You made four new ads. Good ones. You wrote them, filmed them, edited them, and dropped them into the campaign that has been carrying your business. Then you watched Meta spend almost nothing on any of them.
A week later, one of them has forty cents of spend against it. Another has none at all. The ad that was already winning has taken the lot, and you are staring at a report that tells you nothing about whether the work you just paid for was any good.
We wrote about why that happens in You Put Five Ads in the Ad Set. Meta Only Really Runs One. Meta crowns a favorite fast, on thin evidence, and pours the budget into it. That part is settled. What almost nobody explains is the follow-on question, which is the one that actually costs you money: knowing all that, how do you run a test that means something anyway?
There are three levers. Two of them are settings sitting in your account right now that most owners have never opened. But before any of them, there is a question worth asking, because the most expensive testing mistake is testing at all.
Before You Test Anything, Ask Whether You Should
Here is a question that will save some of you a month of work: can you spend more money tomorrow than you spent today, and still make it back?
If the answer is yes, stop reading about testing and go increase your budget. You have already solved the hard problem. Launching a batch of new ads into a campaign that is working is close to the worst thing you can do to it, and the reason is mechanical rather than a matter of taste.
Every new ad you upload starts from nothing. Meta has no idea how people will respond to it, so it has no track record to price it with, and it treats it accordingly. To find out, it does the sensible thing and shows it first to the people most likely to react, which is your warmest audience. That is the same small pool of people your existing winner is already converting.
So your new ads do not politely wait their turn. They compete with your best ad for the same buyers. Your winner's numbers get worse because a slice of its warm audience is being handed to an unproven ad, the new ad's numbers look mediocre because it is unproven, and your cost per result drifts up across the board. You respond by launching more ads to fix it, which is the thing that caused it. That is the loop, and once you are inside it you are not running an ad account anymore. You are firefighting one.
Testing earns its place when the answer is no. When you have pushed the budget and the numbers break, when your cost per lead has been climbing for three weeks, when the winner is fatiguing and nothing you have is ready to replace it. That is the moment to open the account and start experimenting. Not on a Tuesday because it feels productive.
Meta Built a Test Button. Most Owners Have Never Seen It.
If you do need to test, the first lever is a tool Meta built specifically for this problem, and it is called creative testing.
It does not live where you would expect. Go into an existing campaign, drop down to the ad level, and scroll until you find a section called Creative test. Click set up test. Four settings follow, and each one matters more than it looks.
How many copies of the ad to make. Between two and five. You update the creative on each copy after you have confirmed the setup, so these become your test versions.
How much of your existing budget to spend across the test ads. Meta's own guidance here is worth quoting: use no more than twenty percent, to reduce the chance of damaging your campaign's performance while the test runs. It aims to spend that amount per day on the test ads. There is a right and a wrong way to find that twenty percent, which is the next section.
How long the test runs. Worth knowing that this setting only controls when results are measured. Your test ads start running on the ad set's own schedule the moment you publish them, whatever you put in the test duration.
Which metric decides the winner. Set this to your cost per result. Leads, purchases, whichever your campaign is optimizing for. Do not judge a creative test on engagement or video views. Those numbers feel like progress and say almost nothing about whether the ad produced a customer.
What this actually changes is delivery. Drop new creative into a live ad set the normal way and it competes for budget against an entrenched winner, which keeps eating. In a creative test, delivery is provided to the new test ads, so they reach enough people to give you a real read. You stop asking Meta to be fair and give it a structure where fairness is the default.
There is a second, quieter benefit that answers the loudest criticism of Meta's older A/B testing tool. Experienced buyers have avoided that tool for years on the grounds that it runs in its own environment, and when you move the winner into a real campaign afterwards, all the learning resets and the delivery system starts from scratch. Creative testing is set up inside your existing campaign precisely so that does not happen. High performers carry on running with the delivery system's learnings intact, and there is nothing to merge anywhere.
Two things to know before you rely on it. It does not work on campaigns using a bid cap strategy at all. And when the test ends, Meta emails you the results and flags top performers, but it makes no automatic changes. Your test ads simply keep running in the campaign until you turn them off. Winning is not a promotion and losing is not a removal, so the tidying up is your job.
One more piece of honesty from Meta's own documentation, and it matters for a piece about knowing when a result is real: the results identify your best-performing test ads against your chosen metric, but no confidence level is included. Meta is telling you which ad did better over the window. It is not telling you the difference was statistically meaningful. On a small budget with a handful of conversions, treat a narrow win as a hint rather than a verdict, and be willing to run it again.
Raise Your Budget Before You Test, Not After
Now back to that budget field, because this is where most tests are quietly sabotaged before they start.
Read Meta's instruction closely. The money for the test comes out of your existing campaign or ad set budget. It is not additional. So if you are spending thirty dollars a day, stable, bringing leads in at a price you are happy with, and you hand twenty percent to a test, your proven ad is now running on twenty-four dollars instead of thirty.
That is manageable at twenty percent, which is exactly why Meta recommends that ceiling. It stops being manageable the moment you get ambitious. Push a third or a half of your budget into a test and you have starved the ad paying your bills in order to interrogate ads that have not earned anything yet. Your results dip, and the natural conclusion is that the new creative is bad, when the real cause is that you defunded your winner.
The fix takes two days and no cleverness. Before you set up the test, raise your daily budget by roughly what you intend to spend on the test. Move from thirty to thirty-six over a couple of days, gently, so you do not knock the winner out of its stride, and let it settle. Now when you assign twenty percent to the test, that money is genuinely new. Your winner keeps the thirty that was working and the experiment runs beside it rather than out of its pocket.
The Setting That Stops One Ad Eating Everything
The second lever is older, less known, and blunter. It lives at the ad set level, under budget and schedule, and it is called ad set spending limits. You will only see it if your budget is set at the campaign level rather than per ad set.
It caps how much of the campaign's budget a single ad set can absorb. Put a maximum of eighty percent on the ad set that keeps eating everything, and the remaining twenty percent has to go somewhere. Where it goes is the newer ad sets you are trying to give a chance.
Three details decide whether this helps you or hurts you.
It is a weekly cap, not a daily one. This trips people up constantly. Meta manages the limit as an average across a Sunday to Saturday week, and will not exceed seven times your figure across that week. Which means an individual day can and will run over your number, as long as the week comes out right. If you set a maximum expecting a hard daily ceiling, you will look at Tuesday, see it blown through, and think the setting is broken. It is working exactly as designed.
Set it as a percentage, not a cash amount. Meta lets you choose either. Choose the percentage, because there is a trap in the cash version: Meta cannot spend an increased campaign budget if the ad set is still pinned to its old maximum. Raise your campaign budget from fifty to sixty dollars while a fifty-dollar cap sits on the ad set, and that extra ten simply never gets spent. A percentage moves with every budget change you ever make. A fixed number waits quietly for you to forget about it.
Set the maximum on every new ad set too. The cap is per ad set. Cap the incumbent, leave the next batch uncapped, and in a few weeks the new batch is the one hogging everything.
One more piece of housekeeping worth knowing: pausing a capped ad set reduces what the campaign can spend in total. Five ad sets capped at ten dollars each on a fifty-dollar campaign adds up neatly, but pause one and your campaign can now only spend forty, even though the budget still says fifty. And spend limit edits take up to fifteen minutes to take effect, so do not panic in the first ten.
When You Genuinely Want to Force It
There is a matching control on the other side, a minimum, and the honest advice is to leave it alone most of the time.
A minimum tells Meta to push money into an ad set whether it wants to or not. Meta is not being arbitrary when it declines. It is weighing more than conversions, including how people respond to the ad, and it is usually right. Overriding that means paying to show people an ad the system has already predicted they will not like. Meta's own position, stated plainly in its documentation, is to use as few spend limits as possible, because they stop a campaign budget chasing the best opportunities available to it.
Four things to respect if you do it anyway.
A minimum is a target, not a promise. Meta describes it as the lowest range it should aim for, and says outright that no refund is issued for failing to hit one. It is a nudge with no guarantee attached.
Do not use a minimum and a maximum on the same ad set. Meta explicitly does not recommend it, and says the combination can produce a higher cost per action, worse return on ad spend, or underperforming delivery. If you insist, it wants at least a one percent gap and at least a dollar between the two numbers.
Minimums fight with cost controls. If you are using cost per result goal bidding or a bid cap, Meta warns that delivery can be constrained to the point where the minimum becomes unachievable. You end up issuing two contradictory instructions and getting performance nobody can explain.
It will probably cost you in the short term. You are betting against a system with far more data than you have. Sometimes the bet is right, and there is a real pattern behind it: an ad with a higher cost per thousand impressions can still be the better ad if it converts well enough to cover the difference, and a system leaning toward cheap delivery can leave that ad unfunded. Just place the bet knowingly.
If you do want to force a test, set the minimum high enough that the ad set gets meaningful daily spend rather than a trickle, watch it properly, and turn it off after a week if nothing is happening. And change one thing at a time. Do not introduce spending limits, a new bid strategy and a fresh batch of creative in the same week, because you will have no idea which one moved the numbers.
One Campaign, or Two?
The third lever is not a setting at all. It is how your account is laid out, and there are two defensible answers.
The traditional structure is two campaigns. One for testing, where you launch a fresh batch of creative every week or two, and one for scaling, which holds only the ads that have already proved themselves. Winners graduate from the first into the second. The appeal is control: you decide what gets tested and roughly what it costs you, and a test can never destabilise the campaign carrying your revenue.
The cost is real though. You will spend money on ads that were never going to work, your learning is spread across two campaigns rather than pooled, and if both are chasing similar people you can end up bidding against yourself and paying more for the same impressions.
The other answer, which has become much more common, is one campaign that tests and scales at the same time. New creative goes in beside proven creative, all the data pools in one place, and Meta sorts it out. It is less work to manage, the learning compounds in a single ad set rather than being split, and for a business spending in the hundreds rather than thousands per day it usually gets more out of every dollar.
The question that decides it is uncomfortably simple: are you willing to spend money to learn something Meta would have told you for free?
If you have budget to burn and you want certainty about a specific angle, run the separate testing campaign and accept the waste as tuition. If every dollar has to work, consolidate, and let Meta's forecasting do the first round of filtering for you.
One practical note whichever you choose. When you move a winning ad into another campaign, use the existing post rather than uploading the creative again. The ad keeps the likes, comments and shares it has already earned, and that accumulated social proof is doing quiet work for you. Upload it fresh and you throw all of it away and start from zero.
The Rules That Make a Test Mean Anything
None of the levers above rescue a badly designed test. Five rules do most of the work.
One variable at a time. Change the creative or change the copy, never both. If you change two things and the numbers move, you have learned that something worked, which is not knowledge you can use again.
Judge on cost per result. Not engagement, not click-through rate, not hook rate. Those are diagnostics you reach for after the money number looks wrong.
Give it a week. Buyers who run these tests all day tend to allow three or four days for an ad to stop behaving like a new ad, then a few days of clean data on top. Reading a creative test on day two is reading noise.
Remember nobody is telling you the result is significant. Meta reports which test ad performed best on your chosen metric and stops there, with no confidence level attached. A clear win on decent volume is a result. A two percent difference across nine conversions is a coin toss wearing a chart.
Make the ads genuinely different. This is the one people get wrong most often. Five versions of one image with a different headline font, a different background color and the price moved to the other corner are five ads to you and roughly one ad to Meta. It will pick one, run that, and leave the rest unfunded. If you want a real test, change the idea and not the decoration.
Do not stack changes. New creative, a new bid strategy and a budget increase in the same week is not an experiment. It is a mess with a report attached.
A Refusal to Spend Is a Result
Last thing, and it is the one that reframes the whole exercise.
Meta assesses your ad before it spends anything meaningful on it. It looks at the creative, at everything it knows about your account and your buyers, and it forms a view about how the ad is likely to do. When it declines to fund something, that view is usually correct. Most of the time, an ad that cannot get spend is an ad that was not going to work.
So the first response to a starved ad should not be to force money into it. It should be to consider that you have been given an answer, for free, without paying for the impressions to learn it the hard way.
It is not infallible, and it is worth knowing where it slips. The same set of ads relaunched in a fresh campaign will sometimes get funded completely differently, which tells you the first verdict was contingent rather than absolute. And Meta leans toward ads that produce a good experience, measured largely by how people engage, which is not always the same as the ad that sells the most. There is a column for this that almost nobody switches on. Add engagement rate ranking to your Ads Manager view and you can see how Meta rates your creative against comparable advertisers. Average or above and the system will tend to favour your ad and your costs will benefit. Below average and you have found a real reason your ad is not getting distribution.
What to Do on Monday
Open the account and ask whether you can spend more than you are spending now and still make it back. If yes, do that and leave the creative alone this week.
If no, pick one thing to test. One. Raise your daily budget by roughly what the test will consume, give it two days to settle, then go to the ad level of that campaign, find the Creative test section and set up a test. Two versions, no more than twenty percent of your budget, judged on cost per result. Not on a bid cap campaign, because the tool will not run there.
While that runs, open the ad set that has been eating everything, find ad set spending limits, and set a maximum as a percentage rather than a cash figure. Then do the same on every ad set you launch from here, and remember the cap is averaged across the week, so a heavy Tuesday is not a fault.
Leave the minimum empty unless you have a specific ad you believe in and are willing to pay to prove.
When the test ends, go and turn things off yourself. Meta will email you a result and change nothing, so the losing versions keep spending until you intervene. And when you move a winner into another campaign, use the existing post so it keeps its comments and shares.
Testing on Meta is not really about finding the perfect ad. It is about keeping enough genuinely different ideas in front of the system that it always has something good to reach for, and making sure the ideas you paid to produce actually get their turn. Most small businesses are not losing because their ads are bad. They are losing because their best ad never got enough money to show them what it could do.
If you would rather someone else watched the spend split and ran the tests properly while you get on with the business, that is what our Meta ads management does all day. And if you would rather run it yourself, you now know the three levers most people advertising on Meta have never opened. Either way is a good outcome from where we are sitting.
