When to Cut a Campaign: Statistical Patience vs Burning Cash
A worked decision framework for Shopify advertisers on when a campaign has genuinely failed versus when it just needs more data.
Written by Mantas Jurgutis — Founder, Adsify — builds the Google & Meta automation merchants use daily
Editorially reviewed by Adsify Editorial on July 1, 2026 — Reviewed against Shopify, Google Ads and Meta official documentation.
The tension every advertiser faces
Cut a campaign too early and you kill something that would have worked with more data — Google's and Meta's own documentation consistently notes that new campaigns need time and volume to exit the learning phase and stabilize. Cut too late and you burn real cash on a campaign that was never going to work. Neither 'give everything three months' nor 'kill anything under 2x ROAS by day 5' is a real framework — both ignore the actual statistical and platform-mechanical reasons performance data becomes reliable at different speeds for different campaigns.
The minimum data threshold, restated
As covered in the minimum viable test budget analysis, roughly 20 conversions gives a first directional read and 30-50 gives a materially more reliable one. Before any kill decision, check whether the campaign has actually reached even the 20-conversion floor — a campaign with 6 conversions at a bad CPA has not failed, it has produced an unreliable sample, and killing it is a decision made on noise, not signal.
Worked example: the premature kill
A campaign spends $340 over 4 days, generates 5 conversions at a blended CPA of $68, against a break-even CPA of $45. It looks like a clear loss. But with only 5 conversions, the 95% confidence interval around that $68 CPA estimate is extremely wide — a binomial-style sampling error at this volume means the true underlying CPA could plausibly range from roughly $35 to $110. Killing this campaign after 4 days spends $340 to learn essentially nothing definitive, and if the true CPA is actually closer to $40, the store just walked away from a profitable channel based on a coin-flip-sized sample.
Worked example: the justified kill
Compare a campaign that spends $1,600 over 18 days, generating 34 conversions at a blended CPA of $52, against the same $45 break-even. This sample size is well past the 20-30 conversion floor, and the CPA is running 15.6% above break-even with reasonable statistical confidence behind it. This is a legitimate kill (or restructure) decision — the data is now solid enough that continued spend at the same targeting and creative is very likely to keep producing above-break-even CPA rather than randomly correcting itself.
The cost of continuing past a justified kill point
Once a campaign has crossed the 30-50 conversion threshold with CPA clearly and consistently above break-even (not a single bad week, but a sustained trend), every additional dollar spent is a near-certain, quantifiable loss rather than a calculated risk. Worked example: continuing the above campaign for another $1,600 at the same $52 CPA rate produces roughly 31 more conversions and, at a $7 per-conversion loss versus break-even ($52 − $45), a further $217 in realized loss beyond what a kill decision at the 34-conversion mark would have avoided.
Distinguishing 'bad campaign' from 'bad creative'
Before killing a campaign entirely, separate whether the underlying audience/targeting is unprofitable versus whether the specific creative is underperforming — these require different fixes. If CTR is well below the account average alongside a high CPA, the issue is more likely creative (low engagement suppresses delivery efficiency on Meta specifically, and low Quality Score/Ad Rank effects raise CPC on Google); if CTR is normal but CVR after the click is poor, the issue is more likely landing page or offer misalignment with the audience, not the targeting itself.
Worked example: creative fix vs full kill
A Meta campaign shows CTR of 0.6% against an account average of 1.4%, with CPA at $61 against a $45 break-even. Rather than killing the campaign (targeting/audience), swapping in 2-3 new creative concepts while holding the audience and budget constant is the lower-cost diagnostic step — a full audience rebuild costs the sunk learning-phase investment already made, while a creative swap within the same ad set often preserves some of that delivery history. If CPA doesn't improve after another 20-conversion test window with new creative, that's stronger evidence the underlying audience itself is the problem.
Using confidence bands instead of point estimates
Rather than comparing a single CPA number to break-even, build a rough confidence range using sample size: at under 15 conversions, treat any CPA reading as low-confidence and avoid acting on it beyond flagging for continued monitoring; at 15-30 conversions, treat readings as medium-confidence and act only on results more than roughly 25-30% away from break-even; at 30+ conversions, treat as higher-confidence and act on results more than roughly 10-15% away from break-even. This tiered approach formalizes the 'wait for more data' instinct into a specific, repeatable rule.
Budget caps as a patience mechanism
One practical way to enforce statistical patience without open-ended cash risk is setting a maximum test spend cap upfront (as covered in the minimum viable test budget framework) rather than an open-ended 'we'll watch it' approach — a $1,200 cap sized to reach roughly 25-30 conversions at estimated CPA gives the campaign room to reach a reliable sample while creating a hard, pre-committed stopping point if it isn't working, removing the emotional decision-making that happens when a campaign is evaluated day by day with no predetermined ceiling.
The sunk cost trap in the other direction
The inverse mistake is equally common: continuing to fund a campaign well past its 30-50 conversion confidence threshold because 'we've already spent $2,000 on it' or because a founder is emotionally attached to a specific audience or creative concept. Sunk cost has no bearing on whether the next dollar spent will be profitable — only the break-even math and current, statistically adequate CPA data should inform that decision. A useful discipline is writing down the kill threshold before launching the campaign, exactly as described in the pre-test checklist, so the decision at the 30-conversion mark is mechanical rather than emotional.
Automating the patience/kill decision
Because the correct decision depends on conversion count thresholds and confidence bands rather than simple daily spend caps, rules-based automation needs to be configured accordingly — Adsify's optimizer, running every 6 hours, and its auto budget scaling rules can be set to hold a new campaign's budget flat until it clears a defined conversion threshold before allowing any scale-up or scale-down action, which mechanically enforces the statistical patience this article argues for without requiring a person to manually track conversion counts against a spreadsheet threshold.
A simple decision tree
At any check-in: if conversions are under 15, hold and keep monitoring regardless of CPA. If conversions are 15-30 and CPA is within roughly 25% of break-even, extend the test to 30+ conversions before deciding. If conversions are 15-30 and CPA is more than 25% above break-even, pause and diagnose (creative vs targeting) before restarting. If conversions are 30+ and CPA is more than 10-15% above break-even on a sustained basis, kill or fundamentally restructure the campaign. If conversions are 30+ and CPA is at or below break-even, move to scaling decisions rather than further patience.