A common underlying assumption when merchants review orders in a queue is that, if the agent reviewing the order is still unsure whether to release the order after looking it over, the agent will reject it by default.
This course of action is regularly pursued when there is limited information provided to an agent at the time of review, and oftentimes will happen with newer customers or guest checkouts where you don’t have historical information to work from. This might make sense, based on the situation and the desire to eliminate chargeback risk, but it’s important to weigh the pros and cons.
Let’s do a little math to illustrate the point. Say I’m a manual reviewer, considering what to do about a $200 order with a 20% profit margin. I’m completely on the fence—in my judgment, there’s a 50/50 chance it’s fraud. If I reject the order by default once I’ve finished my investigation (and I’m still stumped), I have a 50% chance of gaining $200 (the avoided fraud) or losing $40 (the profit on the lost sale). Over time, rejecting these orders should net me $80 in gains on average (50% of $200, minus 50% of $40).
Each order is $200 with a 20% profit margin.
Reviews have determined that the orders have a 50% chance of being fraudulent.
Your net gain will be $80 on average, if each order has a 50% chance of being fraudulent.
Now, suppose instead that I’m leaning toward it being a legitimate order—I'm not certain, but I estimate there’s a 90% chance that it is. If I reject the order, I have a 10% chance of gaining $200, and a 90% chance of losing $40. Over time in this case, rejecting these orders should net me $16 in losses on average (10% of $200, minus 90% of $40).
Reviews have determined that the orders have a 10% chance of being fraudulent.
Your net loss will be $16 on average, if each order has a 90% chance of being fraudulent.
A good practice is to give anyone doing manual review on held orders a variety of options to indicate why they did what they did. In other words, your case management tool should have the ability to create customized labels per action. When there’s no doubt whatsoever about whether the order was legitimate or fraudulent, something like “accepted with certainty” and “rejected with certainty” could be helpful. If there is uncertainty, “accepted with reservations” and “rejected with reservations” can add context to a decision.
Review your team’s decisions regularly to get a sense of how accurate their gut feelings are, and give guidance appropriately. This will provide much needed context for the review, and allow management teams to query and assess labels/label groups later on—a critically important step in understanding your true false-positive and false-negative rates.
Prevent all types of online fraud and abuse with a single fraud detection solution for end-to-end protection.