How to prevent return fraud with machine learning models
Discover how you can detect and prevent return fraud with advanced fraud solutions backed up by machine learning models.
Eric AlegreVP of Business Development France & UK
31 January 2022
6 min read
Some merchants around the world have reported that they have seen an increase in return abuse fraud by as much as 51% (accounting for, on average, 8% of their total fraudulent transactions). Likewise, 54% of merchants indicated they have suffered lost revenues of $5+ million annually due to return fraud. Such figures can be tear-inducing for eCommerce merchants and without adequate responses, can lead to an increase in the problem, or worse, a knee-jerk reaction from merchants trying to stamp out this and other forms of fraud affecting their businesses. Such measures may do more harm than good, as we will explain shortly. But first, what impact can return fraud have on a company?
The immediate costs to a company initially revolve around the standard procedures of items being returned, processed, shipped, potentially disposing of goods if they are no longer viable for reuse due to damage etc. But our main concern is with the internal policies a company may adopt to combat return fraud - ineffective responses can lead to a poor customer experience (UX), which in itself can impact a company’s reputation, retention rates and potential revenue growth.
Fraudsters and dishonest customers are using tried and tested methods (now in the online domain), in order to make financial gains at the expense of merchants. To tackle this problem, many businesses have chosen a number of methods to limit their impact. Such measures include:
Unfortunately, the majority of these measures often come at the expense of a positive customer experience and have a major impact on sales growth. Denying or limiting refunds in the digital age where customer focus is at the forefront of a positive UX, eCommerce shoppers can simply purchase goods through another online store. One negative experience can be damaging enough, but having permanent and unfavorable return policies in place can do more harm in the long run with customers unlikely to return if they feel they’ve had a poor UX, especially when they can find better options elsewhere.
There are, of course, other internal company decisions that are made regarding how to deal with return fraud. Manual reviews for returns can be performed, but often, merchants will base this on rule-based anti-fraud management, which lacks automation, and crucially, accurate real-time decisions. Rules will often focus on a set list of behaviors, adding users to a blacklist, however, this will not accurately identify serial abusers who make efforts to hide their tracks. The result is a slow and ineffective list of procedures that require more staff to monitor manual reviews and some refund fraud attempts may slip through the net. There is a much more efficient way to tackle return fraud.
Staggeringly, despite the rising costs of return fraud, only 3% of merchants use advanced machine learning solutions to combat the problem. As always, there is never just one reason for such a state of affairs. Merchants must weigh up the costs and benefits of implementing payment processes that may cause negative customer experiences at the checkout through intrusive authentication methods (checkout friction). Trying to maintain the payment process is one potential obstacle, but the customer care element at times outdoes any effective fraud measures. Many merchants appear to be unaware that they can ensure frictionless checkout experiences without limiting their anti-fraud capabilities.
The best way forward is to improve internal company policies, however, this process can be complemented by the use of an advanced fraud solution that can determine genuine users from fraudsters and dishonest customers. This is where the benefits of digital profiling and behavioral biometrics come into play, where a deep understanding of every single user behind payments and returns is gained, all backed up by AI/ML models that can perform the task of scanning 5000+ pieces of data automatically and in real-time, completely unseen by users. In one online luxury marketplace example, from a sample of 330,000 transactions, of which 7,000 were refunds (spanning 8 months), Nethone’s ML model was able to prevent up to 60% of fraud, rejecting only 2% of the most suspicious transactions - saving the company $3M.
The benefits of using ML models are huge and so impactful on a company’s potential continued sales growth and customer success that listing them is necessary:
If you are interested in this topic, and you wish to combat return fraud effectively with an advanced fraud management solution, we are here to help. Click 'book a call' at the top of this page or contact Eric directly via email at email@example.com or via LinkedIn.