Segmentation feels like control. More campaigns, more ad sets, more lines in the reporting. It looks like rigour. In most accounts it is the single biggest reason the system never learns anything worth acting on.
Here is the uncomfortable version. Every campaign and every ad set is its own machine learning model. Conversion data does not pool across them. Split a budget into twelve ad sets and you have not built twelve experiments. You have built twelve under-powered ones, each starved of the signal it needs to make a confident decision, all paying the same platform fees to learn more slowly.
Concentration is the lever that fixes this, and almost nobody pulls it because pulling it feels like doing less. It is not. It is doing fewer things with enough budget behind each one that the answer is actually reliable.
What over-segmentation actually costs you.
The cost is not wasted spend in the obvious sense. The cost is signal. A campaign needs a meaningful number of conversions before the algorithm settles on a stable way to spend the budget. Fragment the budget and most of your campaigns never get there. They sit in a permanent half-learned state, re-optimising every few days, never converging.
You can see it in the account. Spend that lurches around on a flat daily budget. Top-performing ad that changes every week. Half the campaigns flagged limited by budget, which is the platform politely telling you it does not have enough signal to spend what you have given it. None of that is a creative problem or a targeting problem. It is a structure problem.
The point of the account is to teach the system, and you, what works. Excess segmentation buys a precision the data cannot support.
This is the part operators miss. Splitting a budget does not give you more insight. It gives you less reliable insight, more slowly, while the fees stay the same. You are paying full price for a worse decision.
Signal per decision.
The frame that fixes this is signal per decision. Before you create another campaign or another ad set, ask what decision it is meant to inform and whether it will ever accumulate enough conversions to inform it confidently. If the honest answer is no, you have not added a test. You have added noise and a line item.
Consolidate until every meaningful decision has enough signal behind it to be trusted. That is the rule. It is unglamorous and it is almost always the highest-leverage change available in an over-built account.
If a campaign will never gather enough conversions to settle, it is not a test. It is overhead.
Before adding structure, ask: what decision does this inform, and will it ever have the signal to inform it confidently? If not, the budget belongs somewhere it can actually learn.
What to consolidate into.
Concentration is not one giant campaign. It is a deliberately small structure, organised around what you actually need to know, with budget heavy enough that each part can learn fast. In practice that is three jobs, not twelve:
- Testing. One campaign whose job is to find the next concept. Segment it at the ad set level by concept, not by audience or product. Force enough budget through each ad set that a winner is legible inside a sensible window, then graduate it.
- Scaling. Where proven concepts go to absorb budget. Concentrated, broad, and fed by the testing campaign rather than by guesswork.
- Retargeting. Kept separate and kept honest, so warm audiences are never quietly inflating the read on cold acquisition.
The structure should also move with the spend. What works at $30k a month is not what works at $300k. As spend climbs you add concentration, not fragmentation, because the larger the budget the more it matters that each decision is built on real signal rather than a slice of one.
When segmentation is actually justified.
This is not a rule against ever splitting anything. There are real reasons to separate budget, and they share one feature: a genuine commercial difference on the other side of the split.
- Geographies with materially different economics or delivery.
- Languages that need different creative.
- Product lines whose margin or AOV is different enough to deserve a different CAC ceiling.
- Audiences with a genuinely different commercial value to the business.
Each of those is a decision. Most of the segmentation we inherit is not a decision. It is a default that was never questioned, copied forward through account rebuilds, quietly draining the signal out of everything it touches.
So if you take one thing away from this
Concentration is not the cautious option. It is the aggressive one. It says you would rather make fewer decisions and trust them than make many and trust none. Done properly it speeds up learning, cleans up the read, and lifts efficiency on the same spend, because the algorithm is finally working from a signal dense enough to act on.
Count the campaigns and ad sets in your account. Then ask how many of them will ever gather enough conversions to tell you something you can act on. Consolidate everything that fails that test. That is usually the fastest structural win available, and it costs nothing but the discipline to build less.