We're delighted to share that Jaywing and Virgin Money have been named double finalists at the 2026 Credit & Collections Technology Awards. We've been shortlisted in both the ‘Customer Engagement Solution’’ and Machine Learning in Credit & Collections Solution’ categories.
These nominations recognise our work in tackling two of the industry's most persistent challenges: predicting customer attrition timing and identifying persistent debt before it becomes entrenched.
Getting customer retention timing right
Attrition models usually tell you one thing: is a customer about to leave? It's a helpful starting point, but it doesn't give you a timeline. Without knowing when a customer is likely to drift, retention campaigns are a bit of a blunt instrument. You're left guessing.
So, we built a single model that predicts risk across 3, 6, 9, 12, 15, and 18 months. It creates a dynamic profile for every customer, showing exactly how risk develops over time.
We built it on Jaywing's Archetype platform using a Gradient Boosted Machine (GBM). Working as one team, we whittled down 1,743 variables to 66 features. During testing, it hit a Gini score of 89.1% for short-term prediction and remained rock-solid on out-of-time data. It means Virgin Money can ditch fixed schedules and talk to customers based on how they behave.
Nick Martin, Model Owner at Virgin Money, noted: "The longitudinal approach, which predicts when a customer might leave, was a true innovation born from a deeply collaborative process. It allows us to understand our customers on another level and engage with them at the most crucial moments."
Spotting persistent debt before it starts
Persistent debt is a significant challenge, with customers often spending months making minimum payments while interest builds before they meet the regulatory definition. We wanted to build a model that flags risk early enough to make a difference, so we built a machine learning tool that predicts persistent debt risk up to 18 months in advance. An 18-month horizon reflects the fact that customers with promotional credit card balances often take longer to reach persistent debt, while a three- or six-month view could miss those who may need support most.
The judges commended our use of a Gradient Boosted Machine, narrowing down 1,725 variables to the 59 that matter most. We also put strict rules on 44 of those variables,ensuring the model is completely transparent and makes logical sense to risk and compliance teams.
The predictive power is strong, hitting an 87.4% Gini score on testing. But the real win is in the customer response. More than 5% of targeted customers increased their minimum repayments after early intervention; a real, measurable boost to financial wellbeing.
Nick Martin commented: "The Persistent Debt model is a pivotal achievement for us. Its predictive accuracy gives us the confidence to act early and decisively, transforming our ability to support customers. We can now identify vulnerable customers with incredible precision."
Applying advanced analytics where they deliver value
We hear a lot of talk about advanced analytics in credit risk. But for us, the value is always in the practical application. Both of these projects show that machine learning doesn't have to be a black box to be highly effective. It can combine tight risk governance with real, positive outcomes for customers.
Being named a double finalist is a great result for everyone involved, and its testament to what happens when client and consulting teams work as one.