Thursday 26 September 2019 5:08 am

Ecommerce needs a new approach to fraud prevention

Assaf Feldman is co-founder and chief technology officer of Riskified.

In 2018, global ecommerce sales reached £2.27 trillion – a staggering number, and an 18 per cent increase over 2017.

That growth will continue, as ecommerce reaches new markets and consumers become more comfortable with shopping online. 

It’s a huge opportunity for retailers, but there’s a risk that threatens to hold it back – fraud. 

Ecommerce purchases are, by definition, “card not present” (CNP) transactions; your credit card is not physically with the retailer when it’s charged. This is an important distinction, because it shifts the liability for fraudulent transactions to the retailer.

When fraudsters obtain stolen credit card information, they buy items online for personal use or resale. Those charges are typically noticed by the legitimate cardholder, who then files a dispute with their bank, which then takes the money out of the retailer’s account to credit the customer. 

It’s the correct outcome for the customer, but it means that the retailer takes the loss. The order has been fulfilled, and the fraudster isn’t returning the goods.

CNP liability means that retailers have to carefully consider which ecommerce orders to approve. If they’re too lenient, they risk major losses to fraudsters. Too cautious, and they’ll decline good orders and frustrate their customers. 

Many retailers choose the seemingly safer option – in 2017, the average ecommerce retailer turned away 1.5 per cent of revenue due to fear of fraud. 

But even that understates the impact. Unnecessarily declined customers often go to a competitor and are less likely to return to the retailer that initially rejected them. The lifetime impact of an incorrect decision may be many times larger than one lost sale.

To combat this, retailers are trying a number of approaches to fraud. 

They may use basic analytics to assign a fraud “score” to each order based on perceived risk, and then approve orders that score well and decline those that score poorly. Or they may hire analysts to manually review orders that show risky characteristics. Some may even contact you and ask for a picture of your driver’s license and credit card. 

The challenge is that none of those solutions keep up with ever-inventive fraudsters. 

So with retailers stuck between a rock and a hard place, what’s the right solution? In my view, it’s technology. Machine learning, for example, can look at hundreds of different attributes across millions of orders to learn how to differentiate legitimate customers from bad actors nearly instantaneously and with extremely high levels of accuracy. 

By embracing new tech, retailers can approve more good orders, avoid fraudulent losses, and eliminate the pain of trying to verify a customer’s identity manually. And it works – by using machine learning, I’ve seen retailers decrease fraud declines by up to 80 per cent, which translated to a significant increase in revenue, while cutting fraud-management costs in half.

With fraud under control, retailers can turn their focus back to providing customers with a truly great experience. Knowing that good orders are approved and bad orders are blocked means that retailers can safely sell to anyone, anywhere, and help ecommerce realise its full potential. 

Ecommerce is a revolution in how retailers and customers interact. It’s time for fraud prevention to have a revolution of its own.

Main image credit: Getty