How AI Fights Fraud During the Holidays
Christmas is just over a week away, which means the holiday shopping season is in full swing. Consumers are spending billions of dollars per day on gifts in anticipation of the big day. But the fraudsters are also out in force to steal a piece of the action. Luckily, AI and machine learning are getting better at identifying these grinches before they ruin things for the rest of us.
The math is pretty simple: The bigger the holiday buying season, the bigger the pay day for fraudsters. Deloitte predicts that online sales during this holiday season will increase between 14% and 18% compared to the previous year, accounting for $149 billion in sales. Overall holiday spending is expected to top $1.1 trillion, it said.
According to data from the Forter Fraud Attack Index, the overall dollar volume of fraud rose 12% between the second quarter of 2018 and the second quarter of 2019, with activity soaring once holiday shopping began. “Since the holiday season represents about 20% of annual retail sales, expect an increase in attempted fraud,” Forter writes in its holiday white paper.
In the UK, 94% of retailers say they’ve witnessed a rise in fourth-quarter fraud since 2016, according to GBG, with 31% saying that the increase is “significant.” The uptick in activity during the “gold quarter” provides cover for fraudsters to blend in with the holiday traffic.
That provides plenty of room for fraudsters to do their dastardly deeds, according to the folks at Terbium Labs, a provider of data security solutions. “Criminals rely on consumers to stay distracted during the holiday season, and less likely to notice unauthorized transactions on their accounts during the bustle of holiday parties and gift-giving,” Terbium writes in its November 2019 report, “How Fraud Stole Christmas.”
Fraudsters have become extremely diverse in their techniques over the years, thanks to an inherent curiosity in developing innovative approaches to separate hard-working people from their money. According to Terbium, card-skimming and stolen cards are the types of fraud that people fear most, followed by phishing scams and (everybody’s favorite) card-not-present fraud.
Card-not-present transactions soar during the holiday season, as many people prefer to stay at home and shop online as opposed to going to the store. While shoppers may not think there’s a big difference between shopping online and in-person, it makes a big difference to the merchant and the firms processing the transaction. Card-not-present transactions are inherently riskier, and therefore banks will compensate consumers who fall victim to fraud, but the merchants are held liable.
Regardless of the way fraudsters worked, about two-thirds of people say they would hold their bank at least partly responsible for fraudulent activity, regardless of how the compromise occurred, Terbium found. That puts the onus on retailers (which lost 3% of revenues to fraud in 2018, per LexisNexis) to ensure that fraud doesn’t happen on their watch.
Detecting the fraudulent activity quickly is the key to stopping it, and that’s where AI comes in. Machine learning gives the good guys (banks and merchants) a big technological advantage against the fraudsters in several ways.
First, traditional rules-based approaches may erroneously flag unusual purchasing activity. But the problem is that holiday shopping, by its very nature, is unusual. People buy all sorts of goods and services that they wouldn’t normally buy during the other 11 months of the year.
For example, gift cards and jewelry are holiday staples for good-intentioned citizens looking to make a friend or relative happy. But these are also the items that the bad guys commonly buy when they have a stolen credit card number, a compromised account, or are gearing up to execute some good-old “card-not-present” fraud.
Machine learning algorithms are able to identify correlations in the data that could indicate fraudulent activity is underway. Based on these predictions, a bank can either refuse to run a transaction, or it could initiate another round of authentication, such as by requiring customers to enter a special code that’s sent via a pre-registered mobile device (i.e. multi-factor authentication). That can help protect the merchant, as well as the consumer.
The big advantage of machine learning is that it’s more adaptable to those imminently versatile fraudsters than old rules-based approaches that the industry used to use. The challenge is that machine learning requires sophisticated data science modeling techniques and the human expertise to tweak and apply the models, according to TJ Horan, a vice president for FICO.
“If done properly, machine learning can clearly distinguish legitimate and fraudulent behaviors while adapting over time to new, previously unseen fraud tactics,” Horan writes a recent blog post. “This can become quite complex as there is a need to interpret patterns in the data and apply data science to continually improve the ability to distinguish normal behavior from abnormal behavior. This requires thousands of computations to be accurately performed in milliseconds.”
MasterCard is taking steps to protect consumers and merchants with its new Identity Check program. According to Chris Reid, Mastercard’s executive vice president of services for North America, the company is feeding AI models (including neural networks) with upwards of 150 different data types.
MasterCard’s models take all sorts of factors into account — everything from the IP address of the buyer and the value of the item being purchased to the brightness of the mobile device being used to initiate the transaction.
“The more data that you can apply into making a decision, the greater level of confidence you can have about that decision, and as a rule the more accurate you’re going to make that decision,” Reid told Datanami earlier this year.