How long should I let a Facebook ad run before killing it?
Judge on spend, not on time. A workable default is to let a creative spend roughly 2x your target CPA before killing it for zero conversions, and to wait for a meaningful event count before judging its CPA at all. Days elapsed is the wrong axis, because an ad that spent five dollars in three days has told you nothing, while an ad that spent three times your CPA in one day with nothing to show has told you plenty.
Last updated 2026-08-11
Why time-based rules fail
The same 48 hours means completely different things at different budgets. On a small ad set, two days may not buy enough impressions to fill a single conversion's worth of signal; on a large one, two days is an exhaustive verdict. Delivery pacing also varies day to day with auction conditions, so a calendar rule fires on quiet Tuesdays and misses expensive weekends. Spend is the common currency: it measures how much question-asking the ad has actually done. Convert every kill rule you currently phrase in days into a multiple of target CPA in spend, and the rule starts behaving consistently across budgets, seasons and ad sets.
A default rule set
Kill a creative that reaches roughly 2x target CPA in spend with zero conversions, because the arithmetic of what it would need to do from here to hit target is already improbable. For creatives with some conversions, wait until the event count is large enough that one order either way would not flip the verdict, then compare CPA to your break-even with room to spare. Scale decisions want more evidence than kill decisions, since scaling multiplies the consequence of a wrong read. Whatever thresholds you choose, write them down before the test launches; rules decided while staring at results bend toward whatever the results already suggest.
The learning phase distorts early reads
A new ad set's delivery is deliberately exploratory while Meta works out who converts, so early performance is both unstable and typically worse than the ad's steady state. Killing during that window systematically executes ads that were about to improve, and the ad set as a whole never stabilises because every kill and replacement is itself an edit that re-opens learning. The practical accommodation is patience with structure: fund the ad set so learning can complete at all, avoid mid-test roster changes, and treat the first stretch of delivery as provisional. The 2x-CPA zero-conversion rule survives this caveat because an ad that spent that much with nothing has failed under any delivery regime.
When fast kills are correct
Some failures are visible before any conversion data exists. A creative with a hook rate far below the account's norm is failing at the first frame, and impressions will not fix it. A broken destination URL, a wrong product, a policy rejection on relaunch, or comments filling with confusion about the offer are structural faults, not statistical ones. Kill those immediately regardless of spend. The distinction to hold is between evidence problems, which need more spend to resolve, and defect problems, which need none. Most premature kills happen because a noisy early CPA was mistaken for a defect; most expensive slow kills happen because a visible defect was given a fair chance it did not deserve.
Killing the right thing
Before killing a creative, check whether the creative is actually what failed. If every creative in the ad set is dying together, the shared elements are the suspects: audience, offer, landing page, price, or a tracking fault making everything look conversionless. Killing creatives one at a time inside a broken ad set relitigates the same mistake repeatedly and burns a test budget proving it. Similarly, an ad whose spend collapsed without being paused was deprioritised by allocation rather than killed by you, and its sparse data says little. The kill decision is only clean when the ad received real spend inside a functioning structure.
The cost asymmetry worth remembering
Killing too early and killing too late have different price tags. A premature kill wastes the spend already invested and possibly discards a winner, but the loss is bounded by what was spent. A late kill bleeds indefinitely, since a loser left running spends every day. That asymmetry argues for decisive kills at your written thresholds and for resisting the temptation to grant losing ads one more day repeatedly, which is how bounded losses become unbounded. It also argues for launching tests you can afford to complete: the accounts that agonise over every kill are usually the ones running more creatives than their budget can adjudicate.
These thresholds are rules of thumb, and high-AOV accounts with sparse conversions should run them on upstream events or longer windows. No rule replaces checking that tracking works before concluding an ad cannot convert.