First, measure what the extra money bought
Average return across the whole budget is the wrong lens here. Take the period before the increase and the period after, matched as closely as possible on season, promotions and pricing, and calculate the change in spend and the change in sales from your own order or CRM data, not from platform reports. The ratio of the two is the marginal return of the added spend (Marginal ROASExtra revenue returned by the last increment of ad spend.). If it is near zero, the extra money bought almost nothing; if it is positive but below target, it bought something at too high a price.
Then break the added spend down by where it went: which channels, which campaigns, and within each, whether it reached new audiences or people already in retargeting lists and brand searches. Check whether other things changed in the same weeks, such as a price rise, a stock-out or a site release. Finally, check timing: whether the review came before most of the buying cycle had passed. Those three checks point to most of the causes below, and they keep this page separate from a slow, months-long rise in acquisition cost, which the CAC rising diagnosis handles.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Added spend sits in channels that were already large; cost per extra sale far above the average | Marginal cost of the increase in each channel against target | Extra spend past the point of diminishing returns | Move the increment to channels whose marginal cost is below target |
| Most new budget went to retargeting and brand campaigns; platform conversions rose, sales did not | Share of the increase spent on audiences already close to buying | Budget absorbed by existing demand | Cap capture spend; put the increase into reaching new buyers |
| Performance dipped then wobbled after a large overnight budget change | Size and timing of the change against bid strategy status and volatility | Large jumps unsettling automated bidding | Raise in steps; hold each step until results stabilise |
| Frequency climbed while reach barely moved | Reach and frequency for the target audience before and after | Audience already saturated | Widen the audience or add channels before adding weight |
| Traffic and enquiries rose but orders did not; stock-outs, slow pages or full sales calendars | Conversion by step; stock and fulfilment; sales capacity during the period | Business unable to convert more demand | Fix the bottleneck before spending more |
| The decision was taken within days or a couple of weeks | Typical time from first contact to purchase; attribution windows | Sales read too soon | Set the review date by the buying cycle; test with a control |
Fix, verify and prevent each cause
Extra money landing where returns had already flattened
The first pounds in any channel reach the most responsive buyers; each additional pound reaches people less likely to act. An increase applied to channels that were already well funded therefore buys far fewer sales than the average suggests, a pattern of diminishing returnsEach extra unit of spend or effort buys less extra result than the one before it. that the average never shows.
Fix: compute the marginal cost of the increase in each channel and compare it with the target, using the approach in the marginal returns and forecasting guide; keep the increment only where marginal cost is below target and move the rest to channels with headroom.
Verify: after reallocation, total sales rise by more per added pound than in the first period.
Prevent: before approving any increase, ask for the expected marginal cost by channel, not the current average.
New budget absorbed by people already on their way to buy
Retargeting lists and brand search terms show the best ratios in every report, so new money often flows there first, by automated budget rules or by habit. Those audiences were mostly going to buy anyway, so platform conversions rise with spend while total sales do not.
Fix: check how much of the increase reached retargeting and brand campaigns, cap them, and put the increment into reaching new buyers. To settle how much those campaigns truly add, use the design in the remarketing holdout test, which compares an exposed group with a held-out one.
Verify: sales from new customers rise as prospecting spend grows, and the holdout shows the incremental value of the capture campaigns.
Prevent: report new and returning customer sales separately whenever budget changes.
A large jump that unsettled automated bidding
Automated bidding sets bids from predictions learned from past conversion data; Google's Smart Bidding documentation describes strategies that use machine learning to optimise toward conversions or conversion value in each auction. A sudden, large change in budget asks the system to bid at a scale and in auctions it has little recent evidence for, and results can swing while it adjusts.
Fix: move budgets in steps and hold each one long enough to read, as the testing and scaling guide advises for automated campaigns, rather than doubling overnight.
Verify: results at each step settle within the expected range before the next step, and the marginal cost per step can be read.
Prevent: write the step size and holding period into the scaling plan, and never change budget, target and creative on the same day. Inside a Google Ads account, the guide to scaling Google Ads spend orders the alternatives to a single jump, from budget-limited campaigns to looser targets and wider queries.
An audience that was already saturated
If the campaign already reached most of the people it could, extra spend mostly shows the same ads to the same people more often. Frequency climbs, reach barely moves, and the additional impressions add little.
Fix: compare reach and frequency before and after the increase, using the reach and frequency planning described in the media objectives guide, and add new audiences, placements or channels before adding weight to the same one.
Verify: reach among the target audience rises with the next increase, and frequency stays within the range the plan set.
Prevent: set a frequency ceiling in the plan and treat a rising frequency with flat reach as a signal to stop adding budget to that audience.
Demand the business could not serve
Sometimes the ads worked and the business did not. More visitors arrived to slow pages, out-of-stock products or a checkout that failed on mobile; more enquiries reached a sales team with no room in its calendar. The extra demand leaked away before it became sales.
Fix: read conversion step by step for the period, as the funnel analysis tutorial sets out, and check stock, delivery dates and sales capacity for the same weeks; fix the constraint before spending more.
Verify: conversion at the failing step recovers and sales per visit return to the earlier level at the higher traffic.
Prevent: check stock, site performance and sales capacity against the planned volume before any significant increase. For a business that sells booked jobs rather than stock, the guide to unit economics for service businesses shows why the next job is worth less when the diary is full and how to lower the target in those weeks.
Results read before they could arrive
Many purchases follow the first ad contact by days or weeks, and considered purchases by months. Meta's help centre describes attribution settings as a window after a click or view in which conversions are credited, which is a reminder that results keep arriving after the spend. Judged too early, an increase looks like a failure.
Fix: set the review date by the buying cycle and, where the stakes justify it, run the budget increase incrementality test, which raises spend in some regions and holds it in matched ones.
Verify: sales in the test regions are compared with controls over the full window, and the marginal cost per incremental sale is calculated.
Prevent: agree the review date, the control and the success threshold before the budget changes.
A prevention routine for budget increases
Most failed increases could have been predicted, so the routine sits before the money moves. Before any material increase, write down the expected marginal cost by channel, the step size and holding period, the frequency ceiling for each audience, the review date set by the buying cycle, and the control that the result will be compared with. Check stock, site speed and sales capacity against the planned volume. During the increase, track weekly the share of new spend going to retargeting and brand terms, reach and frequency, and conversion step by step; stop adding budget where any of them moves the wrong way. After the review date, compare sales from your own systems with the control, calculate the marginal return, and record the result beside the budget decision so the next increase starts from evidence. Where budgets are large, run an IncrementalityConversions caused by marketing. test at least once a year on the channel most likely to absorb increases.
A homewares brand raises its budget by half
A homewares brand raised monthly paid media from £60,000 to £90,000 ahead of autumn, and sales over the following four weeks rose only from £410,000 to £418,000. Splitting the extra £30,000 showed £17,000 had gone to retargeting and brand search through automated budget rules, where platform-credited revenue rose sharply. A further £9,000 went to a prospecting campaign whose budget was doubled overnight and whose cost per order swung widely for two weeks. Frequency in the main social audience rose from 2.1 to 3.8 a week with reach almost flat. The brand capped retargeting at its earlier level, moved the money into two new audiences in £3,000 weekly steps, and ran a two-region test. Over the next eight weeks, test regions sold 9% more than controls, a marginal return of about 2.6.
Common mistakes
Judging the increase on the average return of the whole budget. Trusting platform conversions over order data. Letting automated rules send new money to retargeting. Doubling a budget overnight. Adding weight to an audience that has already seen the ads many times. Spending more while the site or stock cannot cope. Deciding within days.
