Best practices for manual fraud reviews

Conducting manual fraud reviews is how merchants or their fraud teams check on orders to determine which are fraudulent. Unfortunately, it’s a tedious process. And that was before the 2020 global rise of e-commerce transactions — and the wave of digital fraud that came after. Yet, many businesses responded by dedicating more resources to the manual process.

48% of respondents in Kount’s “Digital Payments in 2021” survey said their companies have between three and four people on their fraud teams. And almost half said between three and four people review orders for fraud as their primary job function. A third said more than five people do this.

“As more customer interactions shift online, many fraud teams, especially those with limited opportunity to expand headcount, are struggling with an increased volume of manual reviews,” explained Brady Harrison, a Kount Senior Data Analyst. “That means people are taking on manual reviews in addition to their other responsibilities to help shoulder the load of fraud.”

If yours is among the businesses that have taken on more manual fraud reviews, follow these three best practices to make the most of the process. Then learn how to scale manual fraud review strategies with machine learning.

1. Understand the true cost of manual fraud reviews

The true cost of a fraud review includes the cost to allocate resources to the task. The average cost to a fraud review, Harrison says, is between $2 and $5 each, depending on the interaction.

“This fully loaded cost accounts for payroll, taxes, overhead, etc.,” he explained. “So an agent in the United States or Europe might spend closer to $5 per manual review. But the opportunity costs could be even higher, as these agents can have other high-value responsibilities.”

The number of interactions you review will depend on your business’s order volume. But multiply that $5 cost by thousands of manual fraud reviews, and the costs per month start to add up. Now, the time and money it takes to review orders are competing for the same time and money it takes to grow the business.

“In my experience, when businesses have to choose how they spend their time, tasks like fighting chargeback fraud, reducing false positives, and improving customer experiences get put on the back burner,” Harrison said. “So add higher opportunity costs to the mix, and you’re using more resources to protect your business than grow it.”

2. Right-size your fraud review team

If you’ve decided manual fraud reviews are the best course of action, the next best practice is to set up your team for success. That means having the right number of people on your team and allocating their time appropriately.

“There are no hard and fast rules that apply to all industries,” Harrison says, “But, in general, shippable goods merchants should be reviewing less than 10% of orders and ideally less than 5%.”

Generally, Harrison suggests that fraud team members without other responsibilities can conduct between 1,000 and 2,000 manual reviews per month or at least 50 per day. So a business that processes 10,000 orders per month might need at least one full-time agent reviewing orders.

Right-sizing is essential, given that too many reviews and too few agents can negatively impact a customer’s lifetime value. For example, a too-small team with too many reviews may falsely decline good orders to keep up. But Kount’s research has shown that many Americans won’t return to a website that declines their legitimate transaction.

Another facet of fraud review success is ensuring a lean team knows what it’s looking for when it reviews an order. Harrison suggests that many top fraud indicators are visible in a customer’s purchase history.

For example, if a customer’s information and order value are consistent with previous orders, it’s less likely to be fraudulent. But if the customer’s information isn’t consistent and the order value is significantly higher than average, the fraud risk is higher.

3. Measure your review-then-decline rate

Once you have the right number of people in place, the last best practice is to measure your company’s review-then-decline rate. Calculate your review-then-decline rate by dividing the total number of orders your agents declined for fraud by the total number of manual reviews done by your agents.

For best results, Harrison says, the ideal review-then-decline rate is between 30% and 60%. If your review-then-decline rate is less than 10%, then you unnecessarily reviewed more than 90% of the transactions you reviewed.

So if you marked 100 orders for fraud review, you could have outright approved more than 90 of them, which would have eliminated unnecessary friction and reduced operational costs.

However, if your review-then-decline rate is too high, essentially, you’re spending a lot of time and money reviewing orders you could have just declined. Moreover, declining those orders outright would have reduced the operational costs of manual fraud reviews.

Improve and scale manual fraud review strategies with machine learning

If you want to scale your manual fraud review strategies, optimize the kinds of transactions you mark for review. First, analyze all the reasons you mark transactions for review. Then calculate the review-then-decline rate for each reason.

Adjusting your business policies around reviews — for example, increasing or decreasing the minimum order value that triggers a review — can reduce the number of orders you mark for review.

However, the easiest and most effective way to improve your review rate and strategy is to incorporate a fraud prevention solution that uses advanced AI and machine learning into your business.

A solution that uses AI and machine learning can use data about your review-then-decline rate, chargeback rate, and business policies to quickly and accurately decision orders. It even helped one large retail organization eliminate more than 15,000 hours in manual reviews per year and save $1.4 million, according to the latest Forrester Total Economic Impact report.

Kount Command uses AI and machine learning to produce a safety score for every transaction. As a result, Omniscore can effectively optimize manual fraud reviews by automatically weighing the risk of fraud against the customer’s value and approving or declining the order.

Kount scores transactions on a grade scale, so the higher the score, the safer the transaction. Knowing this, for example, Kount users can automatically decline orders with Omniscores lower than 80 to reduce manual reviews significantly.

“The machine learning that powers Kount’s Omniscore seeks to replicate or improve on the fraud expertise of human agents,” Harrison concluded. “You can use that expertise to ensure the right transactions are marked for review, avoid reactive fraud rules, and substantially reduce your manual reviews during peak seasons.”