There’s no arguing that self checkouts (SCOs) are an essential part of a retail operation. They save labor, increase productivity, shorten customer waiting times—increasing satisfaction in the process—and other benefits. However, the lack of oversight by employees creates significant risk of mistakes or even fraud, especially transactional fraud. There is an unfortunate perspective amongst would-be SCO thieves that SCO theft is low-risk, easy to pull off, rewarding and carries few-to-no consequences even if they are caught (see Dr. Adrian Beck’s excellent 2018 research paper Self-Checkout in Retail: Measuring the Loss). Every year, retailers lose millions of dollars in profits to SCO fraud. Many are completely unaware, while others treat it as a cost of doing business, without calculating the cost itself.
The necessity of SCO is obvious—but the losses associated with it need to be calculated constantly. In addition, there must be a clear understanding of potential improvements in order to calculate the total benefits of this key element of retail.
There’s Luckily, there is a solution available that can identify even the most subtle forms of SCO fraud. Prescriptive analytics is a software methodology that uses AI and machine learning to analyze data and determine:
- What happened
- Why it happened
- How much it costs
- What to do about it
- Who should do it
Here are several common instances of SCO fraud that prescriptive analytics can easily recognize and flag by using data.
Product Scanning Substitutions
SCOs present golden opportunities for price switching. Without trained cashiers to ensure scanning accuracy, dishonest customers can easily reduce the prices they are charged for more expensive items. Common examples include tag switching (attaching a cheap product’s barcode to a more expensive one’s barcode) and PLU code switching (punching in the PLU code for bananas and placing a rib roast on the scale, thus ringing up the $18/pound roast at $0.49/pound—a loss of more than $122 for a single, seven-pound roast!). Even if this behavior is noticed by the SCO attendant, proving it to be fraud (versus a simple, honest mistake) can be very difficult.
A retailer’s best bet is to turn to the data. A simple way to identify price switching involves monitoring inventory movements. All those false sales will result in unusually high movements for the cheaper products and unusually low movements for the stolen products. Together, these two behaviors create a distinctive indicator of pricing fraud, especially if they are only occurring at a select number of stores.
Another way to detect price switching is to look at item quantities per order. When a typical customer buys bananas, for example, he or she will place all the bananas onto the scale, enter the code, and move them to the conveyor belt. A customer committing SCO fraud, on the other hand, may ring in multiple expensive products as bananas, resulting in the fruit appearing on multiple lines throughout the transaction. Prescriptive analytics can easily detect any of these behaviors and alert asset protection to launch an investigation, mark the customer scanning the bananas as “high-risk” (i.e. of committing fraud) or another action as appropriate.
In some cases, prescriptive analytics can be used in combination with digital imaging cameras at self-checkout. The camera identifies whether the item placed on the scale matches the product code entered by the customer. If there’s a mismatch, the register alerts the customer to the mistake, inviting him or her to re-enter the item. Prescriptive analytics can advise loss prevention of customers who frequently trigger the alerts.
In a previous Voice, I spoke of how cashiers can commit “sliding” by passing items over their scanners without actually ringing them in. Customers can do the exact same thing at the self-checkout register, by scanning one item and placing it plus another item on the belt, or simply walking out with the item still in their cart.
Prescriptive analytics can narrow in on the shoplifters and discourage them from future offenses. The solution clusters customers based on their typical transaction behaviors by type of checkout. For example, the solution may identify that Joe Smith only buys meat when he checks out at a staffed register—he never buys meat at self-checkout, despite the transactions being relatively similar in all other respects. This could indicate theft, and from there the solution can alert LP to launch an investigation, increase Joe’s risk status or flag him for a full or partial audit at his next SCO transaction.
Markdowns are an essential part of retail as they encourage customers to buy near-expired or slow-selling products that would otherwise go in the trash for a loss. However, markdowns are rather easily abused by dishonest customers who can’t resist the allure of a massive discount on expensive products. This is especially prevalent after holidays, when unsold seasonal candy is marked down as much as 80 percent. Unfortunately, these tags are easily removed and added to more expensive items like meat or even energy drinks. If bought through self-checkout, there are no cashiers to recognize the suspicious markdown.
Prescriptive analytics provides a solution to this common problem. The best solutions leverage machine learning to identify the average benchmark level of markdowns per store, and from there consistently monitor the actual number of markdowns at any given time. Should a store experience an above-benchmark number of markdowns, an opportunity is generated and asset protection is alerted to launch an investigation. Alternatively, a prescriptive analytics solution can be configured to monitor the monetary value of markdowns (i.e. the amount of money the retailer is losing to markdowns). Again, if this value reaches abnormally high levels, asset protection is immediately alerted.
Stolen Credit Cards
Stolen credit cards are an increasing concern in retail. When a thief successfully steals one, his first stop is often the nearest retail store, where he spends the funds on gift cards before the credit card’s owner can cancel it. This is an increasingly common trend.
To avoid getting stuck with these fraudulent transactions, retailers can leverage prescriptive analytics to identify certain combinations of data behaviors that indicate credit-card fraud. One approach that has proven successful involves flagging transactions for which multiple declined credit cards were used to purchase multiple gift cards of increasing value (ex: a $50 gift card purchased, with a $500 gift card immediately after). The logic is that the credit cards were stolen and the thief is transferring the funds to gift cards, a little at a time. This incremental behavior determines which of the stolen cards are still active and tests the available credit on those confirmed active. When this happens, LP is alerted to launch an investigation, and any gift cards confirmed to have been purchased with stolen credit cards can be cancelled. The retailer avoids significant losses and does its part to help combat credit-card fraud.
With its ability to assess the risk of any basket containing fraudulently priced items and trigger audits or investigations as necessary, prescriptive analytics is an invaluable asset in retail’s battle against fraud in the self-checkout lanes. Having this robust solution in-house empowers retailers to minimize risk and losses, which translates into increased profits, margins, and revenue.