There has been a lot of buzz in the payments industry about machine learning being a great tool to automate fraud prevention. This is when a computer is programmed to use data to recognize patterns and identify potentially fraudulent transactions in real time. It seems only natural for companies accepting digital payments to want to take advantage of this type of program in order to bolster fraud prevention efforts. However, machines can’t completely eliminate the human element needed to ensure that your business doesn’t process – and then lose revenue because of – fraudulent transactions.
How does machine learning work?
Machine learning analyzes your past transactions and predicts future outcomes based on patterns and behaviors. The more data the machine learning system reviews, the more accurate its predictions will be.
There are three different types of machine learning.
1. Supervised – Supervised machine learning is when the program is given the data and the answer it needs to find, so it can learn how to make accurate predictions. The computer program is taught to identify specific patterns and to take note of something that deviates from that norm, such as a payment from an obscured IP address.
2. Unsupervised – Through unsupervised machine learning, the computer processes raw data and identifies patterns without any assistance from examples or human interference.
3. Reinforcement – With reinforcement machine learning, the computer arrives at conclusions on its own and receives a “reward” if the answer is correct. The algorithm then adjusts future results based on that reward feedback. Rewards encourage more of that same result, while no reward discourages false answers.
Each type of machine learning has its benefits and its appropriate application in the payment verification process. However, all of them are missing the most important element to effectively prevent payment fraud: the trained human eye and experience.
Human checks are vital for fraud prevention
A fraudulent transaction, or a falsely blocked legitimate one, could mean a costly mistake for your business.
If a fraudulent transaction is discovered and flagged after too much time, your goods and revenue could be long gone by the time the false charge is identified and reported. On the other hand, incorrectly identifying a legitimate transaction as fraudulent could present a difficult customer service situation, or the customer could simply give up and go to a competing company.
A quick and smooth payment process is important for customer satisfaction, retaining customers and increasing profits. According to Offer Gat, chairman of Direct Pay Online, “The cost of acquiring a new customer is five to 10 times more expensive than retaining an existing one, so this activity has a significant impact on your long-term revenue streams.”
Three machine learning vulnerabilities
There’s very little room for error when performing these important checks. With your revenues on the line, machine learning is still too vulnerable to handle them for the following three reasons.
1. Machine learning only works when there’s enough accurate data.
The more accurate information a machine learning program is given to review, the better it works. Without enough accurate data, machine learning – whether supervised, unsupervised, or reinforced – can make some crucial mistakes.
Employing a team of risk-management professionals is the only way that you can be sure that the machine learning is working the way it’s supposed to. Risk-management professionals not only have the sufficient and correct data needed to make machine learning work, but they also have the experience to recognize and flag fraudulent transactions.
This is why the most experienced and vigilant payment service providers still use trained and experienced risk professionals to perform manual checks and manage payment fraud prevention.
2. Machines need rules.
Fraud prevention methods rely upon a set of rules that help identify fraudulent transactions, such as geo-location identification, IP tracking and AVS. Machine learning mechanisms need to follow those same rules to provide the necessary information.
Some scenarios, such as when payments are made from one account but originate from different devices, could be flagged by machine learning programs because this transaction could technically break a programmed rule. However, with the flexibility of mobile devices and the increase of on-the-go consumerism, some of those situations may be completely legitimate.
Machine learning can’t be relied upon to catch such a distinction; only an experienced risk-management team can determine if these transactions are legitimate and if they should go through. This ensures that the highest amount of valid transactions get approved and aren’t rejected by the computer program.
3. Humans still need to check machine learning algorithms.
With machine learning, the data must be accurate and the rules must be set properly for it to work. But even then, when legitimate transactions don’t match what the machine has “learned,” there is a greater chance of error, and the algorithms are likely to make mistakes.
An employee’s trained eye needs to identify those issues, whether in the results or in the data input process, and communicate to the machine learning system that there has been a mistake.
Because this adds another step to the transaction review process, companies should depend on the experience of a payment service provider’s trained risk-management team to help consolidate resources and speed up the authorization process of legitimate transactions.
With the advent of – and disadvantages to – a machine-learning approach, online payment solutions with a human-driven risk-management division are the safest and most secure mechanisms for thorough fraud prevention. While machine learning has some potential, it can’t yet be relied upon to deal with the complexities of effectively protecting your business’s revenue.
Eran Feinstein is the founder of Direct Pay Online, a global e-commerce and online payments solutions provider for the travel and related industries. With over 14 years of experience leading technology, sales, marketing and operation teams, Eran is an authority in the East African e-commerce and payments arena.