Nobody on our team decided to cut the creative that had been driving our pipeline...

The algorithm did.

And based on the number it was looking at?

It made the right call.

In early July, we launched a new creative concept.

Almost immediately, it started producing leads cheaper than anything else we had running.

So Meta did what Meta is built to do...

It fed the winner.

Within about ten business days, that one concept had grown to roughly 28% of our total ad spend.

Sounds great.

Except that money had to come from somewhere.

And a big chunk of it came out of another creative concept that had been steadily driving real opportunities.

Spend on that original concept fell by roughly half.

A few business days later?

Our opportunities fell too.

That got our attention.

Because the new creative wasn't bad.

Quite the opposite.

It was doing exactly what the dashboard said it was doing.

It was generating leads cheaply.

The problem was what happened after the lead.

When we dug into it, we saw something we hadn't been measuring closely enough.

The new concept was attracting people who were curious about the problem.

The old concept was attracting people who were living inside it.

Those sound almost identical when they become a form fill.

They are very different when they become pipeline.

Interest is not urgency.

And cost per lead can't tell you the difference.

That's the part that should make any growth leader a little uncomfortable.

Because CPL wasn't just sitting in a report.

CPL was quietly deciding where our money went.

The platform saw one concept producing cheaper leads and kept moving spend toward it.

Again, perfectly rational.

It had no idea that another concept with a higher CPL was producing more of the business outcome we actually cared about.

It only knew what we had taught it to value.

So it optimized harder.

And harder.

And harder.

That is where a useful proxy can become dangerous.

Not because the number is wrong.

Because the number can be right about the wrong thing.

And when you're spending millions across dozens of campaigns, creatives, platforms, and ad sets, finding that distinction isn't always simple.

In our case, the answer was scattered across creative variations running under different names, campaigns, and structures.

To see it manually, we'd have been bouncing between Meta and HubSpot, downloading spreadsheets, matching creative back to pipeline, and trying to reconstruct what changed.

Instead, we could group the same creative concept together across where it was running and follow the trail from spend to leads to actual opportunities.

That's when the pattern became obvious.

One concept gained budget.

The other lost it.

Then the opportunity mix changed.

We didn't suddenly discover that the new creative was a failure.

We discovered something much more useful.

You cannot defund a proven pipeline engine just because a new idea produces a cheaper proxy.

So we changed the way we test.

The proven pipeline concept gets its own lane.

New ideas get incremental budget on top of it until they prove they can drive the outcome that actually matters.

Simple change.

But we only knew to make it because we could finally see what the algorithm couldn't.

And that's the uncomfortable question.

Your platforms are optimizing something right now...

Maybe CPL...

Maybe CPA...

Maybe ROAS...

Maybe another perfectly reasonable number.

But what if the creative getting starved is actually the one creating the customers, pipeline, or revenue you care about most?

The platform won't stop because it's wrong.

It will optimize harder.

The only question is how much budget moves before somebody realizes what the winning number has been hiding.

‍