The Simplest Campaign Structure for Testing Creatives
How to build a basic campaign with bid cap or CBO, let the machine optimize, and why Facebook needs less tweaking today than it used to.
The simplest structure that works for testing creatives
One campaign. Bid cap to control the test, six ads per ad set, and you touch it once a week. That's it. People think they need some massive engineering feat of ad sets, budget sliced down to the penny, daily adjustments. You don't. To validate creatives on Meta Ads today, the cleaner your structure, the faster you read what works and what doesn't.
The most common mistake isn't building the wrong campaign. It's building too much campaign.
Single campaign: bid cap or CBO, and nothing else
The base rule: one campaign only. Launch it, create the ad sets, drop in the creatives, and let it run. If your pixel already has history, go in with a bid cap just to hold the cost during the test window. That gives you predictability while you read the creative, without letting Meta spend wide before you understand what's happening.
Now, if the pixel is new, forget the bid cap. A new pixel with a bid cap stalls, can't find an audience, goes nowhere. In that case you go with CBO, one campaign, budget at the campaign level, and let the algorithm distribute. It's what Facebook itself recommends for an account that's still being built, and here the machine's recommendation is a good one.
The logic is simple: a mature pixel can handle bid control, a new pixel needs freedom to learn.
Why six ads per ad set?
Six is the number that gives you enough test volume without spreading the budget too thin. Each ad set supports up to 50 ads under Meta's limit, but dumping 50 at once to test creatives dilutes the spend: every ad gets scraps of delivery and none of them gather enough data to make a call.
With six per ad set, the algorithm has room to pick the ones that respond and concentrate delivery on them. You read the hook, you read CTR, you read what holds attention. And you can rotate in new creatives without rebuilding the whole structure.
When the test runs across several accounts at once, building six ads in each ad set by hand, account by account, becomes the bottleneck. That's exactly where a test structure replicated in bulk across BMs kills the friction: you set the six creatives, the config comes out identical across every account the first time, no naming errors and no ad sets missing an ad.
Should you let the machine drive? Yes or no?
Today, yes. Fully.
The question that trips a lot of people up: "how am I going to outsmart all of Meta's engineering?" You're not. There's an absurd amount of data and optimization behind that recommendation. Direct response has its quirks, it's different from a traditional branding campaign, but overall what the platform suggests today is built to work.
Touching the campaign once a week is enough. People can't pull this off, not because it's hard, but because they can't stop tinkering. They keep adjusting ad sets every day, resetting the learning phase, killing delivery too early. And then it goes nowhere.
The real test of discipline is opening the ads manager and not touching a thing.
Why today is different from a year ago
If you'd asked me this six months or a year ago, the answer was the opposite: do the opposite of what Facebook tells you. Back then that's how it actually worked. You had to hack the platform, force delivery, trick the system into giving you what you wanted.
But follow the timeline to understand what changed.
First we had a pixel that worked beautifully, tracked everything. Then governments started tightening up on privacy, they broke the pixel, and Meta had to swap direct tracking for AI modeling. It spent a good while with a busted pixel, bad data, unpredictable delivery. In that phase, hacking was the only way out.
Today the machine is much smarter. It was a move toward the Google-ification of delivery.
How the algorithm works now
YouTube always worked this way: the campaign starts at a slight loss while the system learns the audience. Once it figures out who converts, it takes off. Facebook now runs on the same pattern.
In practice, here's what happens:
- First week: spends little, barely finds an audience, looks like it's stuck. Normal.
- It starts finding: one sale comes in, then two, delivery calibrates.
- Hits ~50 conversions: exits the learning phase and the rocket takes off.
That number of 50 conversions per week to exit learning isn't a guess. It's Meta's own criterion for stabilizing delivery. You used to day trade inside the ads manager, in and out all day long. Now it's the opposite: whoever keeps their hands off and lets the learning phase close wins.
Takeaways
- Launch one campaign only. Bid cap if the pixel has history, CBO if the pixel is new.
- Put six ads per ad set, enough to read creatives without diluting the budget.
- Touch the campaign once a week and let the learning phase close.
- Wait for the 50 conversions before judging performance. Before that the algorithm is still calibrating.
Frequently asked questions
Do I use bid cap or CBO to test creatives?
Depends on the pixel. A pixel with history can handle a bid cap, which holds the cost during the test. A new pixel stalls with a bid cap, so go with CBO and let the algorithm find an audience with no restrictions.
How many ads do I put per ad set?
Six is the number that gives test volume without spreading the budget too thin. Meta's limit is 50 per ad set, but dumping them all at once dilutes delivery and no single creative gathers enough data to decide.
How often should I touch the campaign?
Once a week. Daily adjustments reset the learning phase and kill delivery before the algorithm calibrates. The less you touch it early on, the faster it stabilizes.
Why does the campaign barely spend in the first week?
It's the learning phase. The algorithm is still mapping who converts. It spends little, finds little audience, looks stuck. Once it gets close to 50 conversions, delivery takes off.
Is it still worth hacking Facebook like before?
Not like a year ago. Back then the pixel was broken by privacy changes and forcing delivery worked. Today the delivery AI is much smarter and following the platform's recommendation usually gets you more results.
