With Christmas approaching, online retailers are facing an onslaught of fraudsters, and machine learning could prove vital to pattern detection.
Stripe has revealed insights from its data to help online businesses fight online fraud this Christmas, and has recommended that retailers add machine learning tools to their respective arsenals.
Stripe examined transaction data from hundreds of thousands of its customers across 25 countries.
‘We recommend using anti-fraud tools based on machine learning trained on large amounts of data, to ensure businesses are making the right trade-offs between battling fraud and maximising profits’
– MICHAEL MANAPAT
While chip-enabled credit cards have made bricks-and-mortar shopping safer, fraudsters are increasingly targeting online stores.
However, unlike physical stores, online businesses are unfortunately responsible for paying the associated costs.
On average, every $1 of a fraudulent order costs an online business an additional $2.62.
The key to fighting back against fraudsters is the use of machine learning to spot patterns, said Michael Manapat, engineering manager for payments intelligence and experience at Stripe.
“While there are some consistent patterns to fraudster behaviour – for example, their high-purchase velocity, their propensity to work late at night and their desire for cheap or immediately deliverable goods – we’ve found that the predictive strength of these patterns varies widely depending on the location of the business and the fraudster,” said Manapat.
“Because of this, we recommend using anti-fraud tools based on machine learning trained on large amounts of data, to ensure businesses are making the right trade-offs between battling fraud and maximising profits.”
Stripe revealed that fraud rates vary by a factor of two or three based on the country where the credit card is issued. Purchases from cards issued in Argentina, Brazil, India, Malaysia, Mexico and Turkey are particularly fraudulent, although those issued in the US, Canada and France are also susceptible.
‘The most effective providers leverage global datasets from hundreds of thousands of other businesses to train their machine-learning algorithms and identify even subtle fraud patterns’
– JORDAN MCKEE
Stripe, the e-commerce payments platform created by Irish brothers John and Patrick Collison, said that the highest online fraud rates occur during days and times when many people aren’t shopping, such as Christmas Day or late at night. Fraud rates as a percentage of overall traffic increase in the summer and in late December, but not on heavy shopping days such as Black Friday, as might be expected.
An interesting fact about fraudulent transactions is that they are often small. This is surprising, given that fraudsters are not paying for the products they buy.
In the UK, Stripe data shows that fraudulent transaction amounts are only slightly larger than regular transaction amounts. However, in many other countries, such as France and Singapore, fraudulent transactions are significantly larger than normal.
Fraudsters give themselves away by making rapid additional charges at the same businesses on the same credit card, initiating repeat purchases 10 times more quickly than legitimate cardholders.
Typically, fraudsters prefer products that don’t need to be delivered, can be delivered to locations like public buildings or parks without raising flags, and can be obtained quickly before transactions are invalidated. These considerations can explain the prevalence of fraud among on-demand services as well as low-end consumer goods.
“It’s crucial for online businesses to have robust fraud defences, especially during the busiest shopping season of the year,” said Jordan McKee, principal analyst at 451 Research.
“Because online fraud is highly complex and increasingly global, merchants should consider outsourcing fraud tooling to trusted third-party providers that have access to large and robust data sources.
“The most effective providers leverage global datasets from hundreds of thousands of other businesses to train their machine-learning algorithms and identify even subtle fraud patterns.”