Fraud prevention: the industry’s myths and hypes
Roberto Valerio & Dustin Clinard with RISK IDENT reveal the myths and hypes surrounding the fraud prevention industry and best practices in leveraging machine learning and smart data
There is a lot of talk about leveraging Big Data and the value of using it for fraud prevention. What is your take on this?
Big Data plays a major role in fraud prevention, however an equal focus should be directed towards how much data is necessary to properly execute the fraud prevention process. It is often said that the more data one adds, the better results one gets. We have also noticed a trend in many marketing messages that claim the same thing: the more data, the better. Nobody talks about millions of transactions anymore, but about billions of transactions; a dozen data points for a transaction is replaced by tens of thousands of data points for a transaction.
This isn’t good. The numbers are becoming inflationary, and we have the impression that the figures are growing. However, this doesn’t mean that the quality or accuracy of the fraud prevention results are getting any better. Certainly, there are numerous patterns that can be found with Big Data, but we have also seen many fraud patterns that can be derived from a rather small set of data.
When so much data becomes available, it is important to analyze all aspects to see which part of the data can be helpful for one’s goal. And if the goal is fraud prevention, which and how much data is it needed to be analyzed in order to get the best results? Are bigger numbers key for a sophisticated and good working fraud prevention solution? Isn’t the quality of data the main thing that matters? Our answer to this is that data quality and relevance are much more important than the amount of data.
For example, we receive huge amounts of data from our customers, who rightly believe it is all relevant for our fraud prevention tools and results. However, when performing the data analytics processes, we sometimes observe that a merchant has over 150 to 200 data points, of which when analyzed we find only about 30 or 35 are valid predictors in the fraud prevention process.
Can Big Data be turned into smart data? If yes, how?
This is a matter of analytics. There is a lot of research conducted in the area of automatic data analytics – data analytics based on machines and algorithms. Machines have become more powerful in analyzing information and the algorithms have become more sophisticated as well. Data science plays a critical role, and this type of data science is not easy to process – it requires a lot of know-how, is computationally intense, and the techniques are rapidly changing. However, there are many proficient people operating within the payments and fraud prevention industry who know how to get smart data. The interesting part is to see what you can do with data analytics before you put the data into the machine learning algorithms. When it comes to feature extraction, the solution lies in taking information out of the relevant data so that machine learning algorithms can make the best predictions.
To sum up, the combination of domain knowledge, people that understand the business, good mathematicians and data scientists is key to having a good development within our product space.
Will machine learning overpower human intelligence?
Beyond its scientific meaning, machine learning has become a buzzword. But essentially, machine learning is a vast area of expertise. The models that are being used are now very similar and some of them have existed for a long time. Machine learning can support humans to make the right decision and can replicate good decision making. Yet making decisions, building a model and bringing it to a high-quality standard implies human effort. Therefore, we believe machine learning is not going to replace humans, especially on such a complex topic like fraud prevention. However, we may be partially replaced by artificial intelligence, but only with our permission. For instance, in the fraud prevention area, there is a lot repetitive work that can be run by robots. Looking for account takeover, for example, may rely on some common signals to catch the basic fraud attempts – this can be automated. However, account takeovers often have a unique signature that wouldn’t repeat over and over – this will inherently need human input.
What learning points may the ecommerce industry consider in the future, to streamline the potential of artificial intelligence for fraud prevention?
One learning point would be the prioritization of needs. When a merchant is looking for a solution provider, they can set up the piece of the solution is the most important for them. In some markets, organizations are mainly focused on preventing fraud, while in other markets you can observe merchants looking to avoid false positives, or that markets organizations are struggling to adapt to the changing customer patterns, especially in emerging markets where customers display a versatile behavior.
At present, it seems like there’s still an amazement about what is technologically possible. Eventually, this state of surprise will fade and we will be able to have better discussions around the problem that a company is trying to solve, which will help in deciding what technology or company may be the best one to partner with.
As an Overall Sponsor of MRC Dublin, could you share with us some key takeaways from this event? What was the pulse of the industry?
The MRC has always been a source of open and transparent communication. MRC events are a good environment to exchange strategic ideas, to make room for diversity in terms of discussions, to address long-range topics like GDPR and PSD2, and to learn more about new technologies. This event in Dublin was no different. The MRC hosted several practice-based debates on machine learning and the power of data analytics in consolidating the fraud prevention. The ‘black box’ is becoming less opaque, especially as everybody is trying to figure out how to make this work in their businesses and this transparency is a significant improvement.