AI PoCs for the counter-fraud hall of fame

And those set to shape the future.

Welcome to Part 2 of our special blog series exploring the valuable role of Proof of Concept (PoC) projects in AI innovation for fraud prevention. For Part 1, click here.

In this edition, our Chief Client Officer Osman Khurshid, and Head of Data Science Rob Bevington, reflect on some of the stand-out projects we have been involved with at Synectics, and talk about emerging applications of AI-driven fraud detection and prevention - from identifying money mules to preventing APP scams.

Hello again.

Q - Perhaps to start this second interview, it would be interesting to hear which PoCs stand out as pivotal moments in terms of improving AI models?

A - Rob: In addition to the PoCs mentioned in the first part, I’d say a big one was with a major high street bank where we split a single Precision model into separate 1st party and 3rd party models for the same current account product line. The characteristics of opportunistic 1st party fraud and more organised 3rd party fraud are quite different, so having two focussed models rather than one yielded much better results. In fact, the 3^rd party model we implemented helped them identify impersonation fraud with an exposure of £355k in just the first four months of usage. This would probably have gone undetected previously.

A - Osman: I think I’d add two more to that list. A personal favourite we carried out was focused on identifying claims that would result in large losses. A diverse data set was required, but collaboratively we identified that the point of receiving personal injury notification was key to predicting the loss value. We were able to optimise our model based on this intelligence in order to predict the likelihood of a claim becoming a large loss at FNOL stage. In fact, we were suddenly able to identify over 50% of the large loss claims at FNOL stage – quite a result. Whilst we embarked on this before the pandemic, it gave us a headstart on modelling this behaviour for organisations especially, when we look at today’s claims inflation rate.

We also ran a number of PoC projects that resulted in successful ‘No Intention to Pay’ models going live. This is a traditionally challenging issue to tackle because there are often very few indicators which show signs of correlation within opportunistic behaviour. Leveraging National SIRA in combination with historic bad lends our models are able to identify over 20% additional 1^st party fraud cases around this area.

Q - Have you found that there are certain types of AI models that always yield better results?

A - Osman: It really does depend on the use case. However, another project we found extremely valuable was where we tested unsupervised learning vs supervised learning with motor insurance data. It sounds cool and futuristic to say that every model should automatically optimise itself and learn on an ongoing basis, however, in this PoC we found that - in terms of stand-alone results - the supervised model produced significantly higher financial savings.

Q - How do you help customers start using Precision, and what’s the role of PoC projects here?

A - Rob: It is always about understanding the customer need, i.e. the business case, and the data they have. We advise on and build the best model accordingly. In some cases, the organisation will be data mature and be able to provide information suitable to the problem statement being addressed. However, even if they have no historical data – say when entering a new market segment – we are able to help by leveraging the different data sources we have access to in order to build a model specific to their market, industry, product, and target customer base. As their actual customer base grows, we can optimise the model to tailor it more specifically to them.

PoCs provide a hugely valuable starting point. Because clients get a full suite of results and also the granular outputs, they don’t have to take our word on the efficacy of these models. Which can also be a big advantage in terms of helping organisations convince internal stakeholders who may be a bit sceptical of AI or the negative perception.

Q - Finally, are there any new, exciting developments around the corner?

A - Osman: Always! We know one of the most impactful components of a AI model is the data we have access to and an exciting one for us is to Integrate Public Sector data into Precision models to support trend analysis. We’re also working with several organisations and industry bodies to build models that help identify money mules and prevent Authorised Push Payment (APP) fraud – the latter being particularly timely given the Payment Systems Regulator’s (PSR’s) new requirements for APP fraud reimbursement.

The truth is we are always looking at new use cases and different ways to innovate our modelling in order to help customers improve operational efficiency whilst managing risk. Which is why PoC projects will always be something we love.

To learn more about integrating AI into your fraud prevention strategies, click here.