later on
    • Cold start model used bureau data
    • Refined model used own campaign data
  • Approach involved:
    • Horseracing regression vs. Gradient Boosting (GBM)
    • Evaluating sub-models
    • Testing ~2k variables


Hypertuning of Parameters (GBM Model)









Model Performance of one segment (comparison of Techniques)










  • Hypertuning of GBM was performed and compared against CVLR approach
    • Bayesian optimization done
    • Final 150+ variables used
  • Overall model had 80%+ accuracy; GBM model outperformed regression models by 5-10 Gini points
  • Final DM campaign led to ~25% more applications for same mktg. budget