In the world of online commerce, the number one consideration for any new merchant is to get sales through the door, while at the same time growing business and increasing net earnings. Fraud prevention may not be a key consideration upfront but as the business grows and evolves, merchants realize that the risk of fraud also grows.
You may instinctually ask fraud prevention solution providers and industry experts alike: what then is the quickest and easiest way to approach my business’ fraud prevention strategy? What is the silver bullet approach to preventing fraud? And, you will probably hear a number of different perspectives.
There are generally three approaches to fraud prevention that have evolved over the last decade: people, rules and machine learning.
The people approach
No one knows your business like your own team. For many merchants the most cost-effective way to offset the cost of fraud is to hire a dedicated team of analysts to monitor traffic through their website. The team can use internal resources and the history with the customer to determine the validity of a transaction. They can also verify the purchase or cancel orders manually if they are deemed suspicious. For instance, the international retailer Urban Outfitters had a dedicated fraud team of 17 fraud analysts in 2009. They started their fraud prevention strategy on the basis of a “high order review rate to stop fraud.” This is the traditional approach common to all form of trades - from the cashier at the till dealing with the physical customer and payment, to the fraud analyst on the online platform looking out for unusual behaviour. People must apply themselves to analyse fraud and to develop tactics to be one step ahead of the fraudster. While this approach is effective on a case-by-case basis, it is not scalable enough for merchants to use it as the business grows.
The rules-based approach
A person reviewing orders can likely look at up to 140 orders per day. To manage the flow of reviews, your company may have to choose between reviewing orders based upon value and reviewing all orders.
When the transaction volume grows, this is not sustainable. It then makes economic sense to introduce a system-based rules approach because this approach enables merchants to increase the sophistication of controls. These controls ease transactions for legitimate customers and facilitate the identification of risky behaviour. Rules can automate the best practices leveraged by manual review, enabling people to focus on more complex cases that require authentication or research. A rules-based approach allows your team to work more efficiently and also enviably reduces false positives.
Yet again, as volumes keep growing, your company may slowly add more analysts to review orders until the company seeks optimizations in decisioning above and beyond rules.
The machine learning innovation
Rules-based systems evolve naturally to leverage statistical models within risk decisioning. Statistical models enable multiple factors to be weighed simultaneously versus in sequence as in rules. They also allow the review of interactions between factors. Statistical models have proven to be very efficient with the rapid growth of ecommerce and the parallel increase in attempted fraud.
To further improve profitability of ecommerce, our fraud team has moved from a rules-based approach to machine learning statistical modelling,” says Brian Whitney, Director Contact Centre at Urban Outfitters Inc. We experienced immediate results after the model went live.” Urban saw a drop in order reviews to about 20% and false positives shrank by 8% helping to reduce ecommerce order cancelations. The fraud analyst team has been reduced to 7 people. With increased computer powers, the factors a merchant can capture and leverage to make a decision have also increased. Today, machine learning is able to continue the advancement in sophistication of decision making as it enables merchants to evaluate the high volume of characteristics available per order. The sophistication of statistical techniques supports real-time risk decisioning on a multitude of factors. Machine learning enables a near replication of the typical behaviour of the consumer as well.
People, rules and statistical models work best when combined together
Having said that, no single approach has proven to be the most effective in isolation. People can understand operations and fraud trends that today’s rules and statistical models cannot. Rules can be used to apply business policy decisions consistently and systematically, a thing which would be difficult to scale using people.
Statistical modelling, such as machine learning, enables the use of sophisticated algorithms to detect and rank order risk on a scale that people and rules cannot. Machine learning is best used in conjunction with your people and your rules for the most refined risk decisioning. Without these two key strategies, a machine learning model will not be as effective – and certainly will not be your silver bullet to fraud prevention.
About Andrew Stolz
Accertify’s very first Account Manager, Andrew Stolz is responsible for the day to day management of the European Account management team. He has over 9 years of fraud and ecommerce experience, holding fraud analyst roles at Viagogo and Arcadia before joining Accertify.
About Accertify
Accertify Inc., a wholly owned subsidiary of American Express, is a leading provider of fraud prevention, chargeback management, and payment gateway solutions to merchant customersn spanning diverse industries worldwide. Accertify’s suite of products and services, including machine learning, help ecommerce companies grow their business by driving down the total cost of fraud and protecting their brand.
For similar stories, please check out our Web Fraud Prevention and Online Authentication Market Guide 2016/2017 here to get access to an insightful outline of the global digital identity and web fraud ecosystem.
Every day we send out a free e-mail with the most important headlines of the last 24 hours.
Subscribe now