Which Facebook Ads Should I Turn Off, Keep or Scale?

You open Ads Manager on a Tuesday and every row is asking for a decision. One ad has spent $90 with nothing to show, another looks brilliant after two days, and a third has barely spent at all. If you act on each number as it arrives, you kill healthy ads, reset learning you paid for and scale lucky streaks.

Most Facebook ads on most days should be left alone. Judge every ad in multiples of your target cost per acquisition rather than days. Turn off only what has spent two to three times that target with nothing to show. Scale only proven winners, in small reversible steps, and judge them on blended efficiency rather than platform ROAS.

How much spend before you can judge an ad?

Judge an ad by what it has spent relative to your target CPA, not by how many days it has run. A day is not a unit of evidence. An ad on $10 a day and one on $200 a day have learned very different amounts by Thursday. What you are really waiting for is conversions, and you buy those at roughly your Meta ads cost per result. So the honest measure of patience is how many times over an ad has spent what one result should cost.

The arithmetic is simple and worth knowing. Suppose an ad truly converted at exactly your target CPA and purchases arrived at random. After spending one target CPA, it would still show zero conversions about 37% of the time. At two times target that falls to about 14%, and at three times to about 5%. That is why two to three times target CPA is enough to cut a non-converter.

Naming a winner is a different and far more expensive question. Telling ad A from ad B usually takes a few dozen conversions each, which at a $40 target means roughly $1,200 per ad. The full working, including when to move the threshold, is in how much a Facebook ad should spend before you judge it.

That asymmetry is the spine of every kill or scale decision for Facebook ads. Turning an ad off is a cheap decision to get right, and scaling one is expensive. Between the two sits a wide band of ads that have not yet earned either verdict. The correct action for them is to keep running and do nothing.

Time still matters at the edges. Give any test a few full days so weekday and weekend behaviour both show up. Three situations stretch the threshold further:

• High-ticket products with long consideration cycles. Wait for conversions to catch up after the spend is in.
• Substantial edits mid-test. Earlier spend now describes the old version, not the one running.
• Small retargeting pools. These hit the threshold fast but decay as the same people see the ad again. That decay is fatigue, not a failed test.

Faster signals arrive sooner than purchases and help when conversions are thin. Click-through rate, cost per landing page view and add-to-cart rate all fall into this group. They tell you which creative earns attention, though, not which one earns money. Treat a proxy leader as a candidate rather than a winner.

Write your thresholds down before launch. The exact multiple matters less than whether you still honour it when an early number tempts you.

How many ads can your budget actually test?

Your budget can test as many ads as it can fund to a verdict. For most accounts, that means three to five distinct ads per ad set. The ceiling comes from one division: weekly test budget, divided by target CPA multiplied by the conversions you need before judging an ad.

Take a hypothetical account spending $1,400 a week, with a $35 target and a 10-conversion rule. It can fund four ads. Upload eight and you have not doubled the test. You have halved the evidence behind every decision. The detail article on how many Facebook ads your budget can test at once walks through the adjustments.

The three-to-five range matches what John Ho recommended in audit findings reported by PPC Land in September 2025. Ho is a demand generation specialist at Labelium Singapore. In his view, one ad per ad set tests nothing, and 15 or more fragments the data.

Some high-volume creative agencies load far more into a single testing ad set. Read side by side, the two positions agree more than they appear to. Those frameworks assume budgets large enough to fund each ad, and they cull within days. The right count scales with spend.

The cluster's budget rules need reconciling here, because they quietly fund different verdicts. There are three tiers:

• Cutting a loser needs only two to three times target CPA per ad.
• A 10-conversion rule gives a sturdier read on whether an ad is roughly on target.
• Crowning a winner needs a few dozen conversions per ad.

At that last tier, the same $1,400 does not cover even a two-ad winner comparison in a week. Decide which question this week's test is answering before you count how many ads your Facebook ads budget can hold.

Two structural habits keep the maths honest.

First, ring-fence testing in its own campaign or ad set, separate from the one scaling proven ads. New creative then cannot steal delivery from winners, and winners cannot starve the test. This also sidesteps a learning-phase trap. Adding a new ad to an ad set restarts its learning, so dropping test creative into your best performer resets the thing you are trying to protect.

Second, count concepts, not variants. A new hook or format is a test. A headline swap on the same video mostly competes for the same delivery.

If the division comes out below two, test fewer ads rather than cheaper ones. Alternatively, judge on a faster signal and give the survivor a purchase-level test before it earns scale.

When should you turn off a zero-conversion ad?

Turn off a Facebook ad with no conversions once it has spent about three times your target CPA without a purchase. Pause it rather than delete it. Before that point a zero is mostly noise. After it, a healthy ad would show nothing only about one time in 20.

With a $40 target, that line sits at roughly $120 of the ad's own spend, counted on completed days. The checks to run first, and the cases where the line should move, are in when to turn off a Facebook ad with no conversions.

Knowing when to turn off Facebook ads is as much about those checks as about the threshold:

• Today's data. Spend registers faster than conversions, so any window that includes today understates results.
• Attribution lag. A conversion is credited back to the click, so yesterday's numbers are still filling in.
• The attribution window. If your buyers take most of a week and you judge on a one-day click window, the ad can be selling just outside the frame.
• The funnel. If every ad in an ad set shows zero purchases at several times target, suspect the pixel, the checkout or the landing page before the creative. New creative sent into a broken funnel fails the same way.

The spend and zero-conversion articles differ slightly on where in the two-to-three range to cut. The difference comes from how often you look. Every check gives a healthy ad another chance to trip the line through bad luck. A weekly manual review can cut at the lower end. A daily habit, or an automated rule, should sit at three times or above. Otherwise you will kill a steady fraction of good ads purely by looking.

They also seem to disagree about starved ads, the ones Meta barely funded. One says an ad that spent $8 has not been judged. The other says let it go. Both are right, for different goals. In an ordinary creative rotation, Meta's preference is itself a signal, and rebuilding structure around a passed-over ad is sunk cost dressed as testing. When you genuinely need a head-to-head, test in separate ad sets with matched budgets instead.

Ads that convert, but badly, follow the same logic. One sale at three times target means the ad is running at roughly triple your cost. That is not proof it can never work, but it is enough to stop funding it while better ads wait for budget.

Judge prospecting and retargeting separately, though. Cold audiences convert later and less often. Held to a retargeting standard, they look dead when they are only slow.

Should you edit ads during the learning phase?

Usually not. Unstable, higher costs are what the learning phase looks like. A significant edit does not pause Meta's count of optimisation events; it zeroes it.

These all restart the clock: changes to targeting, placements, the optimisation event, the bid strategy or the creative, new ads, large budget swings and long pauses. Renames, small budget nudges and short pauses generally do not.

The test is simple. If the problem would still be a problem once learning ended, fix it now; otherwise, wait. The cases that justify a reset are set out in whether to change your Facebook ads during the learning phase.

This is the "keep" in keep, cut or scale. It is the hardest of the three, because it means doing nothing on purpose. During learning, the delivery system is exploring audiences, placements and times of day. Judge it on day two and you are judging the experiment, not the result. Edit every couple of days and the ad set never builds a clean week of events, so it never exits. Plenty of ad sets that look stuck are simply over-edited.

Budget is where the guidance gets vague. Most practitioners quote a percentage above which a budget change resets learning. None of the guides we reviewed points to Meta documentation for the exact figure. Assume a large swing is a reset and a small nudge is not. Don't toggle an ad set off and on to "refresh" it either, because a long pause followed by a restart counts as a significant edit.

Two situations justify acting anyway.

A broken setup. The wrong optimisation event, a dead URL, conversions that Events Manager shows are not firing, an unintended geography, or a spend rate the business cannot absorb. These are wrong at any stage, so the reset is the cheaper loss.

Learning Limited. This status is a forecast rather than a penalty. Divide weekly budget by cost per optimisation event. If the answer falls well short of the threshold, waiting will not help. Consolidating ad sets, broadening the audience or loosening a cap is worth its reset. Read the event threshold Ads Manager shows for your own ad set rather than one from a blog. Practitioners report the requirement has changed unevenly for some campaign types.

The cluster surfaces a genuine disagreement here. When conversions are thin, one fix is moving the optimisation event up the funnel, to add-to-cart for instance. Some practitioners treat that as the fastest route to signal. Others find it lowers customer quality, because the system finds people who add to cart rather than people who buy. It is the same tension as judging tests on proxy metrics.

Our position is consistent across both. Move up only if the earlier event reliably predicts revenue in your account. Then judge the outcome on sales, not the label in the Delivery column.

When you do edit, batch every change into one. Three tweaks on three days restart learning three times.

How do you scale a winner without breaking it?

Raise the budget in modest steps spaced a few days apart. Widen the audience before it saturates, add fresh creative as frequency climbs, and judge the result on blended efficiency rather than in-platform ROAS. ROAS almost always dips as spend rises, because each additional customer costs more than the last. That dip alone is not failure.

The most widely shared rule of thumb, Robb Fahrion's, lifts budgets by 20–30% every few days. No published guide backs any increment with test data. Treat the percentage as a habit and reversibility as the principle, as scaling a winning Facebook ad without breaking it explains.

Scaling is the only one of the three decisions that changes the conditions the ad was judged under. Double the budget and three things shift together:

• Delivery can drop back into learning.
• The audience gets colder, because the most responsive pocket was reached first.
• Frequency climbs inside a fixed pool.

A big jump triggers all three at once and leaves you unable to tell which one hurt. A small step lets you see. A large budget change is also a significant edit, so the learning-phase rules apply here too. That is the real reason the steps stay small.

Two checks come before the first step. First, confirm the win is real. An ad that has not cleared the few-dozen-conversion bar from the first section is a lucky streak, and scaling multiplies the luck along with the spend. Second, consolidate before you spread by cutting the weak siblings that dilute the winner's conversion volume.

Audience and creative need to move with the budget. Exclude past purchasers and other warm audiences from prospecting. Otherwise scaled spend flatters ROAS by reaching people who would have bought anyway.

Practitioners disagree on whether fatigue is creative burnout or audience exhaustion. When you are scaling, both are usually true. Build new hooks and formats on the working concept and run them alongside the original. Placements help as well, because a concept worn out in the feed can still have room in Stories or Reels.

Diagnose fatigue from signals moving in sequence: rising frequency, then falling CTR, then rising CPA. One bad day proves nothing.

Finally, measure on the business. For e-commerce, that means MER: total revenue divided by total ad spend across every platform. If Meta's ROAS drops but MER holds above target, the extra spend is paying for itself. If CPA drifts after a step and stays there, return to the previous budget.

Calculating MER means having spend and revenue in one place. SproutMe Companion answers it from a data lake that joins Meta and Google Ads spend to Shopify revenue, rather than from three exported reports.

Conclusion

Most Facebook ads optimisation goes wrong on timing, not judgment. Ads get cut before a zero means anything, edited while learning, and scaled on a handful of conversions. The fix is one consistent unit, target CPA, applied differently to each decision.

Cut at two to three times target with nothing to show, once tracking and attribution check out. Leave learning ad sets alone unless something is broken or Learning Limited arithmetic rules out recovery. Scale only what has cleared a winner's bar, in steps you can reverse, measured on MER or acquisition cost. Everything in between is a keep.

See how SproutMe Execute adjusts budgets and creative continuously within the spend limits and approval rules you set, rather than waiting for a weekly review.