Credit scoring model for personal loans – Spotter Loans

Published on March 15, 2018 by Damon Rasheed

We have been working with Spotter Loans for some time now developing a credit model calculating the probability of a customer going into arrears on a personal loan (loans up to $5000) based on several features including the customer’s credit file, 3 months’ worth of bank statements and their online application. The data is based on Australian customers, and we have also overlayed relevant census data broken down to postcode level for additional data points to enrich the data.

Spotter Loans asked us to use machine learning to try to calculate the arrears probability to reduce the number of files going into collections and help them improve their responsible lending obligations by using AI/Machine learning to better target customers who are more suitable for the loans. The short-term lending market is also susceptible to a significant amount of fraud, where customers take out short-term loans with the intention of not paying anything back to the lender. As part of the brief, we also looked at trying to reduce the level of fraud by using machine learning to try to identify applications that are more likely to be fraudulent.

Without going too much into the IP that has been developed, there were a lot of interesting and unexpected findings. Theo, our Senior Data Scientists, has previous experience working for a bank in detecting credit card fraud, so his insights were extremely powerful. He managed to find obscure features that were significant in determining the probability of loan arrears. One of the more interesting areas of research was around the loan reason. Spotter Loans elected to use free text when capturing the loan reason and using text sentiment applications considerable insights were gained by analysing the text string. That finding was one of many “out there” series of features.

The other interesting finding is that machine learning can help tailor loan terms to help the client avoid arrears. For customers – the shorter the loan term the less fees a customer pays, all else being equal, however, there is a trade-off as shorter loan terms mean higher repayments for a given loan amount, which increases the chance of arrears which attract additional fees and chargers.  Machine learning has shown to help identify suitable loan repayment terms which can be discussed with the customer prior to taking out the loan. As a customer you typically want to choose the shortest possible loan term, that will not put you into arrears. This is also desirable for the lender, as considerable expenses and resources are put into arrears collections.

The IP that we have developed partnering with Spotter Loans is available for licensing for any short-term lenders looking to improve their arrears and minimise collection efforts. Although the data was collected in Australia, the methods would be equally applicable to most regions in the world. We have developed an interface that makes it easy to administer the process as well which can be integrated into most CRM system. Contact us if you are interested in hearing more.