Dusseldorf Fraud in online trading has never been easier than it is today. Plenty of stolen data is available on the Darknet and the criminals’ scams are becoming more and more diverse.
Hilla Peled, Chief Data Scientist at the anti-fraud platform Riskified, knows how easily scammers get the data via Telegram channels: “It only takes five minutes, for example, and costs around five US dollars per credit card.” Some fraud tools such as shopping bots and VPN services are even available as subscriptions.
The fraud attempts can hit anyone, online retailers as well as customers. A current survey by the information service provider CRIF shows: In Germany alone, 94 percent of the online shops surveyed stated that they had already been confronted with attempts at fraud in the past. The resulting damage is even estimated by experts at up to 1.8 billion euros.
In addition to data theft, identity theft and the sale of plagiarism are particularly popular, which can quickly make even honest online shoppers victims. But now there is new hope: With artificial intelligence (AI), specialists can track down the scammers more quickly.
Identity theft is particularly difficult to detect. The fraudsters disguise themselves with a trustworthy customer history of an existing customer and appropriate the customer’s buying behavior. This is how scammers often manage to trick manual fraud prevention programs.
$48 billion fraud loss: AI could break trend
Riskified manager Peled is convinced that there have never been so many attempts at fraud as there are now. Globally, too, the number of affected online retailers is over 90 percent. The American market research company Juniper Research estimates the total cost of e-commerce fraud for retailers in 2023 at around 48 billion US dollars. In 2022, the losses were still around 41 billion US dollars, in 2020 only 20 billion US dollars.
But advances in the application of artificial intelligence could now break this trend: “AI-based fraud detection systems make it possible to process huge amounts of data and to detect irregularities and potentially fraudulent behavior that are simply invisible to the human eye,” explains Peled.
Payment service providers such as Checkout.com, Stripe and Paypal, but also anti-fraud platforms such as Red Points or Riskified are already using these new AI methods. Huge amounts of data are analyzed in real time and fraudsters are identified.
This can even uncover the dreaded identity theft: Using AI, the specialists recognize relevant deviations in the customer’s behavior and, in combination with all other available data points, develop an accurate risk indicator for existing customer abuse. This is not possible manually in real time.
Margin of error of the AI is vanishingly small
AI-based fraud prevention programs evaluate more than 1000 characteristics of a potential transaction within 100 milliseconds to determine the likelihood of fraud while enabling secure transactions at the same time. The customer does not notice the fraud check in the purchase process and despite the real-time check, the error margin of the AI programs is already negligible today.
To date, two types of algorithms have primarily been used to combat fraud: on the one hand, artificial neural networks (ANN), which imitate the structure of the human brain and link existing knowledge with new data. With the help of ANN, fraud data, biometric data and fraud patterns can be evaluated in real time, on the other hand with the help of methods based on complex decision tree networks such as Gradient Boosted Trees, which ask specific questions and based on the answers, the AI can make a decision.
Frank Schlein, Managing Director of the financial service provider CRIF, says: “The success of all AI solutions in e-commerce is based on the quality of the data used.” According to Schlein, however, it is not possible to say exactly which and how much data is required for reliable fraud detection. The basic rule is: The algorithm should be fed with as much data as possible.
>> Read also: “Data is more important than artificial intelligence”
Important data provide deviations in the purchasing behavior of customers, unknown delivery addresses and suspicious payment methods, but also suspicious IP addresses, account data already associated with fraud or customer profiles. The AI can evaluate this mass of information in milliseconds and thus convict the fraudsters more quickly.
Monthly fraud prevention training
In order to keep the margin of error as small as possible, the AI must be trained regularly. This is what feeding the AI with new data and fraud patterns is called. Ido Lustig, Product Manager of the payment service provider Checkout.com, says: “We train our algorithms at least once a month to be up to date.”
The AI can already recognize fake reviews by analyzing the language used and make payment processes more secure through biometric identification. Thanks to the AI, the fraudsters’ data is then automatically transmitted across platforms to financial institutions and law enforcement authorities. This not only protects the dealers, but also the customers.
Because every attempted fraud discovered is reported and can be punished as a result. Customers whose data has been stolen or whose customer accounts have been compromised can block their cards and implement additional security measures such as two-factor authentication. In the future, all accounts could be additionally secured by customers’ biometric data such as fingerprints or face scans.
Thanks to the use of artificial intelligence, Riskified was able to secure 71.6 million euros in fraud attempts within 16 months. This is particularly pleasing to Riskified’s customers, who include the camera manufacturer GoPro.
Retailers like Snipes or Zalando and brands like Bonprix, Hugo Boss and Puma also use AI solutions. Because they perceive an increase in fraud in online trading. Bonprix already introduced its AI software Fraud Detection in 2019. E-Commerce Managing Director Markus Fuchshofen says in retrospect: “Twelve to 15 percent of the cases would not have been discovered in the first place without fraud detection.”
Puma relies on AI fraud prevention
The German sports goods manufacturer Puma has been using AI-based software to check credit cards and payment systems since April 2023. Puma’s Head of E-Commerce Jackson Fernandez also justifies the company’s decision with the success rate of the AI programs: The proportion of transactions that cardholders reported to the credit card companies as fraudulent has fallen to a consistently low level since the software was introduced , he reports.
>> Read also: Internet Fraud – How to Report It
The company had previously used Red Points’ AI software to identify counterfeits of its own products on marketplaces. Red Point’s algorithm now scans 65 marketplaces and platforms worldwide and checks whether counterfeit products are in circulation. In this way, Puma has already pulled $500 million worth of goods from stores in 16 months. No wonder Real Madrid, Fila and Hugo Boss also use the software from a Spanish start-up. In total, around 4.6 million illegal products are said to have been discovered every year thanks to Red Points.
However, AI programs are still most successful today when they are supported by human experts. While AI makes it possible to identify fraud patterns, human experts can take into account the complexity of the situation and customize the algorithm for the use case. To this end, the data experts are developing AI-supported decision logic to ward off attempted fraud depending on the product. Because the purchasing behavior of customers when it comes to travel or digital goods varies greatly.
More: When the cleaning machine thinks for itself