Rethink portfolio segmentation with an AI-powered decision-making tool
Challenge
The First Microfinance Bank (FMFB) is a nation-wide microfinance bank, established in 2002, with an aim to alleviate social exclusion by providing opportunities to the underprivileged sector of the society. It operates under the Aga Khan Rural Support Programme (AKRSP) which has an exceptional track record of enabling thousands of poor households with the necessary economic and social relief.
The Bank wished to increase its lending portfolio to customers who would have a timely payback rate. The current segmentation methodologies were not able to capture all relationships within the data and provide an accurate solution for this, and that is where LFD's artificial intelligence models came into play.
Solution
Using a non-linear approach which utilized all the possible links between the data variables, we were able to dissect the bank's customers into various segments. We merged the bank's customer information with our proprietary collated public data, and then bifurcated the information into two sets, one to build the model, and the other to test it. The testing dataset was used to predict which segment the customer would belong to, and then the actual results were compared with the predicted results. This helped ensure a high degree of accuracy when segmenting future customers. Consequently, each segment displayed a certain payback rate, and thus we were able to identify which segments were beneficial to the bank versus those that were posing to be a liability. While developing these segments, we not only looked at the delay in payments of principal, but also the loss made on interest payments, and the cost of delayed payments, considering that money could have been lent to some other customer.
Result
Hence, LFD made loan-giving decisions simple and clear for FMFB. The bank is now aware which category of customers it should lend loans to, for how many days and for which amounts. Moreover, this will directly influence the bank's rating because the model is also able to predict loss and lower write-offs. The model can be easily integrated into the bank's current loan application systems for seamless workflow and ease of use. Simple to use and efficient, the model uses machine learning to automatically improve itself as more data is entered into the system.
Hence, new trends can be identified with time and integrated into the system.
