Cutoff Rules by Metric: How Not to Kill Creatives Too Early
Learn to build cutoff rules by play, checkout, and sale with a sampling buffer so you don't kill winning creatives or overspend on testing.
Why killing a creative at the exact metric cost is a mistake
A metric-based cutoff rule works when you give it a sampling buffer before you kill the creative. If your ideal cost per play is $8 and you set the rule to kill at $8, you'll kill good creatives every single day. The number swings throughout the day, spikes, drops back, and only lands at the real value by the end. Cutting at the metric's floor is like rolling the dice once and deciding the result is final.
If you run volume, you know the scene: cost per play spikes at 2pm, you panic, you kill 30 creatives. By 10pm that same play would be at $8. You just killed a good offer over noise.
The metrics that matter, in order
The analysis flow follows a ladder. Each rung costs more than the one below and tells a different story about the creative.
The order that makes sense for anyone running Direct Response:
- Play: did the video hook them? Cheap metric, you can get volume fast.
- Bit (early engagement on the creative): did they get past the hook?
- Checkout: did they reach the payment page? Now the cost starts to weigh.
- Sale: they closed. The only number that pays the bills.
You look at them in this sequence because each one filters the one below. A creative that doesn't generate plays won't generate checkouts. But the cutoff rule for each has to respect how much it costs to gather that sample.
Minimum spend floor before any cut
Here's the part most people get wrong. An ideal cost per play of $8 does not mean a cutoff rule at $8.
Think about the math. Your CPA is $500. If you let the creative run until it spends the full $500 with no sale before deciding to kill it, you burn budget for nothing. Cost per play gives you a cheaper, earlier signal. But $8 spent with zero plays isn't a signal, it's noise.
The fix is a minimum spend floor. In practice it works like this: you set a minimum amount the creative has to spend before any cutoff rule fires. For play, a floor around $150 usually does the job. Spent $150 and the play is still expensive? Kill it. Before that, let it run.
That floor protects the creative from dying over a momentary spike. It also protects your budget from running a bad creative all the way to full CPA.
Sampling buffer: the cost multiplier
Every operator lands on their own number, but there's a multiplier logic that works well. You let the creative run until it spends 5 to 10 times the metric value before deciding.
Cost per play of $8? Let it run to around $60, $80. If you work with a 10 to 1 buffer, a $10 cost becomes a $100 floor. The idea is simple: you need enough of a sample to trust the number.
Rolling the dice once and landing on 1 proves nothing. Rolling it ten times and landing on 1 every time is statistically rare. Same thing with play. A play at $8 doesn't get killed. Wait for 10, 20 plays to stack up so you're sure the cost is real and not just an unlucky start.
The minimum spend calculation is what gives you that sample volume without having to count plays by hand. Spend $100, $150, and now the rule can act.
Cheap metrics and expensive metrics need different treatment
The 10x multiplier works beautifully for a cheap metric. For checkout, the math falls apart.
If your ideal cost per checkout is $200 and you applied 10x, you'd have to let it spend $2,000 before cutting. Nobody runs a test like that. Expensive metrics don't allow the same sampling luxury as cheap ones.
That's why the smart play is to use the cheap metrics as your main filter. Play and bit give you an early signal, with low spend and a generous sample. You kill the bad creative up there, before it reaches checkout and costs a fortune to prove the obvious.
Checkout comes in as confirmation, not as your first line of cutting. You reach it with a creative that already passed the cheap filters.
Offer-level study defines the numbers
There's no universal number. The $150 floor for play and the 10x multiplier are starting points, not law.
The right move is to study each offer. Every offer has its own cost per play, checkout, and CPA. You pull those numbers for that specific offer and calibrate the floor and buffer from there. An offer with a $500 CPA takes a different buffer than an offer with an $80 CPA.
This work of setting a rule per metric, per offer, repeats with every new structure you launch. When the test involves dozens of creatives running in parallel across several accounts, keeping naming and configuration consistent becomes a critical part of the control: with standardized naming across accounts (a stack like DirectAds solves that friction), you make sure each creative launches the right way and your cutoff rules hit the right creative, with no mix-ups in targeting or a swapped audience.
Watch out for cost swings during the day
The point that ties it all together: cost per metric is not a straight line. The play that lands at $8 passes through $15, $20, $6, $11 over the course of the day.
If your cutoff rule looks at the instant number, it'll fire at the peak and kill a creative that was about to win. The spend floor and the sampling buffer exist precisely to ride out that swing. You wait for the creative to accumulate enough spend and sample for the instant number to close in on the real one.
It's the difference between watching the sea on a single wave and watching the average tide level. The rule has to decide by the tide, not the wave.
Takeaways
- Never set a cut at the exact metric cost. Add a spend floor and a sampling buffer of 5 to 10 times the value.
- Study each offer to define floor and multiplier. One offer's numbers don't carry over to another.
- Use cheap metrics (play, bit) as your main filter and checkout as confirmation, never the other way around.
- Wait for the creative to ride out the day's swing before killing it. Judge by the tide, not the wave.
Frequently asked questions
What's the minimum spend floor to cut by play?
It depends on the offer, but a floor in the $100 to $150 range usually gives enough sample for cheap metrics. The ideal is to calculate it as a multiple of your cost per play: 5 to 10 times the metric value.
Why not use the 10x multiplier on checkout?
Because checkout is an expensive metric. A $200 cost per checkout times 10 would mean $2,000 in spend before cutting, which makes the test unworkable. Expensive metrics call for confirmation with a smaller sample, and the heavy filtering stays on the cheap metrics.
How do I know if a high cost is noise or the real number?
Sample. Instant cost swings during the day. Only after enough spend accumulates (the minimum floor) does the number close in on the real one. Before that, you're looking at a wave, not the tide.
In what order should I analyze a creative's metrics?
Cheapest to most expensive: play, bit, checkout, sale. Each one filters the next. A creative that doesn't generate plays never reaches checkout, so you cut early and cheap.
