Credit risk and Artificial Intelligence (AI) specialist, Jaywing, has announced the results of an exploratory piece of work undertaken for Newcastle Building Society, to increase the effectiveness of its risk application models using Jaywing’s AI-based product, Archetype.

Newcastle Building Society recently undertook some enhancements of its customer risk scores and wanted to understand whether applying AI-based techniques to mortgage risk modelling could yield additional benefits. That’s why it appointed Jaywing, with its considerable experience both in traditional and AI risk modelling. Jaywing took Newcastle Building Society’s existing scorecard development samples for both Buy to Let and residential mortgages, using Archetype to create the models from scratch in just a few hours.

Dr J. Serradilla, Senior Data Modeller at Newcastle Building Society, said “Jaywing’s flexible approach and comprehensive track record in both traditional and AI risk modelling made it a sound partner for us to work with to boost the effectiveness of our existing models. Archetype’s flexibility provides us with a range of model implementation options, from deploying model code in our own decision platform to using Jaywing’s API, and we are working with Jaywing to create the most advantageous approach for us to ensure the benefits are realised ASAP”.

Martin Smith, Jaywing’s Head of Product Development said “We are delighted to partner with Newcastle Building Society on this exercise, and our work with them adds to a consistent pattern of uplift from running proofs of concept for our clients.  Without fail, we are seeing an uplift in Gini compared to an optimised linear model, often as high as Newcastle’s 18.6%, dependent on the data, the portfolio and the number of cases available for modelling.  In most cases, this translates to a huge improvement in bad debt.  It’s not just an improvement in model validation either – results on out of time samples show that the models are robust and continue to be more predictive, with a lower level of degradation compared to a traditional approach.