Fairness in AI: applying deep learning to credit scoring
AI brings huge advantages... So, what’s preventing adoption?
Jaywing's Archetype wins its third and fourth award at the 2019 Credit Awards
13 September 2019
Having been shortlisted for two awards at last night’s award ceremony, which honours all high achievers within the credit industry, our AI team were pleased to pick up both awards for:
- Artificial Intelligence in Credit & Collections solution
- Innovation in Credit
Credit scoring is a well-established field, but most organisations hit a wall in terms of the predictive power of their data using traditional techniques. Designed by modelling experts Jaywing, Archetype was the first commercially available Neural Network-based scoring product which solves the ‘black box’ problem around explainable and controllable models powered by Artificial Intelligence. The software enables analysts to squeeze much more insight from predictive models and has proved particularly popular in credit risk, where lenders have seen up to an 18% uplift – which could save potentially millions in bad debt enable lenders to make better consumer decisions.
Archetype is unique in the marketplace, because it offers the user full up-front control of how the model is generated, and guarantees an explainable output without any loss of predictive power.
Archetype offers a mathematical guarantee that its output will adhere to behaviours specified by the user for each individual variable, an innovation for which Jaywing has a patent application pending. Archetype enables clients to get the most out of their investment in data whilst, crucially, retaining a significant degree of control over how the model behaves. In just a few steps, Archetype generates a highly predictive model that adheres to the analyst’s rules, coupled with a series of charts and graphs that enable the analyst to understand how the data has performed within the model, making the models both explainable and intuitive.
An enhanced credit process
Since creating the product in January 2018, we have run proof of concept projects with over 20 lenders. A number of these are now deploying these models for live use.:
- Hitachi Capital – 7.2%-11% uplift in results
- Newcastle Building Society – 18% uplift in results
- Nationwide hails Archetype a “powerful alternative in building credit scores”, and offered a 6% uplift on an already strong model
- Virgin Money – 12% uplift for risk models, 16% uplift in assessing customer churn
Matthew Jones, Head of Retail Modelling at Nationwide said: “Archetype clearly represents a powerful alternative in building credit scores. We were impressed by the ease of use, the rapidly-realised benefits and ability to enforce intuitive behaviour on key variables in neural networks. Jaywing’s approach and depth of knowledge has made us confident that the use of neural networks and similar technologies has reached a level of maturity where it can be deployed safely within the credit risk arena.”
Archetype’s methodology enables lenders to build models that adhere to the FCA’s requirements for ‘sufficient interpretability’, going beyond explainable AI to prevent the model from exhibiting unwanted behaviours. The approach enables users to get better at identifying risk or fraud cases, to identify customers at risk of leaving for other providers, or even to make better marketing offers. The same software is used to monitor and rebuild models as required.
Built by both award-winning web development and Artificial Intelligence teams, Archetype is an intuitive, browser-based system, hosted by Jaywing, with no need for any IT or data integration.
Archetype comes ready to use, and if development data is in good order, an initial model can be produced in under an hour. The system automatically generates model code output in a range of scripting languages for direct integration with any modern decision platform; alternatively, your system can call an Archetype model using an API.