4most have a long-standing relationship working with a leading provider of financial services. With a significant proportion of new loan applications referred for manual review and decisioning to an underwriter team, the business had identified that their decision process was slower and more complex than desired. 4most was engaged to develop a solution which would enhance the acquisition process with a view to improve customer experience, reduce costs and obtain more consistent decisions. All to be implemented in line with current regulations and compliance requirements.
The project was delivered in two phases; firstly, the current decision strategy was optimised. This was achieved using a combination of traditional analytical techniques and Natural Language Processing to identify potential process efficiencies and consistent underwriter behaviour. This alone resulted in a 30% reduction of underwriting effort. In addition, a comprehensive and transparent database was created, from which the target population for the data science exercise could be identified from the remaining referrals, where decisions were found to be either too complex or inconsistent.
The second phase started with a horse race between several machine learning (ML) algorithms to establish the model structure that best described the relationships in the data to predict the target (Accept/Reject). A key advantage of using ML techniques over more traditional methods was the ability to handle large volumes of data effectively and therefore outperforming existing policy rules in the identification of desired customers.
The model to take forward was agreed with the business according to a set of pre-determined criteria. We then used our bespoke tool, Columbus, to facilitate stakeholder review and sign-off of the final ML model and strategy.
Columbus is a flexible and customisable R and Python based tool, which allows the user to explore a model and identify relationships in the data. Regardless of model complexity, the tool allows users to understand the model output at both aggregate and individual case level. Columbus focuses on outcome interpretation, understanding how certain profiles influence the modelling target, also making it a comprehensive model monitoring solution. Its capability to explore individual cases and identify the reasons underlying model decisions and recommendations is a distinguishing feature. This benefit enhances transparency and understanding and supports compliance in heavily regulated industries. In fact, Columbus allowed us to show a higher level of consistency than human decisioning therefore aiding a fairer outcome for the customers.
We were aware of the technical and operational challenges involved in fully integrating ML techniques in the credit risk decision process, but our persistence resulted in impressive results and we demonstrated that underwriting effort could be reduced by up to 80%.
In addition to cost efficiency, we delivered a streamlined and efficient credit risk underwriting process, which resulted in:
Improved customer experience – through faster application processing
More informed decisions – due to the ML models’ ability to utilise more extensive data
Consistent decisions - resilience to human biases and oversights
Simplified decision process – reducing underwriter effort in processing applications requiring manual checks
Flexible approach - can be constantly and easily adapted to changing scenarios.
Underwriting is a time consuming and costly part of the application decision process and being able to automate this was a huge advantage to our client and its customers alike.
For more further information, please view the Machine Learning page within the Services Menu, or alternatively please contact our Head of Data Science: Fabrizio Russo, firstname.lastname@example.org