Defeating fraud is going to be a long war - that much we know. At all parts of the customer and payments funnel it is important that we deploy the most successful approaches to make it as difficult as possible to defraud a merchant. PSPs and payment gateways use a wide variety of techniques to reduce fraud. Yet, sat where we are, working with merchants, we know that these techniques are not fully successful at stopping fraudulent transactions.
However, while it is fair to say that no single approach alone is going to be completely effective, we strongly believe that applying machine learning techniques is the best approach to reducing the maximal amount of fraud at the greatest scale.
This will benefit not just merchants, but will be a significant competitive differentiator for PSPs who adopt it successfully - a differentiator with the potential to be charged at a premium to the merchant in return for the reduction in chargebacks due to fraud. Here’s how.
Machine Learning at a per transaction cost
The most progressive payments companies like Paypal, Apple Pay and Amazon Payments all put great store in the security of their networks. They believe it attracts customers and users and generates trust in the market. And what is it that they have in common? They all use machine learning techniques to give them a better, faster, more scalable solution for fraud detection.
For many companies seeking to attract, retain and motivate the requisite skillsets needed in an effective in-house data science team to address fraud is a difficult and expensive undertaking. Hiring in consulting skills for a one-time build is perhaps easier but probably even more expensive and does little to resolve the ongoing maintenance problem. For this reason Ravelin has worked to develop a PSP partner program that will allow payment companies to use our machine learning and graph network techniques on their own data sets and do it at a very low per transaction price that will scale with their business.
PSP-specific models for outstanding fraud detection
The dataset that PSPs have access to is less rich than that which is available to a merchant because typically they only see the payment page activity, and not all the events in the customer journey. However in the the models we have run on payment hosting page data alone we have seen some very reliable predictions of potential fraud. This has encouraged us to reach out to PSPs in order to help them reduce the number of chargebacks at source. We also know that each PSP requests slightly different data and operates in slightly different ways. We are confident therefore that with full access to a specific PSP’s dataset we can make significant improvements to the general PSP model and provide really interesting and positive results.
For forward-thinking PSPs, this offers a really fast, cost-effective means of adding best-of-class algorithms to their fraud detection portfolio. Adding this capability to the many existing fraud detection and reductions methods PSPs have in place already offers a real chance to differentiate on fraud, to bring a clear value-add to merchants and more importantly, make life considerably more difficult for people determined to defraud online businesses.