Somewhere in your fraud vendor's dashboard there is a number.
It might be called a threshold, a confidence level, a sensitivity setting, or a dial that goes from one to ten.
You are told it is a technical parameter.
It is not.
A threshold is a budget.
When you set it, you are deciding how many real customers you are willing to throw away in order to avoid paying for one more bot. You may not know you are deciding that. The dashboard does not phrase it that way. But the arithmetic does not care what the label says.
Suppose you sell refurbished espresso machines. A real customer arrives after searching "dual-boiler espresso machine review," reading three comparison posts, leaving the tab open overnight, and returning through a retargeting ad. A bot can do every one of those things. The difference is that the customer will eventually buy, and the bot will not. The threshold is the line where you stop paying to find out which one you are looking at.
A threshold is a budget, and it is the only budget in your media plan you have never seen in dollars.
That is the part nobody says out loud. A fraud vendor will happily tell you that 2.3% of your traffic was invalid, or that 0.8% was suspicious, or that a particular click was blocked at 73% confidence. But nobody hands you the invoice for the customers who were blocked at 74%.
Actually, they do hand you the invoice. It is just buried in your revenue.
The vendor's model may be perfectly good. The signals may be real. The classification may be defensible. But the moment the output is a probability and the threshold is a slider, the vendor has stopped doing the job and handed you a business decision without the information a business decision requires.
You are not being asked to tune a model. You are being asked to price a customer.
Most advertisers cannot answer that question. They know what a click costs. They know what a conversion is worth. They do not know what a false positive costs, because nobody reports it. They do not know what a false negative costs, because nobody reports that either. They are setting a budget in the dark.
A false positive on a $1,200 espresso machine is not a two-dollar click. It is a customer who might have bought the machine, the grinder, the descaling kit, and then come back for beans every six weeks. A false negative is not a single fraudulent click. It is a bot that learns your threshold and comes back tomorrow dressed as a different customer. Neither number appears in the dashboard, but both appear in your P&L.
Once the threshold is your setting, every argument about a wrong call becomes an argument about your configuration.
That is a structural consequence, not a plot. The vendor is not hiding anything. They may genuinely believe the dial is empowering. It feels like control. You can turn it up. You can turn it down. You can watch the blocked percentage move. It is the most satisfying knob in the dashboard.
But a dial that changes an outcome without showing you the trade is not control. It is a way of making you responsible for a decision you did not have the evidence to make.
What would actually help is less exciting. Not a probability. A reason.
"Blocked: this browser reported itself as Safari on iOS, then answered a question only Chromium can answer."
That is something you can take to the partner who sold you the traffic. It is something you can argue about. It is something you can price.
We are building toward that. Not a dial that asks you to guess, but a system where every rejected click carries the specific contradiction that killed it. Then the threshold stops being a mystery setting and becomes what it actually is: a queue of claims you can review, dispute, and pay for or not pay for with your eyes open.
You already make business decisions about every other line in your media plan. The fraud threshold should not be the one place you are asked to make one without the numbers.