Top five priorities when considering the use of machine learning in fraud prevention

Fortunately, accessing advanced fraud-protection solutions based on machine learning (ML) has never been as straightforward as it is today. At Experian, we’re at the forefront of developing ML-based services that meet the needs of companies of every size and market position. And ML is continuously proving its value in helping our clients not only drive more benefit and value from their existing products. It’s also directly helping them to reduce the operational burden, prevent more frauds from taking place and streamline the journey for good customers.

Here, we look at the five priorities which underpin those organisations that get the best outcomes from using ML as part of their fraud-prevention toolkit.

1. Embrace machine learning

The first is also by far the most important. Quite simply, invest in, embrace and use machine learning as part of your day-to-day operations. Today, for just about any business operating in the credit, personal finance or retail space, ML capabilities simply have to be an essential part of the fraud-assessment toolbox.

Why? Because your competitors are already doing so and reaping the benefits associated with it. Failure to match their level of capabilities will simply hand them an unnecessary advantage. Some companies continue to believe that ML is still the preserve of a few market leaders at the forefront of technological investment. If this ever was true, it’s no longer the case. Today, even new start-ups are launching with ML-based solutions in place, and the benefits extend far beyond being able to spot fraud quickly. They can also serve their customers better and run a leaner organisation in which machines and people focus on what they are best at.

In short, ML is a competitive necessity.

2. Update your fraud strategy

The second priority is to ensure your fraud-prevention strategy is regularly updated. Some companies put in the technology, set up the rules and then leave everything as it is for several years. But the fraud threats they face mutate every day, meaning their strategy should be constantly evolving too.

Fortunately, the ability of ML to consume and analyse multiple different data-points across a company’s complete product portfolio also makes it an ideal tool for analysing and future-proofing your strategic approach to fraud prevention. So, use the technology for more than simply spotting fraud from day to day. Use it to define the bigger picture too. Adopting ML can be a great point to pause, take stock, update your strategy and turbo-charge the return on investment from your existing solutions.

3. Never stop investing

The third priority is closely related to the second: ensure you are continuously investing in your existing capabilities. Just as with your fraud strategy, failure to invest will lead to a gradual degradation in the performance of your fraud-prevention tools and the returns they deliver.

This is particularly important as companies move into new markets. Look at the example of Amazon stepping into the insurance space, or established banks and financial services businesses effectively entering retail with ‘buy-now, pay-later’ products. The new threats they’re suddenly seeing through these new channels and product portfolios are very different from those they’ve been fighting for years.

Again, ML can be an extremely powerful resource when it comes to protecting against new threats or updating and upgrading their existing estate. But there are challenges involved, especially if you are tempted to think you can build and maintain your own ML model. In mid-market organisations in particular, there is sometimes a failure to realise how expensive and tricky this can get.

For example, the data scientists you’ll need to build your team around are very expensive and hard to find these days. And they’ll be kept busy, as an ML model needs constant monitoring for quality deterioration and to ensure it consistently treats customers fairly and without bias.

In addition, ML models need to be retrained on new data sets and updated much more frequently than you might think – another operational cost. Finally, don’t underestimate the difficulties involved: while building a quick model for R&D is relatively easy, putting it into production and integrating it into existing data sources is a lot harder.

4. Work in partnership with ML

Fourth, it’s essential to ensure that nobody in the fraud-prevention team starts to see ML (or rather Artificial Intelligence in general) as a threat. It sometimes happens that fraud professionals feel that by investing in the technology, companies are suggesting that machines can do their jobs better than they can.

Nothing could be further from the truth. Look at it this way. Companies typically operate ‘supervised’ ML, where data relating to fraud-related observations often gathered over many years is used to create a Machine-Learning model in which data may be analysed at scale. This plays to the strengths of machines and humans working together: ML can consume as many data points as can be fed into it. And humans can look at the results using logic based on their knowledge and understanding of existing fraud outcomes to ensure the strategy is optimised.

In short, allow ML to empower you to do a better job for the business. And free up your team from interminably looking at data, enabling them to focus on those high-risk areas that really matter.

5. Find the Holy Grail

The final priority is what could be called the Holy Grail of fraud prevention. That is, using ML to deliver against the three elements that make up the total cost of fraud: increasing the amount of fraud you catch; reducing the operational burden that tackling fraud can place on your organisation and its people; and improving the journey for your good customers.

To achieve this ideal outcome, you need to have a multi-layered fraud strategy in place, with technology that can run in parallel or sequentially to address email, devices, identities and other sources of risk. The right ML solution will then take all the data from all these sources and analyse all the rules that have been triggered to create a fraud probability score.

This then provides the basis for a strategy that empowers you always to make the best fraud-related decisions. Achieving this and reaching that Holy Grail is virtually impossible without using an ML solution that brings together that total capability.

At Experian, we are constantly evolving our ML portfolio and the individual solutions within it to keep our clients at the forefront of fraud-prevention best practice. Find out more about the next generation of fraud prevention or contact us now.