Kill criteria: when should you turn off an ad (and when should you not)?

Without fixed kill criteria, your gut decides when an ad dies. Here is how to define upfront at what spend and what signal an ad stops, and how to escape the sunk cost trap.

Turn off an ad based on kill criteria you defined upfront, not based on feeling. The simplest version: decide per test how much spend an ad gets to prove itself, which signals you expect at which checkpoint, and what happens when those signals fail to show. If the ad misses the checkpoint, it goes off, no debate. That way every stopped ad becomes a learning instead of a lingering doubt.

Why does your gut lose to fixed rules?

Every media buyer knows the moment. An ad performs poorly, but yesterday brought one good sale, and the concept feels strong. So you give it one more day. Then another. Before you know it, a mediocre ad has eaten three times its testing budget and you still have not made a decision. That is not a character flaw, it is how humans work: we weigh what we already invested more heavily than what the next euro will return.

Kill criteria remove that negotiation. You make the decision once, upfront, with a cool head, and afterwards you only execute it. That is exactly why they work: the moment the rule applies, your opinion no longer matters. In accounts growing from 15K to 150K per month we see the same pattern every time: the tighter the kill criteria, the faster the testing rhythm, and the sooner the winners surface.

How do you set kill criteria that actually hold up?

The biggest mistake is anchoring criteria to time instead of spend. Letting an ad run for three days means nothing if twenty euros flowed through it in those three days. Spend is the currency an ad uses to buy information, so spend should be the unit of your criteria. A logical anchor is your target CPA or the value of one conversion: give an ad, for example, one to two times your target CPA in spend to reach the first checkpoint.

Work in stages as well. Not every signal becomes reliable at the same pace, so at each checkpoint only judge what can actually be judged at that point.

• Checkpoint 1, early spend: look at hook rate and CTR. If nobody watches past three seconds and nobody clicks, you do not need to wait for the rest.
• Checkpoint 2, around one time your target CPA in spend: look at cost per click and the quality of the traffic, such as add to carts or leads completing the form.
• Checkpoint 3, around two to three times your target CPA: only now does cost per result get to deliver the verdict. Zero conversions at this point is a clear answer in itself.

The staged model prevents the two classic mistakes: judging too early on a metric that has no data yet, and judging too late because you kept waiting for a miracle. Mind the exception too: an ad that scores exceptionally well at checkpoint 1 but falls just short at checkpoint 3 sometimes deserves a second life with a different landing page or a different offer. The creative was not the problem.

What is the sunk cost trap in ad accounts?

The sunk cost trap is the urge to keep an ad running because a lot of money already went into it. You hear it in sentences like: we already spent 800 euros on this one, let us give it a bit longer. But that 800 euros is gone, whether you keep the ad on or not. The only question that counts is: what do I expect from the next 100 euros in this ad, compared to the same 100 euros in a fresh test? Once you frame it that way, the answer usually becomes uncomfortably obvious.

The budget you already spent is not an argument. The only question is what the next euro returns.

When should you not turn an ad off?

Kill criteria exist to stop weak ads fast, not to nervously choke your winners. A proven winner having two soft days does not fall under your testing criteria. Scaling ads need wider margins: daily results fluctuate, attribution lags, and an ad hitting its target on a weekly level has nothing to prove on a daily level. Judge testing ads on their checkpoints and scaling ads on weekly averages, and keep those two regimes strictly separate.

Be careful around external factors too. A CPM spike in a crowded auction week, a sold out product or a broken checkout are not creative problems. When the context is off, pause the verdict, not necessarily the ad.

Conclusion

Kill criteria sound like a small operational detail, but they set the pace of your entire testing machine. Predefined spend checkpoints, staged signals and a hard line between testing and scaling: that is the difference between an account that learns and an account that drags. At AdSplicit we build these decision rules into every account we manage, so budget consistently flows toward the strongest ads. Curious how tight your testing process really is? Book a call and we will gladly take a look with you.