Sponsored by Appriss
Is the next step in the evolution of fraud detection technology already here?
To retail customers on couches—if they were to give it a second thought—virtual shopping might seem magical. With a few mouse clicks or screen touches, they can conjure up just about anything to appear on their doorstep.
But while these transactions may seem ethereal, the final push of a “buy” button jump starts a very physical process of product handling and movement—one with a real possibility of theft or loss at any number of points along the way before items “magically” appear in mailboxes.
Grocery fulfillment is an example. It’s becoming increasingly popular for consumers to make their weekly food purchases online, but many hands will have to touch an order before it gets to a customer, noted David Speights, Ph.D., chief data scientist for Appriss Retail, a leader in data and analytics solutions for retail organizations. “There is fulfillment and delivery, someone has to go out and do the shopping, and then you have to drive the order to the customer’s house,” said Speights. “All along the virtual transaction, there are real people touching the order, points at which people can steal things, or else customers can claim a refund—legitimate or otherwise.”
Retail organizations, then, face the challenge of taking their in-store employee fraud and return fraud detection capabilities into the e-commerce space. Appriss Retail is providing assistance in that effort through the power of artificial intelligence (AI).
AI comes in many flavors. General AI may bring to mind a robot that can become increasingly smart as it learns from interactions and its environment. There’s also more narrowly tailored AI, in which models are developed to help machines perform specific functions.
At leading retailers, AI is being leveraged for a variety of purposes. A mix of accelerated analytics and deep learning is helping retailers with pricing strategies, for example, and retailers are running daily profit optimization calculations to know how best to distribute which products to which stores. Online, AI works on customer data to help retailers provide more personalized shopping experiences and to help e-commerce platforms adapt to the needs and interest of online shoppers. AI can also fuel predictive price and forecast simulations to boost revenue by fractions of a percent, which for giant retailers can add up to millions.
For loss prevention, narrow AI facilitates increasingly smart pattern detection in retail transactions. Models become more capable over time at identifying previously undetected fraud patterns, and better at distinguishing between problematic and legitimate transactions.
Appriss Retail has been doing that kind of work for brick-and-mortar operations, helping retail investigators be more productive with tools that identify deviant patterns in refunds and in-store fraud. The company is now applying that same approach to the whole fulfillment chain associated with e-commerce transactions. “Where AI comes in is that it can take that fraud detection to the next level, to where it’s possible to detect patterns beyond what analysts might be able to typically do,” explained Speights.
It’s an area in which many retailers could do better, he suggested. One obstacle has been to get data into a platform in the right format so that LP can do the analysis necessary to identify fraud. As a result, analytics are not typically run beyond the macro transaction stage, lacking the necessary depth to detect fraud. Appriss is pulling data together for retailers from multiple databases so that it’s possible to run queries, identify oddities, and study and learn from patterns—and then turning that capability loose on e-commerce transactions. “It’s been a space that’s been neglected,” said Speights.
Finding Fraud—and Value
As it takes on projects for major retailers, Appriss is learning about how fraud and theft is infecting the physical fulfillment of online transactions. In one instance, AI models uncovered a large collusion case that would have otherwise been nearly impossible to identify because it was spread across a large number of customers. Lying to get free shipping is common, as is e-commerce return fraud, and so is customers claiming that products received weren’t the same type as ordered. “In one case, the same person was getting beer through a retailer’s grocery channel, and over and over, he’d claim he didn’t receive the beer he ordered. The retailer was giving him the same refund week after week,” said Speights. “It was only $10, but it was over and over. That’s one of the many types of fraud that you can easily spot and a type of loss that a retailer can avoid.”
For one client, AI helped identify e-commerce fraud of up to 15 percent, said Speights. Surprisingly, the retailer wasn’t taken aback. “They were actually expecting that level of fraud was there.” With fraud detection models up and running, the retailer now has visibility into those cases and is having success in shutting them down, Speights explained.
Ultimately, value from AI models derives from making all investigators 25 to 50 percent more efficient. “With it, there’s not that need to rely on one or two awesome investigators; now many investigators can be productive,” explained Speights. “It broadens the tool set of the average person to make them more efficient. It’s kind of like the advent of the word processor, where you don’t need to be a computer expert to extract value from computers.”
Retailers largely acknowledge that AI models are the future of LP. “People probably have set the bar really high, in fact,” according to Speights. Regardless of the buzz, “until you can show that you can spend $1 and save $5, many retailers won’t want to do it,” he noted.
This is the threshold that AI-fueled fraud detection in e-commerce transactions seems to be passing. “We have definitely seen it in the projects we’ve done. Whether it’s something that is true for all retailers remains to be seen, but for the ones we’ve done, and for any retailer that has fraud, then it will provide value,” said Speights. “Chances are, as long as you have employees and consumers who are willing to try to deceive you, then you’re going to find value in these models.”