Predicting Delinquencies of a Personal Loan Portfolio ahead in time​

New Age Fintech Company


  • Provide a view on which part of the loan portfolio will likely show signs of stress on repayments at least 6 to 9 months ahead in time
  • Need for using external variables other than account related variables
  • To lower Cost of Acquiring Customers (CAC), the insurer turned to online selling
  • Traditional customers were not used to online purchase and self-service. The web experience had to be designed optimally
  • The insurer wanted to deploy web-chat effectively to make interactions personal


  • Nanobi’s Early Warning Solution was implemented and tuned to meet the specific segment portfolio. The solution scores all live accounts in the specified portfolio every month of their propensity to default or be delinquent over the coming six to nine months time
  • The Early Warning Solution uses macro or external variables other than account or internal variables
  • 32+ variables (internal and external) were used to predict early warning delinquencies (the solution has a library of 90+ variables)

  • False Positive rate – 2% to 6% and False Negative rate – 12% to 15%
  • 75% to 82%  Model  Accuracy

  • Overall model accuracy was between 75 to 80%
  • Low False Positive cases and False Negative cases showing the strength of the model
  • OOB Score of 84% showing that the model can handle scenarios that didn’t occur in the training / test dataset