When to Kill a Facebook Ad: The Decision Tree We Use Inside Client Accounts

Nine times out of ten, when a founder tells me an ad "didn't work", what actually happened is they killed it on day one. No purchase by bedtime, so it got switched off the next morning. That's not a dead ad. That's an ad you never gave a chance to optimise.

Knowing when to turn off a Facebook ad is half the job of running an account, and almost nobody has a written rule for it. They go on feel, they get twitchy, and they kill good ads in a panic while leaving genuinely dead ones running for weeks.

So here's the actual decision tree we run inside client accounts, top to bottom. It's deliberately built so you can paste it straight into your SOPs and stop relying on mood. Three levels: the campaign, the ad set or test, and the individual ad. The rules are different at each level, and that's the whole point.

Campaign level: you almost never kill a campaign

Start at the top, because this is where people overthink it.

I basically never turn a campaign off. When we take over an account, we'll usually consolidate down to one campaign per business objective, and from there the campaign just stays on. You're not killing campaigns to optimise. You're adjusting budget inside them.

There's really only one reason to kill a campaign: the business objective behind it stopped being valid. You were spending into a particular country and you've decided to pull out of that market. You were running a seasonal collection and the season's over. The objective died, so the campaign dies with it.

That's it. If the campaign still maps to something you want to sell, to someone you want to sell it to, it stays on. Killing a whole campaign because last week was soft just throws away learning the account paid real money to build. So at the campaign level, the answer to "should I kill this" is almost always no.

The interesting decisions all happen one level down.

Ad set and test level: settle the 3-day versus 7-day fight

This is where the real question lives. You launch a test, and you have to decide how long to let it run before you call it.

The argument is always 3 days versus 7 days. Here's how I actually settle it, and the deciding factor is your account's daily spend.

If you're spending under ~A$1,000 a day, you need the full 7 days. At that spend level, three days simply doesn't accumulate enough data to trust, and you'll get fooled by noise. There's a second reason too: Facebook traffic runs in cycles. The people who buy on a weekend often behave nothing like the people who buy on a Wednesday, and which days are good or bad swings hard by brand and by who your customer is. I've seen brands quietly crush it on weekends and die midweek, and others do the exact opposite. Seven days is the minimum window that captures a full cycle instead of half of one. So on smaller accounts, give every test a clean week.

If you're spending much more, say ~A$10,000 a day, three days can be enough, because that volume of spend reaches statistical honesty far quicker. You can read a test in 72 hours that a smaller account would need a week to see clearly. The window isn't a fixed rule. It scales with how fast your spend accumulates real data.

Either way, here's how I read the first three days. If a test is steadily building, a bit of spend on day one, more on day two, more again on day three, that's healthy and I'll let it run the full week. But if three days pass and it has barely spent a cent, the odds it suddenly becomes a winner are low. Every great ad I've seen takes off within about 48 hours. A test still flatlining on spend after three days is usually just a weak ad telling you the truth early.

The rule worth defending: no real spend in 7 days means it's dead

This is the one people argue with, so let me defend it properly.

If you launch a batch of tests, give them a full seven days, and they never take meaningful spend, those are dead ads. Turn them off. Not "needs more time", not "the algorithm didn't get it". Dead.

Here's the logic, and it's worth internalising. When you hand Facebook a set of creatives, you're giving the algorithm the chance to spend on whichever ones it thinks will perform. If it had a full week and chose not to put budget behind your ad, that is the system telling you the creative is weak. Facebook is, in effect, refusing to bet on it. I don't fight that signal. I take it as data.

Think of it like a YouTube video nobody finishes watching. The platform sees the weak engagement, decides it isn't worth showing to more people, and quietly buries it. Same mechanism on Facebook: a creative that can't earn its own distribution is answering the question for you. No spend across a fair window equals a dead ad. Switch it off and free the budget for something with a pulse.

The one early kill: when an ad grabs spend and tanks the account

Now the exception, because there's exactly one scenario where I'll break the "give it time" rule and kill inside 24 hours.

If an ad suddenly takes the majority of the account's spend and your overall cost per acquisition falls off a cliff, that's a same-day kill. The test for an early kill isn't "no sales yet". It's the opposite: it's spending aggressively and actively dragging down account-wide performance.

The classic version is the negative-comment spiral. You launch a creative for a product people have a strong reaction to, and the comments fill up with people piling on. Facebook can't tell good engagement from bad, so it reads all that activity as signal and prioritises the ad. It scales the thing fast, eats your budget, and torches your CPA, all off the back of people who hated it. The moment that spend is significant and performance is clearly suffering, kill it. You don't owe a week to an ad that's bleeding the account.

One guardrail even here: I still let it run at least 24 hours before pulling it, because an ad needs a little time to optimise, and I don't want to cut it during a quiet morning and call it dead by lunch. A creative typically takes around three days to fully settle. So even the early kill has a floor of a day, unless the damage is so bad that waiting would be negligent.

The outlier signal: frequency, and why it isn't a kill on its own

There's one metric founders love to kill on, and I want to put it in its proper place: frequency.

Frequency is how many times the same person has seen your ad. An ideal sits around 1.0. If a test is running at, say, 1.68, that means roughly 68% of the people reached have seen it twice already. Compare that to a winner sitting at 1.23, where only about 23% have seen it twice, and you learn something useful. A high-frequency ad with a high CPA is usually showing to a small audience over and over because the creative can't break into a bigger one. That ad won't scale, and reallocating its budget to a lower-frequency performer will genuinely improve the account.

But, and this matters, I do not kill on frequency alone. If an ad is crushing it and happens to run a higher frequency, I leave it be. Frequency is context, not a verdict. It tells me what kind of creative to make next, whether I need fresh angles that open up a larger audience, far more than it tells me to pull the trigger today.

And one more guardrail before you kill anything on cost: don't judge an ad purely on a hard threshold like "spent 1x my AOV with no sale" or "ran to 3x my target CPA" without looking at the time of day. On a big account you might burn a third of the day's budget before 8am, and if that lands in a naturally weak selling window, killing the ad then is wildly premature. We've turned off great ads too early on exactly this mistake, relaunched them later, and watched them go on to perform. The clock is part of the read.

Ad level: leave the winners alone

Bottom of the tree, the individual ads inside your main set. The instinct here is to tinker. Resist it.

I rarely turn off winning ads. What I see people do instead is turn off a high-spending ad that's slightly off target and force budget onto a low-spend ad with a gorgeous return. It feels productive. It usually makes things worse, because those low-spend darlings tend not to hold up once they're carrying real budget.

The trap is reading a tiny sample as a winner. An ad that got A$5 of spend and a 20x return didn't crack the code. It got one lucky sale. The ad that spent A$1,000 at a 2x return is the one telling you the truth at scale. Always read performance at the spend level that actually matters, not at the level that flatters.

So when we're a bit below target, I'd far rather sit at a 5 to 10% inefficiency for a while and hunt for a new winner than start flicking ads off to feel like I'm doing something. Forcing optimisation usually buys you a worse number and a noisier account. The job at the ad level isn't to kill the underperformer. It's to build a new test good enough to beat the existing best ad, and let the winner keep winning until it does.

So the real skill in all this isn't knowing what to turn off. It's having the discipline to do nothing on the days your gut is screaming to act. Most accounts aren't bleeding money because a few weak ads stayed on a week too long. They're bleeding it to a founder killing good ads in a panic, over and over, and calling it optimisation. So before you reach for that toggle tomorrow, it's worth asking: are the rules you actually run written down anywhere, or are you just trusting that today's mood happens to be right?