Canada | Fraud

Uncovering fraud: AI can help, but we still need a human hand

We’ve come a long way since the days when unravelling scams was largely manual. But even with more sophisticated tools, a critical eye remains key to spot crime.

A Facebook IconFacebook A Twitter IconTwitter A Linkedin IconLinkedin An Email IconEmail

 Image of businessman wearing glasses looking at a computer screenAlthough fraud detection methods will continue to improve, experts stress that the human eye will still drive the investigative process (Getty Images/TimeStopper)

New technology and legislation have brought big changes to the fraud detection landscape over the past 25-plus years. And even more profound changes are likely to come as artificial intelligence and machine learning become more prevalent. But whatever new tools the future may bring, experts agree that humans will remain very much in the driver’s seat when it comes to fraud detection.


In the late 1980s, when Roddy Allan, CPA, started working in forensics, it was common to see banker’s boxes of evidence stacked halfway up the walls of examiners’ offices. In fact, for the first file he ever worked on, he had to go through 1,200 boxes of information from a defunct investment dealer. “It took a year,” he says. 

At the time, says Allan (who joined Deloitte as a partner in Forensics in 2010), a case might begin in much the same way as it does now: “A client would call and say, ‘We’ve found a couple of transactions that look a little odd. Or something’s come through on a tip line.’ So we would have a thread—something to start with. Then we would go and pull the thread.’” [See This is the most common method of fraud detection]

In this largely paper-centric world, it was easy to get swamped by the volume of documentation, because there were few tools available to help sift and sort through detailed information. Most examiners relied on spreadsheets, databases and instinct (Excel was first released in 1985). Moreover, there were fewer sources of information: the internet only started to become widely used in the mid-1990s.


Fast forward to 2019, and the entire field of fraud detection and investigation has expanded. Banker’s boxes are a far less common sight, and examiners have access to more tools and information sources. “Now, if I have a person’s home address, I can look it up on Google Earth in a few seconds, whereas 20-plus years ago I’d have to get a private investigator,” says Allan. 

But rapid access to new sources of information brings its own set of problems. As Emmy Babalola, CPA, a partner in Deloitte Canada’s Forensics practice, puts it: “Today we’ve got an explosion of data from email, social media, corporate chat systems, devices and so on,” she says. “You’ve got to find a way to cull it and boil it down because you can spend weeks investigating.” 

Allan agrees: “It’s a little like being swamped in documents; you’re swamped in potential sources of evidence.”   

It’s also worth noting that paper hasn’t disappeared by any means. “We’ve had cases where paper documents really count—in poison pen letters, for example,” says Babalola. “Also, the court system is still largely paper-based.” 


As time goes on, we are likely to see an ever-dwindling use of paper—a natural corollary to the rise of digitization in firms. And we will probably see AI making inroads into detection methods.

Right now, many programs on the market can alert an organization when certain events occur based on specific instructions. These are sometimes called rules- or scenario-based programs. But AI companies are moving past this stage. For example, MindBridge Ai, an Ottawa-based firm, has developed an AI-powered auditing platform. As John Colthart, VP of growth, explains, programs like this not only incorporate rules, but are is able to constantly change and evolve based on input from humans and data (machine learning). “First you tell the data what you want to see, then let the data come back to you and say, ‘This looks really odd for these 20 reasons.’”

Colthart recalls a case in Southern California where a firm was asked to come in and evaluate the books for potential fraud. Using MindBridge’s technology, an employee who was not a fraud examiner was able to uncover about US$2.8 million of fraud—far more than the US$1 million the company thought had gone missing. 

In the future, Colthart thinks advanced data analysis techniques will become a necessity in organizations, including banks: “I don’t think there’s a bank that’s going to be able to operate unless they can prove they have used advanced technologies to sift through all their data and appropriately triage the items that come up as potential issues,” he says.


But although fraud detection methods will continue to improve, experts stress that the human eye will still drive the investigative process. As Allan explains, “An investigative mindset is still the key to a really good investigation. You must be able to critically evaluate the evidence, even though the data sources and tools at your disposal are much more extensive than before.”

Colthart agrees. He sees new AI-enabled programs becoming a co-pilot of sorts. “This is about partnering a really complex technology stack with readable, understandable explanations so that those using it are much more empowered than they are today.” He adds, “Fraud is a very systemic issue—according to the Association of Certified Fraud Examiners’ 2018 Report to the Nations, occupational fraud is responsible for about $7 trillion a year in losses worldwide. And fraud happens because we allow it. Hopefully in tomorrow’s world we will simply make it much more difficult for fraudsters to act.”