When to Kill a Meta Ad: The 72-Hour Decision Rule

An agency owner sent me a screenshot at 11pm last month asking whether he should shut an ad off. Eighteen hours old, $180 spent, zero bookings. He had already killed it before he hit send.

That ad was the one we would have scaled.

I watch this happen every week, and it always shows up as the same two habits inside the same account. Winning ads get killed on day two. Losing ads get left alone for three weeks because nobody wants to admit the creative they loved is not working. Both habits come from the same root cause, which is reading the wrong number at the wrong time and calling it a decision.

My brother Offek and I have built this decision into a fixed rule for every account our team touches, because gut feel is the single most expensive input in an ad account. The rule has two gates: 72 hours of time, and a spend threshold tied to your target cost per booked call. Nothing gets killed, held, or scaled until both gates are cleared.

The number you stare at on day one is noise

When a new ad set goes live, Meta does not know who your buyer is yet. It runs an exploratory delivery pattern, pushing your ad into different pockets of the audience to gather signal. That period is the learning phase, and while it runs, your cost per result swings wildly by design.

Meta stabilizes an ad set at roughly 50 optimization events per week. That is the actual mechanic. Below that volume, delivery stays unstable and your reported CPA is closer to a rounding error than a result.

Here is where B2B agencies get stuck. If your target is a $400 qualified booked call, 50 of those events per week means $20,000 a week in spend. Almost nobody in our client base is starting there. So the answer is not to pretend the learning phase does not apply to you. The answer is to optimize the campaign on an event that actually happens 50 times a week, such as a lead or a qualified form completion, and then judge the ad on the deeper event that pays you.

That distinction matters more than any creative decision you will make this month. You optimize on the event Meta can learn from. You judge on the event your bank account cares about.

Why 72 hours is the honest read window

B2B buying behavior is delayed and messy. An agency owner sees your ad on a Tuesday during a break between calls, does nothing, thinks about it, comes back Thursday through a branded search or a direct visit, and books. Your day-one dashboard shows that as a failure. Your week-one dashboard shows it as a booking.

Attribution lag alone means anything under 48 hours is fiction. We use 72 because that is where the curve flattens enough to trust across the accounts we run. Cost per result on day one is routinely double what it settles at by hour 72.

There is also a human reason for the 72-hour rule, and it is not a small one. A hard time gate removes the 11pm decision. Nobody makes a good call about a creative asset they wrote themselves while looking at a red number on a phone screen. The rule protects the test from the person running it.

Time is one gate. Spend is the other.

Seventy-two hours at $8 a day tells you nothing at all. You need both gates, and the spend gate is anchored to your target CPA rather than to a flat dollar amount, because a $400 target and a $2,000 target are not the same test.

Use these thresholds:

Under 1x target CPA in spend, you have no information. Do not touch the ad. If your target is a $400 booked call, that means an ad with $260 behind it is not a data point, it is an opinion.

At 1x to 3x target CPA in spend, plus 72 hours elapsed, you have your first honest read. At a $400 target, that is the $400 to $1,200 window.

Past 3x target CPA with nothing to show, the test is over. At a $400 target, an ad that has spent $1,200 with zero qualified bookings has answered the question.

That is the whole framework, and it works because it scales with your economics instead of against them.

The kill, hold, and scale tree

Once both gates are cleared, every ad in the account falls into exactly one of three outcomes. No fourth option, no “let it ride another week and see.”

KILL it when the ad has spent 3x target CPA with zero qualified bookings. Also kill it when it has bookings but the cost per booking is sitting above 2x your target and the trend across the last 24 hours is flat or climbing. A CPA that is bad and still rising is not a creative that needs patience.

HOLD it when spend sits between 1x and 3x target CPA with at least one qualified booking, or when the ad has no bookings yet but the leading indicators are strong. Holding is an active decision with an end date, not a shrug. Set the review for the next 24 to 48 hours and put it back through the same tree.

SCALE it when the ad has three or more qualified bookings at or under target CPA after clearing 72 hours. Three is the floor because one booking is luck and two is a coincidence. When you scale, raise budget by roughly 20 percent every three to four days, or duplicate the winner into a fresh ad set at a higher budget and leave the original running untouched. Tripling a budget overnight knocks a stable ad straight back into learning, which is how people convince themselves a good ad “died.”

What to read when there are no conversions yet

Most ads inside the 1x to 3x window have not produced a booking yet, and you still need to make a call. Leading indicators tell you whether the ad is failing at the top or at the bottom, which are completely different problems.

Start with hook rate, meaning three-second video views divided by impressions. Under about 20 to 25 percent, the first frame is the problem and nothing downstream matters. That ad is not being rejected, it is not being seen. We break down what actually earns that stop in ten ads a week instead of one perfect ad.

Then look at outbound click-through rate. Under 1 percent on cold B2B traffic means the hook stopped the scroll but the body never earned the click. That is a middle-of-the-ad problem, not a targeting problem.

Then compare cost per landing page view against your account average. If clicks are cheap but landing page views are expensive, people are clicking and bouncing before the page loads or before it makes sense to them.

Finally, look at landing page view to lead rate. If the ad is delivering cheap, engaged traffic and the page converts none of it, your ad is not the thing that is broken. The leak moved downstream and you would be killing a healthy ad to fix a page.

The edits that quietly reset your test

Half the failed tests I audit were never really tests. They were ads that got edited mid-flight, which reset the learning and threw away the spend that came before.

These edits restart learning: changing budget by more than about 20 percent at once, changing the audience, swapping the creative inside a live ad set, and changing the optimization event. Pausing an ad set overnight to “save money” and switching it back on in the morning does the same thing.

If you want to test a new creative, it goes into a new ad set. If you want a bigger budget, it moves in steps. If you want a different optimization event, that is a new campaign and a new read window. Discipline here is worth more than any hook you will write this quarter.

Judge the ad on the event that pays you

This is where most agencies get the whole thing backwards, and it is worth being blunt about. Ranking creative on clicks or cost per lead produces ads that generate cheap leads and empty calendars. The ads that win on click metrics are frequently the ads that attract people who were never buying.

Rank creative on qualified bookings, on show-up rate, and on revenue closed. Those three numbers sort your creative in a completely different order than CTR does, and that reordering is the entire point. Our funnel does this ranking automatically instead of leaving it to whoever opens Ads Manager on a Monday, which we walk through in how our self-optimizing funnel picks winners.

It also matters more now than it did two years ago. Since the Andromeda update, your creative is doing the targeting work that audience settings used to do, so the creative you choose to keep is effectively the audience you choose to buy. Killing the wrong ad no longer just costs you an asset, it points the algorithm at the wrong person. We covered that shift in what Andromeda did to targeting.

The rule is the product

When we took a client agency from a $215 CPA to $30 in 14 days, roughly $50,000 booked inside the first 10 days, the mechanism was not one brilliant creative. It was a fixed rule applied without emotion to every ad in the account, week after week, so that losers died at 3x target CPA and winners got fed instead of strangled.

Write your target CPA at the top of your testing doc. Multiply it by one and by three. Those two numbers are your gates. Add 72 hours and stop making the decision any other way.

The agencies that win on Meta are not the ones with better instincts. They are the ones who took the instinct out of the decision entirely.