Every day, goods worth more than €11 million are stolen from stores in Germany. But it’s not just brick-and-mortar retail that’s affected – online crime is also rising rapidly. Artificial intelligence (AI) can help detect and reduce fraud.

For decades, retailers have been actively combating shoplifting. According to a study by a retail research institute, retailers lose around €4.1 billion to theft every year. While this is far from insignificant, it represents only a small fraction of total net retail revenue, which reached €563.6 billion in 2023.

In e-commerce, however, the threat to revenue is much greater. The U.S. market research firm Juniper Research estimates that global e-commerce fraud cost merchants around $48 billion in 2023. While precise figures for Germany are not available, estimates suggest that e-commerce businesses lose about 3% of their annual revenue to fraud.

How AI Can Reverse the Trend

Advances in AI could help curb this trend. AI-based fraud detection systems process vast amounts of data and evaluate hundreds of transaction attributes within milliseconds to determine the likelihood of fraud, without the customer noticing and with minimal error rates.

It doesn’t just look at individual transactions. AI can also detect unusual user behavior over time, such as when customers typically shop or where they access services from.

Identity theft can also be uncovered this way. AI detects deviations in customer behavior and combines them with data points such as purchase history and transaction metadata. This makes it possible to develop precise risk indicators that can predict whether a transaction is likely to result in a chargeback.

Payment service providers are already using these AI methods successfully. At Unzer, we use in-house AI models to identify suspicious behavior early on. Our internal risk platform is AI-driven, featuring a risk engine that can detect fraudulent payments in the Buy Now, Pay Later (BNPL) segment. In 2024 alone, our AI blocked more than 30,000 payments worth over €10 million in this segment. In the future, we plan to expand AI-based risk assessment to other payment methods such as credit cards and SEPA direct debit.

Device Fingerprinting and Behavioral Biometrics

A promising technology in this field is device fingerprinting, which is also used at Unzer. This method identifies a device based on various characteristics derived from its settings and usage. The exact data collected for the “fingerprint” can vary depending on the provider.

Another highly promising, though less widely adopted, approach is behavioral biometrics. This leverages the fact that a person’s typing patterns are as unique as a fingerprint. Unlike traditional biometrics, which rely on physical traits such as facial features or fingerprints, this method analyzes behavioral patterns.

Depending on the device, this may include typing speed, keystroke pressure, smartphone orientation and movement, or which areas of a keyboard are used. These data points help verify a user’s identity and prevent fraud.

AI Makes Payments Safer and Fairer

AI doesn’t just make payments more secure; it also makes them fairer. The reason is that, unlike traditional fraud detection systems that rely on rigid rules, AI-driven risk engines learn from past transactions.

Three examples illustrate this:

- A credit card issued by a Nigerian bank may be unusual for a German merchant, but that alone doesn’t indicate fraud.  
- Someone ordering sneakers at 2 a.m. might work night shifts, be returning from a party, or simply have trouble sleeping.  
- A person living in a high-risk area can still be creditworthy.  

In such cases, AI can usually make more nuanced decisions than rule-based systems, which tend to block too many transactions just to be safe.

The Cost of “Friendly Fraud”

That said, fraud can never be completely eliminated. One of the hardest types to detect is so-called “friendly fraud.” This occurs when customers dispute legitimate transactions, claiming they didn’t authorize a purchase, didn’t receive a product, or that it didn’t match the description.

Examples include a child using a parent’s credit card without their knowledge, or customers requesting refunds even though they received the goods on time.

Friendly fraud costs merchants worldwide more than €100 billion, and more than one in three customers admits to having requested an unjustified refund at least once. The most frequent offenders are over 45 years old; a group often under significant financial pressure and stress, and therefore more likely to justify fraudulent behavior.

Understanding Risk Is Key

Merchants need to understand their e-commerce risks just as well as those in physical retail. Knowing their core markets and customer base is essential.

Security measures such as Strong Customer Authentication (SCA) and transaction monitoring by payment providers like Unzer can help reduce risks.

Personal contact, such as a phone call, can also be valuable, not only for resolving suspicious cases but also for strengthening customer relationships. This is especially important for products that are more prone to friendly fraud, such as digital goods. Close cooperation with payment processors is highly recommended in these cases.

Conclusion

In summary, artificial intelligence plays a crucial role in making online commerce safer. While fraud can never be completely eliminated, AI helps detect suspicious activity early and significantly reduce its impact. As such, adopting AI technologies is a key step toward a more secure and trustworthy e-commerce environment.