Does AI Hold the Secret to Stopping Employee Theft?

Sponsored by Appriss Retail

Companies are promising to leverage artificial intelligence (AI) to eliminate cyber threats and diagnose cancer. Other companies describe how it will be used to predict customer behavior and execute marketing campaigns. Elon Musk, of SpaceX and Tesla-fame, warns, if unchecked, AI could doom the human race.

AI clearly has transformational possibilities—but what will it do for loss prevention?

In some ways, the future of AI is reflected in retail’s past. AI applications have been “hidden” for many years in business applications like credit scores, credit card fraud detection, direct marketing, and identity theft protection, according to David Speights, Ph.D., chief data scientist for Appriss Retail, a leader in data and analytics solutions for retail organizations. Indeed, whether LP knows it or not, AI has been critical in helping retailers successfully address the problem of return fraud for years, he said.

AI, or machine learning, represents a wholly different approach to programming. Rather than relying on “coders” to input the knowledge that machines use to execute solutions, examples are used to help software learn how best to reach positive outcomes. The weakness of the traditional approach is readily apparent: solutions can only be as good as the rules that programmers develop.

That limitation is why, in the early 1990s, the credit card industry abandoned an exception-based reporting approach in favor of methods of detection based on predictive modeling and machine learning, explained Speights. Ultimately, predictive modeling proved itself far superior to an approach comprised of rules and queries.

Speights sees retail loss prevention evolving in a manner comparable to credit card fraud detection. And we’re at the point now where machine learning can be turned successfully against the problem of employee theft, he said. “The economy of scale hasn’t always been there, hardware was more expensive, the cost to develop the models has been very high, and the ROI not clear enough,” he said. But with costs having come down, it now makes good sense to leverage AI to develop models for employee theft. “It’s very cost effective now—you’re able to get that positive return,” said Speights.

In testing with one retailer, machine learning models created by Appriss Retail were able to help the department store identify 25 percent more employee fraud cases‑while spending 25 percent less time tracking down the fraud activity. (The case study is available at the Appriss Retail website.) Over time, an approach with a higher success rate makes loss prevention departments more efficient and decreases employee fraud rates; it also proves the potential to reduce manpower requirements.

For now, Appriss Retail is building predictive models for clients based on a retailer’s historical data of employee fraud and employee terminations resulting from investigations. Soon, however, there will be a sufficient amount of collective intelligence to enable effective employee fraud detection without the need to seed models with a retailer’s specific fraud/termination data. One kind of generic predictive modeling should become available for big-box stores, another for grocery stores, and so on.

When that happens, new stores and retailers of all sizes will be able to benefit—not just those with a documented history employee fraud and successful investigations. “In the near term, it’s an adjunct, not a replacement,” explained Speights. “But over time it will become a larger and larger fraction of the of the detection model—from 20 percent to 80 percent.”

The machine learning model has proven itself capable of outperforming an LP department’s best traditional exception report, which might examine two or three variables and successfully identify fraud in maybe 20 percent of cases that the report flags.

The reason for the better performance is obvious: a machine learning approach examines—not one or a few—but thousands of variables. It learns, over time, the most useful elements in employee transactions for identifying those cases that are most likely to result in a finding of fraud. “It’s dollar amounts, the time after a void before re-ringing, it’s four to five thousand variables,” said Speights. Variables are combined, queried, and weighted—allowing the most likely fraud cases to bubble to the top—so that a red flag for a long-term employee may not be one for a new store associate. “We’re able to find exactly the right pattern of variables” that are best able to identify fraud, explained Speights.

By letting the Appriss Retail team of Ph.Ds loose on its employee fraud data, retailers are able to leverage expertise they don’t have in house. “It helps identify more cases faster and identifies problems that a retailer might normally miss,” said Speights. Moreover, machine learning isn’t limited to the factors that indicate fraud. The model fine-tunes its recommended cases to align with the preferences of the investigator and the types of cases they have demonstrated a preference for going after. The solution also doesn’t demand that investigators obtain new set of skills. “It’s not a matter of understanding AI,” said Speights. “You may not know how a car engine works, but you know to step on the gas.”

It’s not exactly clear how AI will ultimately transform all retail operations, but machine learning seems poised to usher in a new world of supply-chain optimization and targeted marketing. It could also put a drastic dent in the long-standing loss prevention challenge of employee fraud. Machine learning technology has already proven itself capable of identifying employee deviance above and beyond what is identified by typical exception reporting.

Comments

Leave a Reply

Enter Your Log In Credentials
This setting should only be used on your home or work computer.

×

Send this to friend