Proactively managing loan eligibility helped our client lower the number of loan defaults.

Snapshot

Effective credit risk management is the key to improving efficiency in the lending industry. Our client had a sophisticated credit model but felt that there was opportunity to improve this using data science and the Azure machine learning (ML) platform to reduce loan defaults and improve customer profitability and marketing efficiency from improved segmentation. A short period of intense analysis yielded a number of immediate opportunities for improvement of the credit model and an estimated potential upside of <30% improvement in performance.

Background

Velrada was engage to give the client a strong capability uplift in the area of data science, big data management and analysis to assess their current credit model, including the algorithm, features, range of data sets considered, to identify potential scope of improvements in the prediction outcome. An exploratory exercise was carried in less than one week, leading to a series of recommendations which would underpin a significant improvement in the performance of the credit model for both new and returning customers.

Approach

Velrada engaged in rapid, iterative sessions with the client team to stress test existing assumptions and the modelling approach, then used existing data sets to test a range of hypotheses related to various components of the model and the relationships between them. In addition, given the highly sensitive nature of credit model performance from real data, we explored several international data science competitions where similar exercises had been carried out by thousands of data scientists across the globe, based on near-realistic data. This was then used to develop and test further hypotheses.

Delivery

The current prediction model has been developed using Microsoft Azure ML platform, leveraging the Azure ML studio as the user interface. A number of custom scripts had been written and injected into the model for various purpose such as data pre-processing, derived feature calculation, plotting, consuming the output (scores) from the model, and profitability calculation. Velrada’s data science team reviewed the technology and data integration approach, and worked with the client to create a road-map for better leveraging the existing technology components as well as provided recommendations on further opportunity to extend the platform to perform more valuable analysis and improve integration with other operating systems.

More like this

Defining the vision and fast-tracking the big data and data science opportunity.

Taking the lead on data strategy and the vision for the connected customer.