Primary vs Secondary: The Conversion You Chase and the Ones You Only Watch
Your account probably has a dozen things marked as conversions. The algorithm can only chase one of them well. Every conversion you let it optimize toward is a vote for the kind of customer you get more of, and most accounts vote for everything.
Your Google Ads account probably has a dozen things marked as conversions. Form fills, calls, page views, button clicks, a newsletter signup somebody added two years ago and forgot.
Here is the problem. The algorithm can only chase one thing well. Every conversion you let it optimize toward is a vote for the kind of customer you get more of. If you vote for everything, you have not voted at all.
That sounds like a small technicality buried in a settings panel. It is actually the most consequential decision in the account, because it decides what the machine spends your money trying to find.
The algorithm optimizes toward what you mark Primary
It is worth being plain about what Google is doing under the hood, because the whole argument rests on it.
When you mark a conversion as the one to optimize toward, the model goes and gets you more of that exact thing. That is its entire job. It studies the signals it has about millions of people, finds the pattern that predicts your chosen action, and pours budget toward the humans who match. It is very good at this, and it is getting better every year.
Notice what that means. The model does not know what a good customer is. It knows what you told it to count. It is not chasing revenue or closings or happy clients. It is chasing the specific event you labeled Primary, and it will chase that event straight off a cliff if that is where the event happens to live.
A dozen goals is the same as no goal
So picture the common setup. An account with ten or twelve things all marked as conversions. Form fills, phone calls, a newsletter signup, a PDF download, a video view, three different button clicks, a visit to the pricing page.
To a human that looks thorough. To the model it is a blur. You have not handed it ten goals. You have handed it a pile and told it to go make the pile bigger, and it will do that in the cheapest way it can find. Newsletter signups and page views are easy and cheap. Actual sales-ready phone calls are hard and expensive. Guess which end of the pile the model drifts toward.
It optimizes toward whichever of the dozen is cheapest to produce, and cheapest-to-produce is almost never the same as most-valuable. You asked for volume across everything, so it gave you volume from the easiest things, and the report looks busy while the pipeline stays empty.
Targeting versus Observation
The fix is a distinction I hold firmly. Every conversion is either Targeting or Observation.
Targeting is the one outcome you actually want more of. The thing worth spending the next dollar to get. That is what the model chases, and there should be almost none of them. Observation is everything else worth knowing about. The softer signals, the middle steps, the events that tell you what is happening without earning a vote on where the budget goes.
Both live in the account. Only one is allowed to steer. The whole practice is deciding, on purpose and against your instincts, which is which.
What Observation conversions are actually for
Demoting a conversion to Observation does not mean deleting it. This trips people up, so it is worth its own section. The Observation events keep doing useful work. They just stop driving.
They are your diagnostics. A page-view conversion tells you whether people are getting deep enough to care. A newsletter signup tells you something about intent, even if it is weak intent. The middle-of-funnel steps show you where the journey stalls. All of that feeds your reading of the account and your attribution, and none of it should touch the bidding.
Think of it like the gauges on a dashboard versus the steering wheel. You want a lot of gauges. You want exactly one steering wheel. The mistake is not watching too many things. The mistake is letting all the things you watch grab the wheel at once.
The discipline is ruthless demotion
Getting this right is not clever. It is mostly the courage to demote things that feel like wins.
Most of the dozen events in your account are Observation. One, maybe two, are Targeting. The hard part is never identifying the valuable outcome. Everyone can point to it. The hard part is being willing to tell the model that almost nothing else counts, and to leave the satisfying, easy-to-move numbers sitting in the Observation column where they inform you but do not drive you.
It feels like giving something up, because it is. You are giving up the comfort of a big conversion count in exchange for the model actually working on the thing that pays. I have never met an operator who regretted the trade once the closings caught up.
”But I want to see everything”
The usual objection is that demoting conversions means flying blind. It does not, and this is the part worth getting right.
More data is good. More steering is not. You can watch every event in the account, report on all of it, and study the whole journey, while still telling Google to optimize toward one thing. Observation gives you the full picture. Targeting gives the model a single clean instruction. Those are different jobs, and conflating them is exactly the error. You do not have to choose between seeing everything and steering by one thing. You get to do both, as long as you keep them separate.
Why this bites harder in real estate and lending
In a fast, cheap purchase, a sloppy conversion setup is forgivable. The cheap conversions and the real ones are not that far apart, and the sales cycle is short enough that mistakes surface fast.
Real estate and lending punish it. The sale is large, the cycle is long, and the gap between an easy form fill and a real closing is enormous. So when the model drifts toward the cheap events, it is drifting toward people who will never close a six-figure transaction, and it can do that for months before anyone connects the healthy dashboard to the empty pipeline. The higher the stakes and the longer the cycle, the more a muddled Primary quietly costs, because the attribution stays green the whole time it is steering you wrong.
What you get back
When only the outcome you actually want is marked Primary, the budget starts flowing toward it. The model has one clean signal to learn from instead of a committee of them. The traffic slowly shifts toward people who resemble your real buyers, because those are the only ones that now count.
And the reporting gets honest. There is a clear line between the thing you are steering by and the things you are only keeping an eye on, so a good month in the dashboard finally has a chance of being a good month for the business. The attribution stops arguing with the pipeline.
The platform will optimize toward whatever you call Primary. That is the whole game. The whole skill is having the discipline to call almost nothing Primary, and the patience to watch everything else without letting it grab the wheel.