Retailers spend all year getting their product selection, inventory, logistics and marketing ready for year-end peak sales seasons, but they often don’t adjust their fraud screening tools for sales peaks. Surprisingly, this oversight may not lead to more fraud but to more false declines. Why would otherwise dependable fraud screening rules and algorithms turn away good customers at the most important time of the year for retailers? It’s because consumer behavior changes on big sale days. Here’s what retailers should know about the problem and how to prevent it by training their machine learning programs to adapt to holiday sales peaks.
No matter where shoppers live, special holiday sales drive behavioral changes that can trip up an online store’s fraud prevention tools. We’ll use Black Friday as the primary example, because it’s been adopted by e-commerce retailers in many countries outside the U.S. Retailers in Brazil and the UK, for example, hold Black Friday sales on the same date as U.S. stores. Mexico’s version of Black Friday is El Buen Fin (The Good Weekend). It’s four days of sales timed to coincide with the country’s Revolution Day, Nov. 20. Argentina, meanwhile, devotes three days in early November to Cyber Monday sales.
What we see in the U.S. and other markets is that deep discounts, the desire to find the right gifts and the time-limited nature of the offers can prompt shoppers to buy in ways that, on an ordinary day, would raise fraud flags.
What specifically confuses fraud programs during sales peaks? Let’s look at the major challenges.
The first challenge is the presence of more first-time visitors shopping in your store. They may even be shopping online for the first time—something that’s common in developing markets where big sales can be the reason shoppers go online. During much of the year, a new visitor making a big purchase is a fraud flag.
The second challenge is higher recurrency. Recurrency refers to how many times someone buys from your store. It can also track how many purchases a customer has made from multiple stores on the same day. Fraud prevention algorithms and rules usually give a lot of weight to recurrency, especially in combination with recency—how recently the shopper made their last purchase.
Together, recurrency and recency make up velocity, another major fraud indicator. Velocity analysis will usually raise flags if a shopper has made multiple purchases from the same store in a short period of time. And most of the time, legitimate customers don’t do that. However, during sales peaks, they may go back to a website to make an impulse purchase, buy a gift after talking with a friend or family member or go back later to finish shopping if the site crashed under heavy holiday traffic.
Another sales peak challenge has to do with credit cards. Many shoppers will use more than one on a peak day like Black Friday to work around the credit limits on individual cards. This behavior is normally a flag, because it can indicate fraudsters shopping with stolen card numbers, especially if the velocity of orders is also high.
Because consumer behavior can raise fraud flags during sales peaks, some retailers temporarily adjust their fraud controls to be less conservative. It’s easy to see how fraudsters can exploit that change. And it’s true that the number of fraud attempts does increase during sales peaks.
However, fraudsters are working with other people’s money, so they’re not worried about getting the best price. And while retailers may see more attempted fraud, the overall percentage of fraud can decrease during sales peaks.
For example, if 20 out of 1,000 orders are fraud on a given day, your percentage of fraud is 2%. If fraudulent orders increase to 30 on a peak sales day, that’s a 50% increase in fraud. However, if your total orders hit 2,000 on that peak day, the overall percentage of fraud decreases to 1.5%.
Data from 2010 to 2020 shows this holiday-related change in fraud percentage at scale. During that time, the average percentage of fraud during the peak sales period from December 24 to January 1 was 46% of the percentage of fraud for the rest of the year.
| Year (Peak Sales Period from December 24 to January 1) |
Percentage of Fraudulent Transactions
|
| 2010
|
49.4%
|
| 2011
|
66.4%
|
| 2012
|
47.6%
|
| 2013
|
46.8%
|
| 2014
|
37.3%
|
| 2015
|
37.2%
|
| 2016
|
34.1%
|
| 2017
|
35.3%
|
| 2018
|
67.0%
|
| 2019
|
37.7%
|
| 2020
|
42.4%
|
That’s not to say an increase in fraud attempts isn’t a problem. Many retailers see an uptick in fraud-related chargeback notices in January, when cardholders discover fraud. But because of the way card companies calculate chargeback rates, chargebacks from November or December may be counted against lower order volume in January. That results in a higher chargeback rate for the retailer, which may mean they pay more for transaction processing.
There’s a better way to handle fraud prevention during sales peaks. Rather than turn away good customers by using rules suited to non-peak days, or letting fraudsters slip in by opening the gates a little wider, you can create special thresholds and special rules just for your sales peaks.
To do this, you’ll need to train your machine-learning algorithms on data from similar sales peaks instead of average days. And you’ll need to verify that training with testing before you switch over to the special rules on big sales days. Then, as each sales peak passes, your system will have more data to learn from and get better at implementing its special rules.
When should you start this? My suggestion is as soon as possible. Most retailers invest so much time throughout the year building toward a profitable Black Friday or other holiday sales season. It’s wise to work year-round on an optimal fraud prevention process during those important sales days, too.
Original article at: https://about-fraud.com/fraud-screening-for-retailers-prepare-for-volume-spikes/