How Machine Learning Speeds Up Fraud Detection

By Russ Banham

Forbes

In their work to unearth evidence of fraudulent activities, forensic accounting investigators dig through diverse data looking for anomalies that suggest something is just not right. But as the massive volumes of data collected by companies balloon, this task has become increasingly arduous, time-consuming and humanly impossible.

The regrettable consequence is the greater chance of a well-thought-out scam slipping through the cracks. A case in point is healthcare fraud, which has been estimated to cost the United States tens of billions of dollars annually.

For forensic accounting investigators, unearthing these crimes manually is an uphill climb. “The fundamental issue is that there is a flawed approach in examining fraud, since fraudsters know the rules that are set up to catch them,” says Justin Bass, chief data science officer at Crowe, the global accounting, consulting and technology firm combining specialized industry expertise with innovative technology solutions.

Bass provides the example of money laundering rules, which require banks to report any cash transactions of more than $10,000 to regulatory authorities. In response, “fraudsters simply break up the cash transactions into smaller amounts,” he explains. “The rules are created to catch these smaller amounts, but then the fraudsters circumvent them with other methods — which leads to creation of other rules and other subsequent actions by fraudsters to evade those new rules.

Machine Learning To The Rescue

Now there is a way to circumvent fraudsters via the use of machine learning(ML), the subset of artificial intelligence giving computers the ability to scan a haystack of data in search of the proverbial needle and progressively improve this capability through continuous learning.

Instead of investigators manually reviewing spreadsheet rows and columns, looking for three or four data elements that together indicate a suspicious transaction, ML can peruse thousands of data elements — instantly.

Applying an algorithm to this massive volume of data to tease out unique interrelationships presents a greater likelihood of detecting anomalies indicating fraud. “Whereas people generally can visualize three or four dimensions when evaluating the accuracy of a purchase order, machines can examine innumerable dimensions to ferret out the truly suspicious activities,” Bass explains.

To that end, Crowe has developed a proprietary ML tool called Crowe Data Anomaly Detection that has allowed the firm’s forensic accounting investigators to focus their efforts on higher-risk cases, reducing the time spent on those that don’t pan out, says Bass, whose team created the fraud-busting solution.

“We let the data tell us where to look, as opposed to us having to look everywhere,” says Tim Bryan, one such investigator and a partner in the Crowe forensic accounting and technology services group.

How It Works

Since the solution is capable of continuous learning, its ability to detect fraud improves by the day, Bryan notes. “Each time the tool is right about an actual anomalous transaction, the information automatically goes into the system, making it smarter. The same applies to when it is wrong, as this false positive also is incorporated.”

To detect the aforementioned money laundering schemes, the data anomaly detection solution examines the underlying data to pinpoint incongruities, clustering like-transactions together. Programmed to identify transactions under $10,000, the tool might highlight, say, if similar sums are deposited in a large number of banks across geographies, instantly detecting this atypical interrelationship. As a result, the customary latency time between when an investigator receives a transaction report and subsequently conducts a hindsight analysis is vastly reduced. “The transaction now comes in and is immediately scored by the tool,” Bryan says.

To test the tool’s ability to identify suspicious and possibly fraudulent activity, Crowe recently used the solution to analyze more than 16,000 contracts from a large telecommunications company. “With human analysts, the project took five professionals four months to complete,” says Bryan. “The machine learning tool enabled professionals to focus their time on investigating only the top 5% of transactions within one month, culminating in a 95% reduction in the amount of data the professionals needed to review, saving significant time and costs.”

Turning The Tide

That’s good news for companies (and bad news for fraudsters). “I’m confident we have a game-changer here,” Bass says.

Having successfully tested and used the tool internally, Crowe recently made it available as both a standalone software product and an add-on to clients’ existing accounting systems. The technology doesn’t just assist in detecting potentially fraudulent activities; it also illuminates human errors that could result in accounting mistakes.

“What Justin’s team has developed is what we in forensic accounting call ‘the brains,’” says Bryan. “It is industry agnostic, in the sense that it can be used in the healthcare space to look at fraudulent billing, in insurance to examine suspicious workers’ compensation claims, in manufacturing to look at fraudulent purchasing and in academia for a university to scope out fraud or errors in their expense processes.”

Since the tool enables continuous monitoring, as opposed to a one-time look back at data, Bryan says it presents the vital opportunity to improve the accuracy of financial statements across the board. “The tool finds things we couldn’t find using our rules-based investigatory procedures,” he acknowledges. “Now we’re leveraging technology to do what we’re good at — only much better.

Russ Banham is a Pulitzer-nominated financial journalist and best-selling author.

Leave a Reply