Why Manual Review is Key to Retailers’ Defense Against Fraud, Part 1

Preventing fraud and chargebacks is a constant challenge for online sellers, especially as card-not-present (CNP) fraud costs keep rising. Fraud-detection technology is evolving fast to keep pace with e-commerce growth and the increasing sophistication of organized fraud attacks. However, one element of online fraud prevention still requires human expertise — manual review of orders flagged as possible fraud.

Some merchants now view manual review as old-fashioned — too slow and costly to be worth implementing. But manual review has evolved. It’s faster and more cost effective than in the past, and the need for manual review is greater now than ever. In this first part of a two-part series, I’ll look at why manual review matters so much now and how it keeps pace with the speed of e-commerce.

Why Use Manual Review for Fraud Prevention?

Because it requires human expertise, manual review of flagged orders is one of the most expensive pieces of the fraud prevention program puzzle. However, the costs of operating without a manual review program can be much higher. Here’s why: Algorithms have gotten very good at flagging possible fraud, but they still generate false positives — i.e., orders that look like fraud because of the location, delivery address, shipping speed, or some other factor, but are legitimate. The problem is that good customers, especially wealthy ones who shop while they travel for work or pleasure, also place orders that fit this profile. Without an expert to examine the order more closely, merchants can lose revenue and alienate customers.

Right now, this is one of e-commerce’s biggest problems. False positives cost U.S. card issuers an estimated $331 billion in 2018, up from $264 billion in 2016. That’s far more than the cost of completed CNP fraud, which Juniper estimates will total $130 billion over the next four years.

Losing more money to false positives than to fraud is a major problem for the industry. For individual merchants, it creates additional problems. False positives create bad customer experiences and diminish customer lifetime value. Even one bad experience will drive as many as eight of every 10 customers to a competitor.

The solution to this problem is manual review of each flagged order. Whether to conduct manual reviews in-house or outsource them depends on the merchant’s order volume, seasonal sales peaks, and internal fraud expertise.

What Manual Review is Like Now

Manual review isn’t just someone looking at an order, reviewing files, and making phone calls to confirm data. Technology has made manual review increasingly precise and efficient. For example, group analysis batches orders by specific criteria to allow one fraud specialist to review multiple orders at once. Another current approach is double (or triple) validation, in which the same order is analyzed by two or three analysts in parallel, and their results are compared to make a final decision.

Large organizations with many reviewers can use best-fit analyses in their manual review process. Best fit factors in the regional expertise and risk tolerance of each reviewer to assign orders for review. Some of these programs use machine learning algorithms to optimize review assignments. In all cases, the resources available to reviewers have expanded in recent years. Social networks, link analysis and data visualization now provide continuously updated data for reviewers to use.

Making Manual Review Work for Real-Time Decisions

Even the fastest manual reviews require some time to complete. What can they do for sellers that require real-time decisions? For purchases like digital content, event tickets or on-demand delivery of groceries, manual review is still needed to prevent false declines and stop fraud, but delays on order decisions will drive customers away.

It's possible to build an in-house manual review program that reduces false declines on real-time decisions. This is contingent upon the merchant having the resources and internal expertise to implement two additional programs. The first is damage control, and the other is control group review of automatic declines.

How do these programs work? I’ll go into further detail in part two of this series, but here’s a brief overview:

• Damage control allows a merchant’s customer service team to add customers to a temporary whitelist if they challenge a declined order. Then the customer or the customer service representative, depending on the decision flow, can resubmit the order for manual review. The obvious risk here is that fraudsters can also contact customer service. Therefore, damage control procedures are advisable only for merchants with access to very well-trained manual reviewers who are versed in the most current fraud tactics.
• Control groups reviews don’t involve customer interaction and don’t carry the risk of temporarily whitelisting a fraudster. In a control group program, reviewers analyze random batches of automatically declined orders to identify false positives. This allows merchants to learn and track their false positive rate, and it also surfaces false positive data that can be used to refine and improve fraud-detection algorithms.

We’ll look more at the interplay between manual review and better machine learning in part two. The takeaway for now is that manual review is key to reducing false declines while preventing fraud. As manual review programs harness technology and a variety of datasets, they’re getting faster and better at separating legitimate flagged orders from fraud. That can help merchants keep revenue and customers they would otherwise lose to false positives.

Rafael Lourenco is executive vice president at ClearSale, a card-not-present fraud prevention operation that helps retailers increase sales and eliminate chargebacks before they happen.