Most accounts are training the algorithm to find the wrong person

Ad platforms have spent the last five years making it easier to launch a campaign and harder to run a good one.

It used to be about creative, audience, and angle. Who you were talking to, and what you said to them. Now the pitch from every platform is the same: give us everything and we will handle the rest.

That is genuinely true. It is also conditional on something almost nobody checks.

The conditional part

The machine optimizes toward whatever event you tell it is valuable. It is very good at this. Point it at a signal and it will go find more of whatever produced that signal, at a scale and speed no human media buyer can match.

So the question stops being “who is my audience” and becomes “what am I telling the machine to go find.”

Most accounts are answering that question by accident.

I have opened accounts where the optimization event was a page view. Not a form, not a call, not a booking. A page view. That account was spending real money every day teaching a very capable system to go find people who look at things. It found them. It got very good at finding them. The reports looked fine, because the number it was reporting went up.

Why it happens

It is almost never stupidity. It is usually one of three things.

The good event does not have enough volume. Platforms need somewhere in the range of thirty to fifty conversions in a week to learn anything. If a business books four jobs a week, optimizing toward booked jobs means the campaign never exits the learning phase. So somebody slides the target down the funnel to get volume, and then keeps sliding, and eventually lands on something that fires often enough to work and means nothing.

The event was set up once and never audited. Someone configured it during a website build two years ago. Nobody has opened it since. I have found a conversion named after a newsletter signup that was actually configured to match any page view on the site. Real signups were around 114. The platform was reporting roughly 710. Nobody was lying. Nobody had checked.

The event fires before the thing actually happens. A form listener attached to the submit action rather than the confirmation counts failed validations, spam, and people who tried three times. Those all look like conversions.

What this costs

Here is the part that makes it worse than a reporting problem.

A broken conversion is not neutral. If you had no data at all, you would be cautious. You would ask questions, spend slower, look for other evidence. Wrong data does the opposite. It makes you confident, and then it tells you to spend more, because the thing you are measuring is going up.

Bad tracking does not hide the problem. It creates confidence in a lie.

And the damage compounds, because the machine is learning the whole time. Every day the wrong event runs, the audience model gets more precisely wrong. Fixing the event later does not undo the training. You start over.

What to actually check

You do not need a measurement overhaul. You need to answer four questions honestly.

What single event is each campaign optimizing toward? Not what conversions are being tracked. What is the campaign actually bidding against. These are frequently different.
Does that event represent a person who could become a customer? Say it out loud. “We are paying to find people who submit this form.” If that sentence sounds fine, good. If it sounds like nothing, you found it.
When does it fire? Before or after the thing succeeds. Go submit the form wrong on purpose and see if it counts.
What is the raw number? Pull the count from the source, not from the ad platform’s reporting. If those two disagree, believe the source.

That last one is the one people skip, and it is the one that finds the problem.

The part nobody wants to hear

If your optimization event is wrong, your campaign performance data is not a report on your advertising. It is a report on a broken instrument.

Every conclusion built on top of it is unsafe. The creative test you ran, the audience you cut, the budget you shifted. All of it was decided using a number that was not measuring what you thought.

That is not a reason to distrust the platforms. They do exactly what you configure them to do. It is a reason to go look at what you configured.

The platforms will find whoever you tell them to find. That is the whole job now.