How Machine Learning Can Deliver Faster—And More Secure—Payments

With consumer expectations higher than ever, analytic tools that can meet those demands while weeding out fraud are more important than ever, says Luke Reynolds.

A recent survey conducted by Juniper Research predicted the overall value of fraudulent online transactions globally will reach $25.6 billion by 2020. Of that activity, retail fraud will account for 65%, while banking fraud will contribute 27%.

This significant jump (up from $10.7 billion in 2015) points to the increased sophistication of fraudsters. Sadly, as consumer, e-commerce, and banking technology advances, it is making it even easier for fraudsters to get better at account takeover, synthetic identity, and creating havoc generally.

So what products can aid banks, merchants, and processors in the fight to better safeguard their customers’ collective data? One of the most significant technology solutions employed today resides in machine learning, or adaptive analytics.

The traditional, rules-based model for fighting fraud creates an environment where consumers must experience fraud first, then businesses analyze and adjust their rules or models for the next attack.

These systems are an excellent method of blocking certain types of fraud. But taking an advanced, layered-risk approach can reduce your manual intervention, block fraud immediately as it occurs, and actually create a stronger customer experience.

Using the Right Data

Machine learning is a form of intelligence that allows your risk-management platform to understand each individual customer touchpoint and actively detect and recognize patterns in a customer’s purchasing journey. It relies on understanding and reacting to behavior in the same way that we do as humans.

Human beings are innately in tune with each other’s behavior. We can tell a local from a tourist in a coffee shop simply by observing how they look around and noting their body language. Machine learning works in a similar way. It analyzes each customer’s behavior in real time, enabling organizations to quickly and accurately detect the subtle anomalies that indicate someone is acting out of character.

By incorporating this type of learning into your risk-management strategy, you are teaching a machine to separate the signals of a legitimate consumer’s behavior from those of a fraudster.

This approach enables machines to make autonomously data-driven decisions in place of being manually programmed to perform explicit tasks. After being exposed to new data, machine-learning programs can enhance themselves over time. No surprise this technology has recently become the center of many technological advances in the payments landscape.

According to a report released by Javelin Strategy & Research, fraud claims a new victim every two seconds, and many of those fraud incidents involve credit cards. In an environment where transactions can take place in milliseconds, identifying and preventing fraud requires more than just manual monitoring. Machine-learning providers are working steadily to innovate their layering technology, aiming to capture fraud without increasing false-positive rates.

Financial institutions, issuers, and acquirers are finding that machine learning has not only helped their customers by improving their risk strategy, but also by reducing their operational costs. These are not just operational costs associated with fraud and chargebacks, but also those stemming from customer service.

For example, if an issuer or acquirer has a low risk appetite, this typically leads to a higher false-positive ratio, which can then lead to good customers having their cardholder accounts temporarily suspended or merchant funds held, pending investigation. When these are actually genuine transactions, this process leads to customer calls and emails criticizing the experience.

With any risk-mitigation tool, you are only as good as your data. Some tools take a unique approach to data by focusing more heavily on good behavior versus bad behavior, assuming fraud will be the exception and not the rule. This approach allows merchants, acquirers, and financial institutions to develop a larger pool of good-behavior data than bad-behavior data. The advantage of this approach is that the models will learn each customer’s character traits more quickly compared to models that focus on fraud labels as their data source.

Many issuers, acquirers, and processors are now seeing a practical application for these analytics tools when layering them over an existing fraud solution and are partnering with machine-learning experts to create this approach.

Return on Investment

At a time when consumers are demanding a friction-free experience, faster checkouts, and speedier approvals, merchants and processors must use the necessary analytic tools to stay ahead of fraud.

The adoption of machine-learning solutions is helping reduce the amount of fraud and chargebacks, improve operational efficiency, and reduce customer friction. Companies that are actively embracing this technology are already seeing a big return on investment.

In the long run, it is widely expected that machine learning will continue to serve the payments industry as an incredible resource for helping financial institutions operate in a safer, more efficient environment.

Routine tasks once handled solely by humans can now be performed in conjunction with machines, allowing us the ability to capture fraud as it occurs, learn more rapidly as fraudsters become more complex, and, ultimately, enhance the consumer experience.

—Luke Reynolds is chief product officer at Featurespace Ltd., Cambridge, United Kingdom.