Secure Trust Bank deploys Jaywing’s ‘explainable’ AI for application scoring

Source: Jaywing

Secure Trust Bank has become the first UK lender to complete live deployment of credit application models built using Archetype, the proprietary ‘explainable’ AI-driven software of credit scoring and AI experts, Jaywing.

The Archetype software generated a significant uplift in predictive power compared to a traditional linear model and is projected to provide substantial reductions in bad debt over the first year and beyond. The company will be working on further predictive model developments throughout the next year.

This landmark deployment is part of a three-year engagement which will see Jaywing involved in regular monitoring and calibration to ensure the models are always optimised. It enables Secure Trust Bank to fully realise the benefits of neural net-based modelling, initially across two of its portfolios. Archetype will generate significant performance improvements using a technique that ensures that the data used within the model fully adheres to the lender’s governance rules - just as their previous scorecards did.

Like most UK lenders, Secure Trust Bank has taken a close interest in the potential for neural networks to deliver double-digit improvements in risk model performance. However, the lender was mindful of the FCA’s guidance that such models need to be based on sufficient interpretability. Jaywing’s software offers complete control over the data that is used by the model, and the way in which changes in the data impact the resulting score.

With a full suite of in-depth reporting to explain how the resulting model performs, Archetype generates easily implementable models which can be deployed through any modern decision system, either by automated code generation or the use of an API.

Commenting on the deployment, Mike Abraham, Group Head of Consumer Credit Risk at Secure Trust Bank said: “The performance improvements that Archetype has generated are significant, and we are impressed by Jaywing’s ability to deliver this performance through a fully controlled model. The in-depth model reporting gave us total confidence that the models are fully reliable and have the stability and robustness that we expect. Archetype prevents the models from exhibiting unwanted behaviours, and after extensive testing we are confident that this represents a considerable improvement in our ability to improve our lending decisions.”

Martin Smith, Head of Product Development at Jaywing, commented: “Many lenders are assessing AI-based models, undertaking parallel runs and monitoring. But Secure Trust Bank has become the first to realise the promise offered by explainable AI and controllable models to tick the regulatory box around sufficient interpretability, and to push ahead with a live deployment. This gives it a huge commercial advantage, and we are delighted to be supporting the business as it transforms its approach to credit scoring.”

Comments: (0)

THE STATE OF DIGITAL BANKING, 2019 - Landscape: The Digital Banking Strategy Playbook