It is the operational equivalent of walking into a restaurant and telling the waiter, "I would like the food."
The waiter, being polite, will bring you something. It may even be good. But you have not given an instruction. You have given a mood.
This is what happens when a buyer of traffic tells a vendor to "block fraud."
It sounds like a clear directive. It is not.
Fraud is not a single, self-evident category. It is a spectrum that runs from a script hammering a landing page to a real human on a shared VPN who clicks faster than your model thinks is polite. Somewhere in that spectrum is a line. The buyer has not said where the line goes. They have simply said: draw it.
So the vendor draws it. And the vendor draws it where the vendor's incentives point.
That is not an accusation. It is arithmetic.
A vendor that blocks more traffic reports lower fraud rates. The fraud that gets through is visible to the buyer in the form of wasted spend. The legitimate users who get blocked are invisible. They do not complain. They just do not convert. The publisher gets less money, but the publisher is not the one who signed the contract.
So the default setting, in the absence of instruction, is to block more. Not because anyone is dishonest. Because the buyer asked for food and the vendor brought the dish that photographs best.
"Block fraud" is not an instruction. It is a request to be surprised later.
What would a real instruction look like?
It would start with the question the buyer almost never answers:
How many real customers am I willing to lose to catch one bot?
That is not a rhetorical question. It is a number. It has to be a number, because every block decision is a trade. You cannot eliminate fraud without eliminating some legitimate traffic. The only way to guarantee zero fraud is to block everything, which is the one fraud solution that also solves revenue.
Then you need to specify the evidence standard. What counts as proof? A browser that contradicts itself? A session that behaves like ten thousand others? A probability score? If you cannot take the evidence to the partner who sold you the traffic and have an argument, the block is just a fee.
Then you need to say what happens to the ambiguous cases. The clicks that are probably fraud but not certainly. The users who look strange but might be real. Do you block them and lose the revenue? Do you allow them and accept the risk? Or do you hold them for review, and who does the review?
These are not technical questions. They are business questions. And they belong to the buyer.
The moment you say "block fraud," you have outsourced the only decision that was yours to make.
We have been asked for a simple switch. A dial. A button that says "block fraud."
We think the button is the problem.
Because the button does not ask what you are willing to lose. It just blocks. And when the bill arrives — in the form of a publisher who stops sending traffic, or a campaign that quietly underperforms — the buyer will say the vendor blocked too much, or too little, and the vendor will say the buyer set the threshold.
Both will be right.
The next time you sit across from a vendor, do not say "block fraud."
Say: "Here is what I am willing to lose."
The silence that follows will tell you everything.