You hire a security guard for your shop. He reports he stopped 99.9% of shoplifters. He does not mention the customers he turned away because they looked nervous.
That is the number nobody publishes.
When you buy traffic, you are not buying clicks. You are buying the people attached to them. A click is just a proxy for a customer. So when an anti-fraud vendor tells you they blocked 99.9% of invalid traffic, you are missing the other number.
How many legitimate customers did they block?
How many people tried to reach you and were told, by a machine, that they were not real?
How many sales did you lose to a false positive?
The vendor does not publish that number. It is not in the marketing materials. It is not in the quarterly report. It is not even in the contract.
Why would it be? The fraud rate is a story about how much money they saved you. The false positive rate is a story about how much money they cost you. One of those stories sells. The other one gets you fired.
Here is the uncomfortable part. A false positive is invisible to you. The blocked customer does not complain. They do not send an email. They just go somewhere else. You never see the click, so you never see the loss.
The publisher sees it. The publisher watches the revenue disappear. But the publisher is not the one paying the vendor. You are.
A blocked click is not a refund. It is a customer who walked away.
The vendor will tell you the percentage of fraud they blocked. They will not tell you the percentage of customers they blocked. One of those numbers is flattering. The other is a conversation you would want to have before you sign.
My email spam filter once ate an invitation to a wedding. I found it three years later. That is the entire false-positive problem in miniature: the system was certain, and it was wrong, and I never knew.
The vendor defines what counts as invalid. So the false positive rate is a number about how often the vendor's own definition was wrong. But the vendor never has to admit it was wrong, because the customer never comes back to argue. The customer is gone.
Measuring false positives is hard. You would have to follow up with every blocked user and ask, "Were you actually a human trying to buy something?" Nobody does that. It is expensive, and the answer would be uncomfortable.
So the number stays unpublished. And because it stays unpublished, the buyer assumes the system is working. The buyer sees a low fraud rate and thinks the vendor is doing a good job. The buyer never asks the obvious question: How many good customers did you throw away to get that number?
The fraud rate tells you what you did not pay for. The false positive rate tells you what you did not earn.
That is the number that matters to you. Because you are not in the business of avoiding fraud. You are in the business of acquiring customers. A vendor that blocks 99.9% of fraud but also blocks a small fraction of your customers is not saving you money. It is costing you revenue. And you will never know, because the customers do not come back to complain.
You might think you are saving money by not paying for those clicks. But you are not. You are losing the revenue those clicks would have generated. A click that never happens is not a saving; it is a missed sale.
Nobody gets a bonus for discovering that the traffic was fine. The vendor's sales team does not get a commission for telling you how many good customers their system threw away. The publisher does not get a refund for the revenue that disappeared. And you do not get a report that says, "Here are the customers you lost." So the number stays in the dark.
The next time a vendor tells you they blocked 99.9% of invalid traffic, ask them one question: What is your false positive rate?
Then watch how long the silence lasts.
Somewhere, a customer tried to buy from you, and a machine said no. You will never meet them.
That is the cost of the number nobody publishes.