Collaborate to Disrupt: Fighting fraud with alternative data

Even the most sophisticated fraudsters leave behind traces of their activity. Revealing these tell-tale signs of fraud requires an alternative approach, as we heard recently in the third webinar of our ongoing Collaborate to Disrupt series.

GBG’s Carol Hamilton and Max Excell were joined by First Abu Dhabi Bank’s Charanjeet Singh and CredoLab’s Michele Tucci for a conversation about harnessing alternative data and machine learning for multi-layered fraud and compliance decisioning.

Alternative data is information that can’t be acquired from credit bureaus or from the application process itself, which are the two primary sources of data used by organisations in the financial services sector.

During the webinar, which you can replay on-demand here, the group heard how alternative data can be used to spot behaviour that might otherwise go unnoticed to help deter and unmask fraudsters.

Understanding alternative data

Explaining the concept of alternative data - including how, where and why a business may choose to harness it, Michele said there are broadly two kinds of alternative data sources: transactional and behavioural.

Transactional alternative data might include things such as telecoms data, payments, ecommerce data, open banking API data and utilities, to name a few. Behavioural alternative data could include things like smartphone device, travel patterns or ecommerce activity.

Michele went on to explain how an organisation might choose which alternative sources of data to use, citing coverage, specificity, accuracy and timeliness as key factors in the decision-making process.

Click here to hear the conversation in more detail.

Upwardly mobile

Speaking to the practicalities of using a specific form of alternative data, Charanjeet explained how organisations that allow customers to onboard via a mobile app can take advantage of some of the unique features of a phone to catch out criminals.

Charanjeet said: “Rather than allowing the registration to happen through Wi-Fi, you can enforce that the registration is only done through mobile data. That way, you capture a lot of information which otherwise will not be available. You can make full use of metadata power and you can check whether the same phone number is being used to pump out multiple applications.”

Similarly, he explains, there are tools that can tell you how long the applicant’s email address has been in, whether it is connected to any social media accounts and even if it’s been previously associated with fraud.

What’s more, email ID and phone logs can be cross-referenced against your own organisational data to reveal fraud involving collusion with your staff.

Machine learning

Also on the agenda was machine learning as a means of going beyond traditional and even alternative data sources to detect and prevent fraud.

GBG’s Regional Director EMEAA Carol Hamilton offered six tips on the best ways to augment your fraud prevention processes using AI.

The first tip was to avoid a ‘black box’ approach with an open, user-controlled interface which helps your experts build and train the score thresholds in a way that optimises detection, which will vary based on your organisation's risk appetite and resources.

The second tip was about transparency and the need for users to have oversight on the contributing factors behind a given score. This is important for audit purposes and for tracking, as well as offering assurance to the user.

For more of Carol’s tips on implementing machine learning and much more insight from Charanjeet Singh and Michele Tucci, click here to stream the webinar on demand.

.red { fill: #b0013a; }