Blog / machine learning
What’s the most important factor to identify fraud?
Our survey results reveal which factors merchants use to identify fraud, and how these differ between markets and businesses. What can these differences tell us about fraud detection?
What’s the most important factor to identify fraud? Every business will have a different answer. With fraud risks still growing amid the pandemic, and constantly changing customer behaviour, it’s increasingly difficult for fraud teams to separate fraudsters from genuine customers. As no two businesses are the same, there isn’t a one-size-fits-all approach. Every business has its own customers and fraudsters, as well as individual buying journeys, checkout experiences and so on.
To understand this more, we surveyed 1000 fraud professionals from large online merchant businesses around the world about which factors they see as the key indicators for fraud. We spoke to merchants in four industries - travel & hospitality, retail, marketplaces and digital goods.
As we can see, there’s not a single number one factor for all merchants, with around a third of customers naming customer profile and order content as the most significant indicator, and the remaining third split between location, device ID and shared industry data. It’s key to note that shared industry data is the least significant factor, with only 40% of businesses placing it in the top three categories. This reflects the importance of merchant differences when it comes to customer behavior and fraud.
There are subtle differences between the industry sectors, and even between individual merchants, which reveal that a blanket approach to fraud detection is flawed. Let’s look at some of the reasons behind this...
Location data is more important to marketplaces
For marketplaces, 64% name location data as one of the top three fraud identifying factors, compared to 62% for digital goods, 58% for retail, and only 57% for travel and hospitality merchants.
Many online marketplaces have thrived amid the pandemic, as Amazon sees explosive growth, eBay’s US sales grew 22% in 2020, and Etsy’s stock tripled last year with custom masks flying off the shelves. But, with increased demand and transaction volumes comes a greater risk of fraud.
Monitoring location data is important to online marketplaces that facilitate multiple sellers shipping products to customers. Certain addresses can be flagged as risky if they are known pick-up spots for fraudsters, or if the billing and shipping addresses look suspicious - they don’t match, are unusually far apart, and/or frequently change.
Food delivery marketplaces can also be subject to ‘pizza plug’ scams. In this case, a fraudster advertises food deliveries at bargain rate, then makes fraudulent online orders for food and arranges delivery to consumers. Multiple address changes or bulk ordering to a suspicious address would be a red flag for fraud.
Order content is more significant to retail merchants
Order content is a more important factor to retail merchants with 34% naming it as the number one factor, compared to 30% in both travel & hospitality and marketplaces, and just 24% in digital goods.
Last year was record-breaking for retail ecommerce, as online sales grew by 46% in the UK and accounted for 21.3% of total retail sales in the US. Retailers were forced online or risked closing their doors for good - the accelerated move to digital left unprepared retailers vulnerable to fraud.
Retail merchants, by nature, involve selling physical consumer goods. Fraudsters look to buy high-value items or buy in bulk to resell. This makes it easy to flag high-risk orders according to order value and velocity: an order for ten $300 dresses is more suspicious than a $5 t-shirt.
Plus, order content often varies more for retail businesses than digital goods and/or subscription businesses in which order content and value will often be consistent.
Alongside industry trends, every business will prioritize different fraud risk factors
Every business is unique, even within these wider industries, and even the slightest difference might affect how you should monitor your fraud, including:
Business maturity: Fraud monitoring priorities often change as a business matures. Newer businesses may start out only monitoring locations, but as a business grows, a more sophisticated approach is required to avoid impacting legitimate users.
Checkout time: The standard checkout time of genuine customers varies from business-to-business but it’s an important fraud indicator. Genuine customers tend to spend longer on checkouts thinking about a purchase than fraudsters who often copy & paste card details.
Region: Where your business operates impacts the type of fraud you will see.
Monitor everything to learn what works best for your business
In our Industry leaders' insights on fraud and payments webinar, Florian Jensen, Global Fraud & Payments Director at Glovo advises merchants to “use all the data points possible and find out what works for you.” Machine learning models are useful for getting an overview of all of your data points and what fraud trends you’re seeing.
Ricardo Ferreira, Data Scientist at Ravelin says "it's not feasible for fraud teams to just look at all of the data and guess patterns. But a model can highlight the most important fraud factors and recognise a genuine from a fraudster - it’s the easiest way." Our approach is to build machine learning models based specifically on your customer data, so that it understands your customers and fraudsters. To learn more about our bespoke fraud solution, contact us here.