Cut where the last pound earned least
The scaling guide explains why returns fall as spend rises: delivery spends each budget on the people it predicts are most likely to convert at the lowest cost first, so each increment reaches people a little less likely to act or a little dearer to win. That is Diminishing ReturnsEach extra unit of spend or effort buys less extra result than the one before it. at work, and it means every ad set's reported return is an average of good early spend and weaker late spend. A cut should remove the late spend first, because that is where the least is lost per pound saved.
The marginal returns guide shows why the average flatters the margin and how to estimate the marginal figure. For a cut the consequence is direct: removing spend that earned near break-even costs little, removing spend that earned well above it costs a lot. An ad set with a strong average may carry a top slice that barely pays; an ad set with a weak average may be weak all the way through and belong at the top of the list.
This guide carries out Meta's share of a cut. How much the whole marketing budget falls and how it divides between channels is decided before it reaches Meta; the guide to cutting a marketing budget across channels covers that decision, and the guide to cutting a Google Ads budget does this guide's job for Google Ads. Whatever the split, record the move in a budget reallocation memo: the reduction, the evidence of marginal returns behind each cut, the guardrails, the review date and who approved it.
Finding the marginal spend on Meta
The scaling guide's method works in both directions. Take two periods at different budget levels for the same ad set or campaign, with similar seasons and no other significant changes, and divide the difference in revenue by the difference in spend. The result is marginal return on ad spendExtra revenue returned by the last increment of ad spend. for that slice of budget; for lead generation, divide the difference in spend by the difference in qualified leads to get a marginal cost per lead. Compare each figure with the economic limit: break-even return from the break-even calculator, or the most a lead can cost. The ROAS calculator handles the arithmetic. The ROAS calculator set up for a budget cut shows how to enter the difference in revenue and spend between two budget levels and how to read each slice against break-even before deciding.
Most accounts have the evidence already, in the budget steps of past scaling. Where they do not, the places marginal spend usually hides on Meta are these:
| Where to look | What marginal spend looks like | Typical action |
|---|---|---|
| Ad sets added during scaling | A recent budget step whose marginal return was never checked | Return to the previous level |
| Overlapping prospecting ad sets | Two ad sets reaching the same people and competing in the auction | Consolidate into the stronger one |
| Ad sets stuck in learning limited | Budget spread too thin to reach enough events | Consolidate, or pause if the result never paid |
| Retargeting | A high reported return built on people who would have bought anyway | Judge on incrementality, then lower rather than remove |
| Long-running ads | Rising frequency and falling click-through | Retire the tired ads; keep the budget on fresher ones |
Retargeting needs care in both directions. The comparison of prospecting and remarketing explains that remarketing's reported efficiency overstates its incremental value, because many of the people it reaches were already close to buying. That makes it look like the last thing to cut when it may be the first. The honest test is a remarketing holdout, which measures the incrementalConversions caused by marketing. conversions retargeting adds; without one, read the reported return as an upper bound. The prospecting it depends on matters too: a retargeting pool is filled by prospecting, so a deep cut to prospecting shrinks retargeting's results later, as people age out of the pool's retention window and fewer new visitors replace them.
Lower, consolidate or pause
Three levers remove spend, and they touch learning differently.
Lowering an ad set or campaign budget keeps it running with less money. The learning phase guide, citing Meta's documentation on the learning phase, lists a large change to the budget among the edits that can restart learning, while small moves are not listed; Meta does not publish where the line falls. Lowering therefore happens in steps, and the size of each step is a judgement rather than a rule.
Consolidating means fewer ad sets sharing the remaining budget. The learning phase guide treats consolidation as the main structural answer to learning limited, and a cut makes it more pressing, since the same number of ad sets on less money means fewer events for each. The safest form of consolidation is subtraction: keep the stronger of two overlapping ad sets as it is and pause the weaker, rather than editing the survivor's audience, because a change to targeting is on the guide's list of significant edits.
Pausing removes spend completely. It is the right lever for spend that did not pay and will not return. For an ad set that will come back, such as retargeting for next quarter's sale, the learning phase guide notes that pausing for an extended period can restart learning and advises lowering the budget instead of pausing for long gaps.
| Lever | Use it for | Learning consequence, as the learning phase guide states it |
|---|---|---|
| Lower | Spend that still pays but at a lower return than the rest | Large budget changes can restart learning; cut in steps |
| Consolidate | Overlapping or thinly funded ad sets | Pausing one leaves the other untouched; editing its audience would restart learning |
| Pause | Spend that did not pay and will not return | An extended pause before resuming can restart learning |
Campaign budget or ad set budgets when cutting
Where the budget sits decides who chooses where the cut falls. The comparison of campaign and ad set budgets, citing Meta's documentation on Advantage+ campaign budget, explains that with campaign budget the system distributes one figure across the ad sets continuously, towards the best results for the campaign's optimisation goal and bid strategy. Lowering the campaign figure therefore lets delivery decide where the cut falls, usually taking it from the ad sets producing the dearest results. The same guide notes that budget moves between ad sets under campaign budget do not count as edits to each ad set, so the ad sets themselves are not edited by the redistribution. Pausing a weak ad set inside a campaign budget simply returns its share to the others.
With ad set budgets, you choose. Each ad set's figure is lowered on its own, which is the right structure when the cut must fall in a particular place, such as one market or one product line, and the wrong one when you would rather the system found the least costly cut. Each manual change is an edit to that ad set, and a large one can restart learning.
Two cautions from the budget types guide apply. Campaign budget follows the cheapest result for its optimisation event, so a campaign that shares one budget between prospecting and retargeting may protect retargeting's cheap results while starving prospecting; keep the two on separate budgets during a cut. And the campaign figure must still cover Meta's minimum for every active ad set, so a lower campaign budget is another reason to consolidate first. For a hard ceiling on the quarter's total, the campaign spending limit the guide describes stops delivery when reached, which a budget does not; a lowered lifetime budget is re-paced across the remaining schedule, so its pacingHow fast a campaign spends its budget across the flight compared with the plan. adjusts without a daily edit.
What to protect
Some spend and settings cost little and protect everything else. Cut around them.
- Measurement. The pixel, the Conversions API and the monthly reconciliation cost almost nothing to keep and are what make the cut's effect readable. A cut quarter is the worst time to lose the signal.
- Exclusions. Customer lists and recent-purchaser exclusions stop prospecting paying for people who already buy. Keep them current; a stale exclusion wastes a larger share of a smaller budget.
- A creative pipeline at reduced volume. Ads still wear out at a lower budget. Fewer new concepts, made steadily, keep the remaining ad sets from running on tired creative, which the optimisation hub treats as the lever to work hardest.
- Ad sets that have exited learning. Stable ad sets carry the most predictable spend. Cut around them, lower them gently if at all, and avoid other edits to them during the cut.
- The prospecting that feeds retargeting. Without it, the retargeting pool empties and next quarter starts from a smaller base.
Sequencing the cuts
The order matters because each change has to be judged before the next, and stacked edits make the result impossible to read. A sequence that respects the learning phase:
- Agree the target and the period, and write them into the reallocation memo with the guardrails.
- Pause what did not pay: ad sets and campaigns whose marginal return was below the limit and that have no reason to return.
- Consolidate: pause the weaker of overlapping or thinly funded ad sets, leaving the stronger untouched.
- Lower what remains in steps, starting with the slices whose marginal return was lowest, with a wait after each step long enough for results to settle and the attribution window to fill.
- Lower rather than pause anything that will return after the cut.
- Judge each step on marginal figures, then decide the next.
Make one kind of change at a time. The scaling guide warns that scaling while changing creative or bids gives three edits and one result; the same is true of cutting. Record each step in the change history with the reason and the expected effect, so that the review can compare forecast and outcome. Before the first step, the marketing budget cut checklist confirms that the aim, the commitments and the restore triggers are agreed across every channel, not only on Meta. Agreeing each restore triggerThe condition agreed with finance when a budget is cut that says when money comes back and to which line: a measured signal, a threshold and an agreed response. before the first step gives the lowered budgets a known route back, so they are less likely to harden into the new baseline.
Setting expectations for lost volume
A cut loses volume, and the business should agree how much before it starts. Estimate it from the marginal figures: the spend removed from each slice multiplied by that slice's marginal return gives the revenue expected to go, and the same arithmetic in purchases or leads gives the volume. Because the cut starts with the least productive spend, the first savings should cost the least volume, and each further step more.
Three effects make the loss arrive unevenly. Reported results take time to settle, because conversions are credited to the date of the ad interaction and recent days fill in. Retargeting results fall after prospecting is cut, as the pool it feeds shrinks. And ad sets with fewer events per week may become learning limited, which the learning phase guide describes as delivering with less stable predictions. None of this promises that efficiency will hold. Report the cut against the forecast, in volume and in return, and judge the business effect on the shop's or CRM's own numbers rather than on Meta's reported figures alone, since retargeting's easy credit falls away faster than the sales it claimed.
A granola brand cuts a quarter's spend (illustrative)
Honeyslate Granola cuts Meta spend by a third for a quarter
Honeyslate Granola, an invented granola brand, had to cut Meta spend from £1,200 to £800 a day for a quarter. Prospecting ran as one campaign with Advantage+ campaign budget at £950 a day across three ad sets; retargeting ran separately on an ad set budget of £250. Last quarter's budget steps gave a marginal return for each slice against a break-even return of 1.9, and a holdout the previous year had put retargeting's incremental return at about 2.1 against a reported 6.8.
| Ad set | Daily spend before | Marginal return | Decision | Daily spend after |
|---|---|---|---|---|
| Interest stack (prospecting) | £200 | 1.2 | Paused | £0 |
| Customer lookalike (prospecting, overlapping broad) | £250 | 1.5 | Paused to consolidate | £0 |
| Broad (prospecting) | £500 | 2.6 | Left untouched; received the campaign's remaining budget | £650 |
| Retargeting | £250 | 2.1 incremental | Lowered in two steps, not paused | £150 |
The team paused the two weaker ad sets inside the campaign, which moved their share to the broad ad set without editing it, then lowered the campaign budget from £950 to £800 and, a week later, to £650. Retargeting went from £250 to £200 and then £150, since it would be needed for a spring promotion. The forecast, which assumed the broad ad set's extra £150 a day would return about 2.0, put incremental revenue lost at about £525 a day, roughly 16 orders a day at a £32 average order, or about 1,490 orders over the quarter. The shop's own data showed about 1,180 fewer orders than the same quarter's plan, while Meta-reported purchases fell by about 2,050, most of the difference being retargeting credit for people who bought anyway. Return on the reconciled figures rose slightly, which the team recorded as a result rather than a promise for next time.
Common mistakes
- Cutting every ad set by the same share. It keeps the weakest spend and trims the best.
- Judging on average return. The cut is decided by what the last slice earned.
- One large cut on a working ad set. Meta documents that large budget changes can restart learning; step down.
- Pausing ad sets that will return. An extended pause can restart learning; lower them instead.
- Editing the survivor when consolidating. Pause the weaker ad set rather than rewriting the stronger one's audience.
- Protecting retargeting because its reported return is high. Judge it on incrementality.
- Cutting prospecting to nothing. The retargeting pool it feeds empties as people age out of it.
- Dropping measurement or exclusions to save money. They are cheap and they make the rest of the budget work.
- No forecast of lost volume. Without one, nobody can tell whether the cut went as planned.
