Why CPMs rise in Q4, and when the ramp really starts

Q4 ad costs start rising in the last week of October, about four weeks before most plans assume. Two separate curves get treated as one: the price of an impression, set by every other advertiser's budget calendar, and the willingness of your customers to buy. They peak at different times, and the gap between them is where the money is.

When do Q4 ad costs actually start rising?

The last week of October. Not December, and not the week of Black Friday.

The reason is that the auction price is set by other advertisers, and other advertisers work to a budget calendar. Annual holiday budgets release at the end of October across retail, and the effect on the auction is immediate: the same inventory, considerably more money chasing it, with the demand it's chasing still a few weeks away. By the time Black Friday arrives the ramp is most of the way done.

That timing has a direct operational consequence. Anything you want to learn cheaply has to be learned before that week, because after it you're paying peak prices for information rather than for sales.

Is the auction getting more expensive, or are your customers?

Both, on different schedules, and nearly every Q4 guide collapses them into one line about CPMs going up.

The auction ramp is a price you pay per impression. It's driven by competitors' budget release dates and it's indifferent to whether anyone is buying. It starts in late October and peaks around Cyber Monday.

The demand ramp is your customers' willingness to buy. It's driven by intent, it arrives later and it holds through the last shipping cutoff in mid-December long after the auction has started to fall.

Because they don't coincide, the quarter has four distinct regimes rather than one expensive month. Late October and early November is the worst ratio of the year: rising prices against demand that has not fully arrived. Cyber Week is maximum on both, so conversion rate is what saves you rather than CPM. The fortnight after Cyber Monday is the best ratio of the quarter for anyone who can still ship. January is cheap on both, which sounds better than it is.

This is the whole argument for not steering on CPM in Q4. Conversion rates rise in November for exactly the same reason prices do, and an account reporting a CPM up 80% with a CPA up 15% is having a good quarter that its dashboard is describing as a bad one. MER, ROAS and what to steer on covers which number to hold onto when in-platform attribution gets noisy, which it does most in the weeks when the most is riding on it.

What did CPMs actually do across the last quarter?

The most-repeated figures come from Jon Loomer's own 2024 tracking: Instagram CPM up 65% year on year by 1 December, Facebook up 38% by 2 December and an average CPM moving from roughly $7 on 8 November to roughly $15 on 3 December. Treat those as unverified. They're secondhand everywhere they appear and we haven't been able to check the precision against a primary dataset. What's well established across every practitioner account is the shape — a roughly two-fold climb inside four weeks, then a fall. The exact multiple isn't established at all.

Meta's own reporting points somewhere useful on the annual comparison, and reconciling the two is the part worth having. In Q4 2025 Meta reported ad impressions delivered up 18% year on year and average price per ad up 6%. Across the full year, impressions up 12% and price per ad up 9%. Those are blended worldwide figures across every objective, placement and country, so they're emphatically not your CPM.

They do rule out one story, though. Q4 isn't an annual repricing of Meta's inventory: the year-on-year change for the quarter is single digits. It's a within-quarter ramp measured against October, which is why the October baseline is the number you should write down before the season starts and the number almost nobody records.

How big is the season, and how much of it gets discounted away?

Adobe Analytics, measuring over a trillion visits to US retail sites, put online spending between 1 November and 31 December 2025 at $257.8B, up 6.8% year on year. Cyber Week came in at $44.2B (up 7.7%), Cyber Monday at $14.25B (up 7.1%) and Black Friday at $11.8B (up 9.1%). Smartphones carried 56.4% of transactions.

Adobe measured discount depths of 30.9% off list in electronics, 29.6% in toys, 25.1% in apparel and 23.4% in computers. Read those next to the CPM ramp, because they're the same squeeze approached from the other end. If your category discounted a quarter of list price while your impressions cost close to double, contribution margin is being compressed from both directions in the same fortnight, and a plan that budgets for the CPM alone has priced half the season.

The practical version is to recompute break-even at your actual Q4 selling price rather than at list. A 25% discount doesn't reduce margin by 25%, it reduces it by 25% of revenue against a fixed cost of goods, which is usually a much larger proportional hit. Break-even ROAS has the arithmetic. Doing it in September is a planning exercise. Doing it in December is an autopsy.

Why did Shopify report 27% growth against a market growing 7.7%?

Shopify reported $14.6B across its own merchant base over Black Friday and Cyber Monday 2025, up 27% year on year, peaking at $5.1M a minute. That's a first-party figure about a company's own platform, so read it the way you would read any vendor reporting on itself.

The two figures do not even cover the same days. Adobe's 7.7% is five-day Cyber Week; Shopify's 27% is the four-day BFCM weekend, so some of the gap is just the window. Not all of it, though: even allowing for that, 27% against 7.7% is too wide to be measurement noise. Two things sit inside the rest. Shopify's merchant base grows year on year, so part of that 27% is new stores rather than existing stores selling more, and nothing in the published figure separates the two. And the independent brand mix skews towards categories and price points that the large marketplaces compete for less directly.

For an independent brand the implication is narrower than the headline. The useful market comparison is Adobe's, because it counts the whole online market you actually compete inside, including the marketplace share you lose to. Plan against your own year-on-year and treat the platform figure as evidence that independent commerce grew, not as a forecast you inherit.

When is the cheapest time to acquire someone you will sell to in November?

September and October, and the reason is arithmetic rather than clever. An impression bought in early October costs roughly what it costs in July. An impression bought on 3 December costs, on the reported figures, about twice that. An email address acquired in October is still yours in November at the marginal cost of a send. So a November sale made to an October subscriber gets paid for in October dollars.

The audience is there earlier than the calendar suggests. The National Retail Federation's holiday survey, published 16 October 2025, found 42% of shoppers planning to begin browsing and buying before November. They are not only cheaper to reach in October, a large share of them have already started.

Klaviyo's benchmarks across more than 183,000 customers show where the value lands once they are on the list. Flows are 5.3% of sends and around 41% of email revenue, and flows earn roughly 18x more per recipient than campaigns. A subscriber acquired in October walks into a welcome flow, a browse-abandon flow and an abandoned-cart flow, and none of those pay an auction price at all. The flows that carry the revenue goes through which ones and in what order.

Here's the part nobody has published, stated plainly because pretending otherwise would be worse. There's no operator study comparing cost per email acquired in September or October against cost per email acquired in November. Not a benchmark, not a survey, not a vendor report. We went looking for one and it doesn't appear to exist.

That absence matters more than it sounds, because the entire recommendation rests on it. "Build your list before Q4" is advice everyone gives and nobody has costed. The direction is obvious — a cheaper impression buys a cheaper subscriber, and the subscriber does not expire on 31 October — but the size of the gap is what decides how much budget to move and when to stop moving it, and neither of those is knowable from a principle.

So the honest version is that the direction is arithmetic and the magnitude is unmeasured. If you're going to do it anyway, measure it: cost per email acquired, by week, from the first week of September through the first week of December, holding creative and offer constant. Four columns and a date. It's a dataset the category doesn't have, and you would be producing it as a by-product of work you were doing regardless.

Is January really cheap?

The trough is real. Advertisers who spent an annual budget in Q4 stop bidding, and the auction falls faster than retail demand does for a few weeks.

The magnitude is contested and the figures circulating are not trustworthy. One series shows Meta CPM falling sharply from December into January, though we have not found a primary source for the size of the drop. The 40-60% reductions attached to the "Q5" framing are unverified — we can't trace them to a primary dataset and they shouldn't be planned against.

What's defensible: January impressions are cheaper than December impressions, buyers are colder and the audience you're talking to has just spent. Those pull against each other, which is why January tends to work for re-engagement, replenishment and subscription conversion, and to disappoint anyone who scales cold prospecting into it on the strength of the CPM alone.

What does the plan look like, working backwards from Black Friday?

Working backwards is the only ordering that makes the deadlines visible. Everything below is positioned relative to Black Friday rather than to a date, because the date moves and the sequence does not.

The four-weeks-out freeze is the row most often ignored and the most expensive to ignore. Restructuring an account resets the learning phase, and the days it takes to recover are the days you have spent all year buying inventory for. The launch arithmetic sets out what the learning phase costs in budget terms, and everything it says gets worse when impressions are at their annual peak. The corresponding row twelve weeks out is where the angle taxonomy earns its keep, because a September test tells you the same thing for half the money.

A Q4 checklist

What to do with all this if you only change one thing

Move the learning earlier. Almost everything above reduces to a single reallocation: spend September and October buying information and email addresses, then spend November buying sales. The auction makes that trade for you whether or not you plan it, and the only choice you control is whether you are on the cheap side of it.

And if you do run the September-to-December cost-per-subscriber measurement, publish it. The gap in the public record is real, the experiment is cheap and the category has been recommending this play for a decade without ever pricing it.