Who kills your losing ads, you or the algorithm?

How do I tell which setup I am running right now?

Open your ads manager and find the budget. Not the amount, the location. That single fact decides who is in charge of your losing ads.

On Meta, the budget sits in one of two places. At campaign level it is called Advantage campaign budget, formerly campaign budget optimisation: one number on the campaign row, with every ad set underneath reading "Using campaign budget". That is one pot. At ad set level, each ad set row carries its own number, and each is a pot that defends itself.

On Google, the budget always sits at campaign level, so the question is not where it lives but how many campaigns you run. One campaign covering everything is one pot. A separate testing campaign alongside it is a second pot, and it exists precisely so that money set aside for learning cannot be quietly pulled into the thing that already works.

Then count the pots. That number is the whole answer. One pot means the algorithm is killing your losing ads. More than one means you are. And if it is more than one and nobody is reviewing them on a schedule, the honest answer is that nobody is.

Do this before you touch a single piece of creative. Most of the accounts we inherit have a creative problem sitting on top of a structure problem, and the structure problem is the one nobody looked at.

What does one pot actually cost you?

It costs you the answer.

I run an account with eight ads spread across four ad sets, all drawing on a single campaign budget. The platform put roughly five sixths of the money into one of those four ad sets and left the other three on scraps. Inside one of the starved sets, a single ad received nine cents of spend across the entire run. Nine cents.

That account was efficient and it asked nothing of me. I killed nothing, because there was nothing to kill. The platform had already made every decision I would have made, faster, and for free.

What actually happened to that nine-cent ad was not one decision. It was two. The campaign budget decided that its ad set was not the interesting one, so the whole set got a twelfth of the money. Then, inside that starved set, delivery decided the other ad was the better bet and took almost all of what little was left. Two different mechanisms, stacked. Only the first one has anything to do with where your budget sits.

I still have no idea what those six ads could have done. They finished on samples somewhere between no conversions and a handful, which is not enough to conclude anything about any of them. They did not lose. They were never tested. The algorithm reached a verdict before the evidence existed, and because it starved them rather than switching them off, the account never showed me a failure. It showed me a tidy result and a blank space where six answers should have been.

That is the real bill for one pot, and it does not appear on any invoice. You saved money and you learned nothing about what you had.

What does many pots actually cost you?

It costs you money and it costs you attention, and the second one is the one people underestimate.

On another account I run, every ad set holds its own budget across a main campaign and a testing campaign. No ad set there dies on its own. Its pot has money, its only job is to spend that money today, and it will find something to spend it on. Last month I opened it, went through it, and switched off nine ads I was no longer willing to fund. No drama, no disaster. Routine maintenance, and it was maintenance only because I was there to do it. Nobody looks, nobody kills, and the account bleeds on schedule.

Now, if the platform is always picking a favourite ad inside each set, why was there anything left for me to kill? This is the part that took me longest to see, and it is the real answer to who kills what.

The algorithm picks the best ad in the ad set. It never asks whether the best ad in the ad set is any good. It sorts inside whatever constraint you hand it, and it does not question the constraint. When each ad set holds its own budget, the constraint you have handed it is "this set spends its money today, no matter what". So it does exactly that, cheerfully, on the least-bad thing in the box.

That is what those nine ads were. Not ads the algorithm failed to sort. Ads it sorted perfectly, inside boxes that should not have had money in them. The sorting was flawless and the question was wrong, and only a person can notice that the question is wrong.

The same account taught me the mirror version of the problem, which is the one nobody warns you about. I had an ad that was genuinely good and it was being starved inside a shared pot by ads that were not better, just earlier. So I built it an isolated lane and forced the money into it by hand.

Read those two together and you can see what many pots really buys. In one direction I was doing the algorithm's job by hand, killing what it would have killed. In the other I was overriding it, funding what it had written off. Many pots hands you the steering wheel, and a steering wheel is only worth having if somebody is holding it.

Is it really as simple as counting pots?

No, and I have been making it sound tidier than it is.

The proper names are worth knowing, because they are what everyone else will say to you. ABO, ad set budget optimisation, is where each ad set holds its own budget. CBO, campaign budget optimisation, now called Advantage campaign budget on Meta, is where the campaign holds the budget and the platform spreads it across the sets. That is the lever this whole article is about. It is the biggest one. It is not the only one, and four other things decide whether any of it means anything.

Ads inside an ad set always compete, in both setups. This is the one that catches people. ABO does not mean every ad gets a fair go. It means every ad set gets a fair go, and inside each set the platform still crowns a favourite and starves the rest. So counting pots tells you who allocates between your ad sets. Inside them, the algorithm is holding the knife no matter what you chose. If you want two concepts genuinely tested against each other, they need two ad sets, not two ads in one.

Volume decides whether any verdict is real. An ad set that is not getting somewhere in the region of fifty conversions a week is still guessing, and the platform will tell you it has a winner anyway. A winner at low volume is frequently noise wearing a hat. Most of the accounts we see are not choosing badly between concepts, they are choosing confidently on data too thin to hold a conclusion. That is a budget problem dressed as a creative problem.

Your own ad sets bid against each other. If two sets target overlapping audiences, you are in your own auction, pushing your own costs up. A second pot is not automatically a second opinion. Sometimes it is just you, twice.

Time changes the answer. Allocation shifts as the platform learns, and an ad set that looked dead in week one can be carrying the account by week four. A verdict reached on day two is not a verdict, it is a weather report.

Put those together and the honest version is this. The structure decides who decides. Everything else decides whether the decision is worth anything. Counting your pots is the first question, and it is the one nobody asks. It is not the last one.

Why is killing a lot of ads not a failure?

Because killing ads is what testing looks like from the outside.

Nine dead ads sounds like nine mistakes. It is nine answers. We were testing concepts on that account: different angles, different promises, different ways in. Nine did not land, so now I know nine things that do not work on that audience, and I know it for certain, because I paid to find out. That is the price of the answer, and the answer stays bought.

The platforms are built to sort. Give the algorithm one ad and you have given it nothing to do, because there is no choice to make. Give it eight and it will go and find your two. Variety is the raw material you feed it, and an account that produces no dead ads is usually not one that is winning. It is one that never tested anything.

So the pile of dead ads is not the mess. It is the receipt. The winners came out of that pile, and there was no other door they could have come through.

When would you deliberately choose the expensive one?

Many pots is the costlier setup and there are four situations where we choose it on purpose.

When you do not yet know what works. A new offer, a new audience, a new market: the platform has no history to sort on and neither do you. Handing it one pot means paying for a verdict reached on almost no evidence. Buy the learning first, then buy the efficiency.

When a concept deserves a fair hearing. If you have an ad you believe in strategically, an isolated lane is the only way to guarantee it gets tested rather than starved by whatever happened to convert first.

When the learning is worth more than the media. Knowing which angle your market responds to outlives the campaign. It goes into your landing pages, your email, your sales calls, your next offer. An account structured only for efficiency will never tell you that, however well it performs.

When somebody is actually watching. This condition governs the other three. Many pots without a weekly review is not a testing strategy, it is just an expensive setting. If nobody is going to look, one pot is the more honest choice and you should take it deliberately.

The reverse holds too. Once you know what works and the job is efficiency rather than discovery, one pot is usually the better tool. It is faster than you, cheaper than you, and it never forgets to check.

Is the platform taking this decision away from me anyway?

Increasingly, yes, and that is exactly why the choice needs making consciously now rather than by default.

Two changes this year point the same way. Google is changing how Target CPA and Target ROAS behave in budget-limited campaigns: from 17 August 2026, a campaign that has quietly been beating the target you set will be pulled back towards that target instead of being left to run, with a Bid Target Adjustment Tool appearing in accounts from 6 July 2026 (Search Engine Land, 22 June 2026). And Google has moved Demand Gen campaigns on Discover with view-through conversion optimisation from cost-per-click to cost-per-thousand-impressions billing, effective 15 July 2026 (Search Engine Land, 16 June 2026). Same campaign, different meter.

Neither is a scandal. Both are the platform taking more authority over where your money goes and how it is counted. That direction of travel will not reverse, which means the part still under your control, where the budget sits and how many pots you run, is a larger share of the decision than it used to be, not a smaller one.

If you are not sure which setup your account is running, or you have counted your pots and the number surprised you, tell us about your business. We will tell you what we would do about it, and whether it is worth doing at all.