Our award-winning experts build powerful fraud models so you can detect and prevent fraud, save on losses, confidently prioritise cases and rest assured that you are dedicating resources where they are needed most.

Since Jaywing's inception, we’ve delivered fraud models for over 30 different financial services brands.

People in a meeting

Fully explainable AI

In a market leading initiative, Jaywing developed Archetype, our proprietary (patent pending) predictive modelling software, that enables the development of fully explainable AI models. Archetype is able to exploit complex data patterns, whilst providing the ability to apply logical constraints to ensure model predictions are explainable hence removing the black box problem and the possibility of spurious predictions.

Stop reacting. Start Predicting.

Two Worlds Combine

The models developed within Archetype to identify credit risk have delivered consistently strong uplifts when compared to freshly optimised linear alternatives, translating into reduced bad debt or increased acceptance.

The same AI based approach is arguably even better suited to the modelling of fraud data where the anomalous (non-linear) patterns associated with duplicitous applications are more prevalent. As a result, we are seeing exceptionally strong results, with multiple models achieving coefficients well above 90% when applied to unseen test data. These models strongly outperform existing scores and/or unwieldly/high-maintenance rule-sets.

Harnessing The Power

The improved ability to discriminate between genuine and fraudulent applications has multiple business benefits. The most obvious benefit is that a powerful scorecard enables the detection rate to improve and so fraud reducing losses. Another clear benefit is that as False Positive Rates improve, underwriters waste less time reviewing the applications of genuine applicants, potentially enabling resource to be deployed elsewhere.

Bespoke Strategies

The re-use of the data captured for credit decisioning should enable the development of a model that can be used to segment applications based on an estimated fraud risk for whom customised customer journeys can be created. This enables the lowest risk applicants to be onboarded with minimum friction, resulting in rapid conversion and a positive user experience. Conversely, applicants for whom a higher level of fraud risk is observed will require closer scrutiny including, quite possibly, the expense of referring to costly additional 3rd party data sources. Moreover, this additional data returned can be ingested by the models to further refine the assessment of fraud risk and inform decision making.

The use of intelligent bespoke data-driven strategies, rather than a one size fits all approach, can drive significant operational and data savings.

Understanding it All

Within Archetype the data scientist the opportunity to control the directional influence of all features included in the model. This provides assurance that the overall model prediction make sense and we refer to this as ‘global’ explainability.

A historic weakness of score-based systems is that, unlike rule-based engines, the reason for referral is not clear. Archetype has this covered too, as along with an overall score the strongest impact factors are supplied informing the underwriter of the risk presented, and we refer to this as ‘local’ explainability.


The models we develop can be deployed in a growing range of codes, and relevant scripts can be deployed within your local environment. Alternatively, simply link to us via a secure API and we can return the scores and codes to you.

Award-winning solutions

Jaywing blend the knowledge and experience of building fraud models
for over 20 years, with muti-award-winning, fully explainable, AI based modelling capability.

Not sure where to start?

Leverage Jaywing's fraud experts and award-winning technology to combat fraud.

Application Fraud

Jaywing has unrivalled experience in application fraud modelling. Since our inception, we’ve delivered fraud models for over 30 different financial services brands. These models exploit the differences within and across application records, using advanced, non-linear techniques to identify pockets of fraud which would be missed by traditional scorecards or systems which trigger warnings based on data flags alone.

The resulting models identify a probability of fraud based on application and bureau characteristics, without requiring the implementation of additional systems or the acquisition of new data. Working with what you already have, our fraud models can optimise your fraud investigation activity by focusing your efforts on the highest-risk cases, reducing investigation and third-party data costs.

Learn about our award-winning work with Virgin Money