Sherlock for fraud detection: How algorithms can support manual credit analysis

In broad terms, buyer fraud refers to an attempt by a buyer or potential buyer to defraud a seller during the course of a transaction. There are multiple types of buyer fraud, ranging from reshipping to phishing. But the result is always the same: the fraudster disappears without paying their debt.

Identifying these scammers before they can complete their operation sounds simple enough in theory. After all, every insurance company runs these checks in order to protect their customers, and themselves, from fraud. In practice it requires meticulously analyzing large amounts of buyer data, which can be both complicated and time-consuming.

Historically, we have relied exclusively on our expert analysts to manually conduct these buyer assessments. But with increasingly large volumes of both data and cases as well as a growing digital marketplace, we needed an extra pair of eyes. That’s why we’ve turned to Sherlock. 

Sherlock is Allianz Trade’s machine learning solution that helps us detect suspicious cases more efficiently. By pulling from multiple sources and using advanced algorithms, the tool is able to analyze huge amounts of data rapidly and accurately. 

Of course, we do not give the tool unrestricted access to data – we prioritize our clients’ data privacy, and do not use any personal information such as names or dates of birth.

Much like a human credit analyst, Sherlock’s buyer assessment checks are run at the credit limit request stage, i.e., when one of our clients approaches us to cover a sale against non-payment risk. Sherlock’s analysis involves thoroughly checking data, and flagging suspicious cases for our credit analysts to investigate further. 

Our philosophy is that people make decisions, and algorithms draw attention to issues. In tandem, they make a powerful combination. While an algorithm can analyze hundreds of features in one second, it can’t review data that it doesn’t have. And real-world human experience offers intuition and knowledge about people that help determine fraudulent cases.

What this means in practice is that Sherlock cannot make a final decision on credit limit approval. Based on its analyses, it determines the probability of buyer fraud and sends a report of its results to a credit analyst who will then review the case.

There is no machine replacement for our credit analysts’ experience, expertise and instincts. Our aim in adopting tech is simply to complement their work, to free up precious time for deeper analysis, and to set them up for maximum success. This, fundamentally, enables us to support our clients and their businesses better.