Cut where the last pound earned least
Every campaign's reported return is an average of good and poor spend. The first pounds buy the most relevant searches from the people closest to buying; the last pounds buy less relevant searches, lower positions and auctions the bidding judged only just worth entering. Because of Diminishing ReturnsEach extra unit of spend or effort buys less extra result than the one before it., the last pounds always earn less than the average shows, and those are the pounds a cut should remove first.
The marginal returns guide explains why the average flatters the margin and how to estimate the marginal figure from the account's own history. For a cut, the consequence is simple. Removing a pound that earned close to break-even costs almost no profit; removing a pound that earned far above it costs a lot. A campaign with an excellent average can still be carrying spend at its edge that barely pays, and a campaign with a mediocre average may be poor all the way through.
The aim, then, is a list of spend ranked by what its last increment earned, cut from the bottom until the saving is reached. The same logic decides cuts across channels, which the budget allocation guide applies when the reduction is to the whole marketing budget rather than to Google Ads alone. The guide to cutting a marketing budget across channels adds what a single platform cannot show: commitments, work that is slow to rebuild and the triggers for restoring spend. The 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. for the Google Ads lines is agreed there too, as part of the whole cut rather than for one platform.
Find the marginal spend
Marginal spend hides in five places, and the account's segment reports show each.
| Where to look | What marginal spend looks like | Typical action |
|---|---|---|
| Campaigns | Average cost per conversion near or above the ceiling; a large recent budget increase whose step was never judged | Trim the budget or tighten the target; pause if poor all the way through |
| Keyword themes | Search terms that convert rarely; broad match reaching far from the core | Negatives; remove or narrow the weakest keywords |
| Networks | Search partners or Display placements with weaker results than Google search | Turn the network off for the campaign, or give it its own small budget |
| Locations | Areas with high cost and few customers; presence-or-interest traffic from outside the catchment | Exclude the weakest areas; set the location option to presence |
| Time | Hours or days when conversions are rare or nobody can respond | Narrow the ad schedule |
For campaigns that are limited by budget, the budget simulator Google provides estimates conversions at other budget levels, which turns the question "what would a lower budget lose" into a number the platform will commit to as an estimate. For campaigns that are not limited, the history of past budget and target changes is the evidence: a period when spend was lower, with similar season and no other changes, shows roughly what the top slice of spend added.
Rank the findings by the return of the spend they would remove, using cost per conversion (CPACost of each conversion.) for lead accounts and return on ad spend (ROASRevenue generated per unit of ad spend.) for sales accounts, set against the ceiling the bidding hub derives from margin: the maximum cost per acquisition or the break-even return. Exclusions such as poor networks, areas and searches usually come out at the bottom of the list, because they buy the least and cost the least to remove.
Lower budgets or tighter targets
Once the exclusions are done, most of the remaining saving comes from two levers that remove different spend.
A lower budget. Google documents the budget as an average daily amount with a monthly limit, and the limit is recalculated when the budget changes, so a lower budget lowers the ceiling for the rest of the month at once. The campaign then spends less, and if the new budget is below what it could spend, it becomes limited by budget. The budgets guide observes that a limited campaign spreads its budget through the day and may stop serving earlier, which removes auctions by when they happen rather than by what they are worth. Under maximise conversions or maximise conversion value without a target, the budget is the main lever and this is the natural way to cut.
A tighter target. Google documents that Smart Bidding sets a bid for each auction from a prediction of conversion and value, aiming at the target as an average. A stricter target CPA or a higher target ROAS means fewer auctions clear the bar; the ones dropped are those the model predicted would pay least. Spend falls as a consequence, and the budget may go unspent, which is the intended outcome of this kind of cut. It aims more precisely at the marginal spend than a budget cut does, at the cost of a learning period after a large change.
| Row label | Lower budget | Tighter target |
|---|---|---|
| What falls | Spend, directly | The auctions entered, and spend with them |
| Which auctions go | Those that would have come after the budget ran out | Those predicted to fall short of the new target |
| Works on | Any campaign; the natural lever without a target | Campaigns bidding to a target CPA or ROAS |
| Fails when | The campaign was not spending its budget anyway | The new target is far beyond what the campaign has achieved |
| Watch for | Limited-by-budget status and lost share to budget | Learning status; spend collapsing rather than easing |
The two can be combined, but not on the same day. A target-based campaign that is not limited by budget will not notice a budget cut until the budget falls below what it was spending; there, the target is the lever.
What to protect
Some spend looks cuttable on the reports and is not.
Brand terms. Brand campaigns report high returns because people searching the name were likely to buy. Whether the ads add sales beyond what the organic listing would have captured is the incrementality question the branded versus non-branded search comparison frames, with a region-based test to answer it. Until that test has been run, neither keep Branded SearchSearch queries that contain a company's or product's brand name. because its return looks high nor cut it because the return is assumed to be free: decide after the answer, and watch whether competitors take the position.
Campaigns whose data feeds a portfolio. The Smart Bidding strategies guide describes portfolio strategies as Google documents them, a way to pool data across campaigns under one target. Pausing one member thins the data the others learn from, so cut within the portfolio before removing a member from it.
Campaigns with incomplete tracking. Phone-led campaigns or those whose sales close offline may look expensive because the account sees only part of their results. Check the business's own records before cutting them on platform figures alone.
Campaigns building future demand. New-customer, launch and demand-creation campaigns are judged on a longer horizon. If they are cut, the decision should say so, rather than letting them fall on a short-term ranking they were never meant to win.
Pausing versus trimming
A trim lowers a campaign's spend and keeps it running; a pause stops it. The two are not interchangeable.
Pause when the spend is poor all the way through: the average, not just the margin, is outside the ceiling, and there is no strategic reason to keep it. Pausing removes the spend cleanly and frees the attention it took. Trim when the campaign is profitable on average but its last increment is not: the campaign keeps its history, its learning and its best auctions, and loses only the edge.
Pausing has costs that trimming avoids. A paused campaign stops gathering the data its bid strategy learns from, and the budgets guide notes that missed days are not recovered later, because the monthly limit is a ceiling rather than a target. Restarting brings its own learning period. For ads, keywords and ad groups inside a campaign, the pausing underperforming ads SOP sets the thresholds in advance, checks the data before acting and logs every pause, which keeps a budget cut from turning into a series of unrecorded decisions.
Sequencing cuts to limit relearning
Google's Smart Bidding documentation lists creating or significantly changing a strategy, changing conversion actions and large changes to targets or budgets among the changes that can start a learning period. A cut made of many large changes at once can put the whole account into learning at the moment it has least money to absorb the noise.
A workable sequence runs in four stages:
- Exclusions first. Negatives, networks, locations and schedule narrow what the campaigns buy without being among the listed learning triggers, and they remove the cheapest-to-lose spend.
- Then trims on campaigns without targets. Lower budgets in steps, judging each step on the month-to-date spend and the conversions it bought.
- Next, target changes one campaign at a time. Tighten in steps, leaving each campaign to settle before the next, and never on the same day as its budget change.
- Pauses last. Only what the earlier steps showed to be poor throughout.
Note every change in the change history with its reason, and judge each step after the conversion delay has passed, not on the days immediately after it. Where the cut is urgent and must happen in one move, accept the learning period and say so in advance, rather than reacting to the noise with further changes. The marketing budget cut checklist is the gate to tick before the first of these changes, with the cross-channel checks beside the account ones.
Setting expectations for lost volume
The people who asked for the cut need to know what it will cost, before it happens. Estimate it from the marginal figures, not the averages: the conversions lost are roughly the spend removed divided by the marginal cost per conversion of that spend, which is fewer conversions than the average cost would imply. That is the case for a targeted cut, and it is also the honest forecast. The CPA calculator does the arithmetic.
Three further points belong in the message. Results will be noisy for a period after any large change, so the first fortnight is not the verdict. Some lost conversions may not show in platform reports at all, if they were phone or offline sales. And the cut is reversible: record the starting settings so that, when money returns, the same steps can be reversed in order. The tutorial on restoring paid media spend after a cut uses that record across Google Ads and the other platforms once a restore trigger is met, bringing each line back on its marginal evidence rather than simply in reverse. A short written note in the shape of the budget reallocation memo, stating the move, the evidence, the expected loss and the review date, turns a cut into a decision that can be checked.
A pet-treats shop cuts a quarter's budget (illustrative)
Biscuitvane Pet Treats removes a quarter of its paid search spend
Biscuitvane Pet Treats, an invented online pet-treats shop, has to cut paid search from £10,000 to £7,500 a month for a quarter. Its gross margin of 45% puts break-even ROAS at about 222%. The team ranks spend by an estimate of the return on its last increment, from past budget changes and the simulator, and cuts from the bottom.
| Area | Monthly spend | Average ROAS | Estimated marginal ROAS | Action | Spend saved | Revenue lost |
|---|---|---|---|---|---|---|
| Dog treats, generic Search (target ROAS) | £4,000 | 450% | 273% | Target tightened from 400% to 480% | £1,100 | £3,000 |
| Cat treats Search (no target) | £1,700 | 320% | 250% | Budget trimmed to £1,100 | £600 | £1,500 |
| Puppy training treats Search (no target) | £1,500 | 300% | 231% | Budget trimmed to £700 | £800 | £1,850 |
| Shopping, best sellers | £2,000 | 520% | Above 400% | Unchanged | £0 | £0 |
| Brand Search | £800 | 1,200% | Untested | Protected until a holdout test | £0 | £0 |
The plan saves £2,500 a month for an estimated £6,350 of lost revenue, a loss of 13% of revenue for a 25% cut in spend, because every pound removed had been earning close to break-even. The target change was made in the first week and the two budget trims a week later. The generic campaign showed a learning status for about ten days and missed its estimate in that fortnight; over the quarter, revenue fell by £6,900 a month against the £6,350 estimated, and blended ROAS rose from 479% to 547%. At the 45% margin, the cut reduced gross profit after ad cost by a little over £600 a month: a modest price for £2,500 of cash, and far less than an equal 25% cut across every campaign, brand and best sellers included, would have cost on the team's own estimates.
Common mistakes
- Cutting every campaign by the same share. Good and poor spend removed alike.
- Ranking by average return. The average hides the edge where the cheap cuts are.
- Cutting the budget of a campaign that was not spending it. Nothing changes until the cut bites; the target was the lever.
- Tightening the target in one large jump. Spend collapses and the learning period hits at the worst time.
- Changing budget and target on the same day. Two causes, one result, no way to tell them apart.
- Cutting brand terms on assumption. Test incrementality before deciding either way.
- Pausing a portfolio member first. The rest of the portfolio loses data it learns from.
- Promising a proportional loss. A targeted cut loses less than the average implies, and stakeholders should hear the honest forecast.
