Before you cut Q4 ad spend, check three numbers

Not on the ad platform's ROAS alone. A cut is a bet that the spend was not causing sales, and in a quarter when ad prices are up and demand is soft, the number that tells you to cut is the one most likely to mislead you.

What is different about Q4 2026?

Three things, all dated:

1. Ads cost more. Meta reported that its average price per ad rose 12% year on year in the second quarter of 2026 while ad impressions grew 14% (Meta, 29 July 2026). If prices stay 12% higher than a year ago, a flat budget buys about 10.7% fewer impressions, because one divided by 1.12 is 0.893. It is a worldwide average across Meta's apps, not a European one, and the latest figure Meta has reported.
2. Buyers are more cautious. Consumer confidence fell in Germany, the euro area and the US in September, and German willingness to save reached its highest since the 2008 crisis (GfK, 25 September 2026).
3. The calendar is crowded. Amazon's Prime Big Deal Days run on 6 and 7 October 2026 in 22 countries, including the Netherlands, Germany and the United Kingdom, though dates can vary by country (Amazon, announced 14 September 2026). Black Friday is 27 November and Cyber Monday 30 November.

Each of those tends to push your ROAS down, and none of them says a channel stopped working. That is the trap: the number is right that spend is buying less, and wrong that the fix is to spend less.

Should I cut ad spend in Q4 when demand is weak?

Not until you know what the spend was causing. A cut removes ads that were creating sales and ads that were only collecting them, and the platform cannot tell you which is which, because it grades its own homework. Three numbers can, in part.

Which three numbers should you check first?

Against the same weeks last year, by market:

1. Blended MER, total revenue divided by total ad spend. It counts each order once, so no platform can claim a sale another one already took. If MER falls while a platform's ROAS holds, that platform's number is not tracking your revenue, and the gap is where credit may be going to orders that were coming anyway. Blended MER vs platform ROAS has the sum, and the blended ROAS calculator does it for you.
2. New-customer CAC against what a first order returns. Blended CAC hides what a new customer costs when existing customers carry the order count. Set new-customer CAC against blended CAC, then set it against first-order contribution margin from the contribution margin calculator.
3. Days to convert. If buyers who saw two or more channels take longer than they did last year, spend switched on late reaches them after the promotion ends (the lag in GA4).

How do you test a cut without guessing?

Switch the channel off in some regions and compare total orders there with the regions where it stayed on. That is a holdout test, and it answers what the channel added to your orders, which no dashboard does. It needs as many days as your order volume requires, and the holdout test planner counts them.

Promotions blur every comparison. With Prime Big Deal Days from 6 October and Black Friday on 27 November, the weeks before 6 October are the cleanest reference you will have this quarter, and the last day to start a holdout before Black Friday counts back from the peak.

If you cannot run one:

1. Cut the smallest channel you can afford to lose first, and watch total orders, not the platform's.
2. Change one channel at a time, and write down the date.
3. Leave tests, tracking and attribution settings alone from 20 November (what to leave alone the week before).

Where a read fits

Later, once you sell through two or more channels and have a few months of GA4 history, a causal attribution read like Causality Engine's can take the channel question off your hands before the peak. It reads the GA4 Attribution paths export and shows what each channel caused next to what last-click gave it, with Direct split back to the channels that sent those buyers. Every channel gets a data-health score from 0 to 100 and a next step. It takes 1 to 2 minutes and costs EUR 99 once per upload, excluding VAT, with a full refund within 30 days, no questions asked. The first finding is free: your browser works out how many days buyers who saw two or more channels take to convert, and the file never leaves your machine.

Sources, accessed 30 September 2026: Meta Q2 2026 results (29 July 2026); Amazon (14 September 2026); GfK, via InvestingLive (25 September 2026).

Related answers

• Blended MER vs platform ROAS: which should drive spend?
• New-customer CAC vs blended CAC
• The last day to start a holdout test before Black Friday
• Confirm a weak channel with a holdout test
• Black Friday 2026: four measurement deadlines come first
• Q4 attribution checkup: what to check before Black Friday