Research Q4 2026 US & Canada Beauty + supplements
Nobody can tell you what a CPM costs
Every Q4 plan I see opens with a version of "CPMs go up 20–30% over Black Friday." Nobody quoting that can tell you whose CPMs, over what window, against what baseline. So I rebuilt the cost picture from audited platform filings first, disclosed-methodology panels second, and nothing else. Three credible sources disagree on the sign, not just the size — and that disagreement turns out to be the most useful thing on the page.
Take every number on this page with a pinch of salt. These are panels and platform aggregates. A panel describes the advertisers inside it, not you. Meta's audited figure covers every advertiser on earth, including cheap international inventory a US DTC brand never touches. Tinuiti's panel is large US retail. Triple Whale's is 33,000 mostly-Shopify stores over four days. None of them is your account, and where they disagree by 17 percentage points in the same quarter, at most one of them is close to it.
Your own store data is the gold standard and the only primary source of truth about your business. Use everything here to check whether your direction of travel matches the market's — rising, falling, flat — and then set every actual number from your own trailing 90 days. Where I've marked a figure as not published, that isn't a hole for you to fill with someone else's guess; it's usually a metric sitting in your own account that you can measure this afternoon.
Amais Bajwa
Fractional growth lead
Here are six numbers I pulled while researching this. Read down the column and see if you can make them agree.
They don't tell one story, and that's the point. Meta's own audited price per ad rose 12% while the largest US managed-retail panel reported CPMs falling 7% in the Black Friday quarter, while DTC brands on Shopify paid 7.6% more over the weekend itself. All three are true. They measure different inventory, different advertisers and different windows. Anyone who reconciles them into a single number is guessing, and the single number they produce describes nobody.
What follows is the cost, margin and measurement layer underneath a Q4 plan: what advertising actually cost over BFCM 2025, what it's doing in 2026, and what margin survives a discount. Two of the calculators are live — put your own numbers in, because as I said up top, yours are the only ones that describe you.
Before the numbers
How to read a benchmark
A benchmark is a measurement of a population. If you don't know the population, you don't have a benchmark — you have a number.
Panel bias isn't a footnote, it's the finding
The three Meta cost series on this page disagree in sign for three consecutive quarters. Here's why, and it's entirely mechanical.
| Series | Population | Window | Q4 2025 reading |
|---|---|---|---|
| Meta average price per ad | Every advertiser on Meta, worldwide, all placements, all objectives. Revenue ÷ impressions delivered. | Full calendar quarter | +6% |
| Tinuiti panel CPM | US advertisers under Tinuiti management, >$4bn annual spend, skewed to large retail. Media CPM. | Full calendar quarter | −7% |
| Triple Whale panel CPM | 33,000 mostly-Shopify DTC shops running conversion campaigns. $607m spend, $2.88bn revenue. | Four-day BFCM weekend only | +7.6% |
Meta's figure includes cheap international inventory and cheap surfaces — Reels, Threads, Audience Network — that a US DTC conversion campaign barely touches. Tinuiti's panel is sophisticated enough to have migrated spend into those cheaper surfaces, which is exactly what produces a negative CPM print alongside +17% impression growth. Triple Whale's window is the four days when every DTC brand in the country bids simultaneously.
None of them is wrong. Where credible sources disagree on this page, both appear with the definitional difference named. They're never averaged into a consensus figure, because a fake consensus is worse than an honest disagreement — it hides the thing you actually needed to know.
CTR means at least three different things
- All-placement CTR counts every click on the ad unit — the link, the profile name, the like button, an expand. On Meta this typically runs two to four times higher than link CTR. It's the number most benchmark pages quote, because it's the flattering one.
- Link CTR (outbound) counts only clicks that leave for your site. This drives cost per landing-page view and is the only one worth planning against.
- Google CTR is impression-share dependent. A 12% Search CTR on brand terms and a 12% CTR on non-brand terms describe two unrelated businesses.
Where this page reports a CTR, the definition is stated. Where a source didn't state which definition it used, I didn't use the figure.
ROAS is an attribution window, not a performance measure
Meta's default reporting is 7-day click plus 1-day view. Switch the same campaign to 1-day click and reported ROAS routinely halves, with no change to the business. Google's Performance Max reports conversions across Search, Shopping, YouTube, Display and Gmail without telling you the split — which is why the finding that PMax delivers "ROAS similar to standard Shopping" should be read as reported ROAS similar, not incremental ROAS similar.
Haus's analysis of 640 incrementality experiments gives the practical translation: a campaign reporting 4.0× on Meta's dashboard delivered roughly 2.8× incremental — an inflation factor of about 1.43×. The break-even calculator below takes that factor as an explicit input rather than hiding it.
Your account will differ, and here's the size of the gap
Northbeam's 2025 review found median first-time CAC up roughly 9% and median MER down just over two percentage points — but with the spread running from sub-$5m brands whose revenue declined on higher spend to $50m+ brands deliberately buying share at 18–23% spend growth. A median that spans that range is a weather report, not a target.
Four tiers, and they're not interchangeable
Reported — the platform or company published its own audited actuals. Meta quarterly results, Shopify GMV, Adobe Analytics, Census. Highest confidence.
Panel — aggregated vendor data covering a subset of advertisers or stores. Usable, but the panel composition drives the answer. Every panel figure here states its size.
Survey — stated intent, not observed behaviour. Always described as intent.
Modeled — calculated on this page from cited inputs. The arithmetic is shown inline every time, so you can disagree with the assumption rather than the conclusion.
Five things worth changing your plan over
If you read one section, read this one. Everything after it is the evidence.
- Google, not Meta, is where 2025 costs broke. Triple Whale's 33,000-shop panel put BFCM Google CPA at $26.31, up 34.2%, with Google ROAS down 21.1% to 3.62×. Meta's CPM rose 7.6% and its ROAS actually improved 3.2%. Most 2026 budget conversations are pointed at the wrong channel.
- Discounting harder correlated with growing slower. Across Klaviyo's 10,000-brand stable panel, the average discount rate fell 10% YoY in BFCM 2025 and daily average depth never exceeded 30%. The brands running the smallest discounts grew fastest, at +14%.
- Peak week is more cost-efficient than November, not less. Cyber Week carried 17.1% of the entire Nov–Dec online season in five of 61 days — roughly 2.09× the daily rate — while modelled Meta CPM runs about 1.59× early November. Demand outruns cost by roughly 31%. The arithmetic is shown in full further down.
- Alphabet publishes no cost metric at all. There is no paid-click or cost-per-click disclosure in Alphabet's 10-Q or 10-K. Any page citing "Google's reported CPC trend" is citing something that does not exist. Panel data is the ceiling of what's knowable.
- There is no beauty or supplement CPM benchmark. Anywhere. Not from Tinuiti, Triple Whale, Northbeam, Varos or Gupta Media. Every vertical-cut "beauty Meta CPM" table I found came from a page with no disclosed sample — several of them quoting ROAS figures attributed to a public ad library, which structurally cannot contain advertiser cost or revenue data.
Meta absorbed the 2025 holiday surge by expanding supply, not by repricing
Meta's weakest price growth in seven quarters was Q4 2025 — the Black Friday quarter — at +6%, against +18% impression growth. A year earlier, Q4 2024 ran the exact inversion: +14% price on +6% impressions.
This does not mean your CPM was flat. It means the aggregate auction wasn't the thing that moved it. If your Q4 2025 costs blew out, the cause was almost certainly your own bidding, creative fatigue, or a narrow audience — not a market-wide squeeze. That's a fixable problem and a completely different remedy.
A · merchant economics
What BFCM 2025 actually cost to run
Ad cost is the part everyone watches. It's rarely the part that kills the quarter.
The discount, the return and the shipping line do more damage than CPM does — and unlike CPM, they're entirely within your control. So start there.
Discount depth: shallower than the reflex, and shallowest among the winners
Salesforce put the average US discount rate at a 28% peak on Black Friday 2025, 27% globally, and reported it as essentially flat against 2024. Underneath that average sits something more uncomfortable: order volume rose only 2% globally and 1% in the US, while average selling price rose 6%. Cyber Week 2025 grew on price, not on baskets.
The beauty discount statistic does not exist
Adobe publishes peak discount depth for eight categories every year. Cosmetics is in none of them, in 2024 or 2025. No holiday measurement provider publishes a supplements or VMS line at all. If you've seen a "beauty discounted X% over BFCM" figure, it wasn't measured — it was produced. It will circulate again this October.
The nearest genuine signal comes from Klaviyo's panel of 10,000 stable brands, which is Shopify-weighted and therefore much closer to a DTC beauty brand than Adobe's enterprise sample: the average discount rate fell 10% year on year, daily average depth never exceeded 30%, and the cohort of brands running the smallest discounts grew +14%. Klaviyo's Health & Beauty category grew 14% YoY over the five-day window, ahead of apparel at 11%.
Where the margin actually goes
The reason discounting bites harder in these two verticals than operators expect is that COGS doesn't move when the price does, and two other lines — fulfilment and returns — are close to fixed per order. On a $60 order, $12 of all-in fulfilment is 20% of revenue before anything else happens. Discount 40% and that same $12 becomes 33% of what you collected.
The 40% wall
A $60 skincare set at 75% gross margin, $12 fulfilment and the NRF all-online return rate of 19.3% produces −$0.67 of contribution at 40% off. Not a thin margin — a negative one, before advertising. Even at a far friendlier 8% return assumption it yields $4.25, a break-even MER of 8.47× against a Health & Beauty panel ROAS of 2.1×.
The sitewide 40%-off beauty promotion is not an aggressive growth play. It's a structurally unprofitable one that needs a second purchase to rescue it.
The first order isn't supposed to pay back
A $45 subscription starter at 30% off yields $1.43 contribution — a break-even MER of 22.1×, which is to say it cannot pay back on order one under any realistic acquisition cost. At full price the same order yields $14.16. The whole model rests on refill two.
Partner data puts subscriber 12-month value at roughly 2.3× a one-time buyer. That's the number that makes a discounted starter rational — but only if the refill actually lands at full price. Discount the recurring price rather than the first box and you've permanently repriced the LTV you were borrowing against.
The method, so you can disagree with the assumption rather than the conclusion. Skincare: list $60, gross margin 75% (your input — no panel publishes beauty COGS), fulfilment $12 all-in, Shopify Grow 2.7% + $0.30, return rate 19.3%, return handling $7.50, 0% resale recovery on opened cosmetics. Supplement: list $45, gross margin 65%, fulfilment $12, same fees, return rate 3% because consumables are rarely returnable. Contribution = R(1−r) − COGS − fulfilment − r·handling − fees, where R is list × (1−discount) and r is the return rate. COGS is set at list price and does not fall with the discount — that's the whole mechanism.
Run it on your own numbers
Defaults are the cited profiles above. Every one of them is a placeholder for a figure you already have. Gross margin in particular is an input, not a benchmark — nobody publishes it by vertical, with any sample.
Discount depth, contribution and break-even
Drag the discount and watch contribution per order, the break-even MER underneath it, and the volume lift you would need just to stand still. The attribution-inflation input is the one most calculators leave out: a campaign reporting 4.0× on Meta's dashboard delivered roughly 2.8× incremental across 640 Haus experiments. Set it to 1.00 if you measure incrementally.
At 20% off, the skincare set already needs a break-even MER higher than the best ROAS any platform delivered over BFCM 2025. That's the arithmetic behind Klaviyo's observation that the brands discounting least grew fastest. It wasn't a branding choice — it was the only cohort whose unit economics let them buy traffic.
The first-order-discount cohort problem
The strongest available evidence that peak-season cohorts retain worse is Northbeam's: across its 2025 dataset, revenue spiked around the promotional moments but first-time MER and first-time revenue deteriorated sharply in November and December, and median first-time CAC rose about 9% over the year. Triple Whale's panel shows the same shape from a different angle — new-customer revenue fell from 59.7% of BFCM revenue in 2024 to 48% in 2025. Brands spent more and bought a smaller share of genuinely new customers.
Deep-discount cohorts retaining worse is asserted everywhere and evidenced nowhere
The specific claim — that customers acquired on a deep first-order discount retain worse than customers acquired at shallower depth — is widely repeated and not published by any source with a disclosed sample. Two commonly-cited pages were checked and rejected: one attributes it to an unlinked third-party study with no sample, the other publishes beauty cohort benchmarks with no brand count, customer count, geography or period.
Treat it as a hypothesis worth testing in your own cohort data, not as an established benchmark. It's a split you can run this week.
Returns, shipping thresholds and stockouts
NRF and Happy Returns put total 2025 US retail returns at $849.9bn, with 19.3% of online sales returned and 9% of all returns fraudulent. 82% of consumers say free returns matter when shopping online.
No body publishes a beauty-ecommerce-only or supplement-only return rate. Every category table claiming one is assembled from cross-category research plus vendor case studies — one such page states this about itself explicitly, to its credit. The nearest defensible figure is NRF's 19.3% all-online rate, which almost certainly overstates supplements and may overstate or understate beauty depending on your shade-matched mix. That's exactly why the calculator exposes it as an input: it's the single most sensitive assumption in the whole margin model.
Free-shipping thresholds are the same story. Prevailing threshold values and their measured effect on AOV aren't published by any source with a disclosed sample — the searches returned only aggregator pages with no methodology. What can be said from the arithmetic above: at $12 all-in fulfilment on a $60 beauty order, a free-shipping threshold set below roughly $45 gives away the entire contribution on a single-unit order. Set it from your own contribution model.
On inventory: shoppers reporting they were "worried it would go out of stock" rose from 33% to 42% year on year. Scarcity is now the dominant purchase accelerator in that survey, which cuts both ways — a stockout during peak costs you a customer who was already primed to buy, and probably buys elsewhere. The dollar cost of a peak-season stockout isn't published anywhere; model it as contribution per order × lost units, plus the acquisition cost you already paid to generate the visit.
The app conversion premium is a vendor claim, not a benchmark
Every app-versus-web conversion and AOV comparison I found for this page sits behind a lead-capture form, published by a vendor selling app software, with no disclosed panel. None of it is cited here. What is citable: app sessions are growing at roughly 1.3× the rate of web visits, 86.5% of US 16–64s say they primarily shop on a smartphone or tablet, 54% say they prefer a brand's app (up from 50%), 84% expect their favourite brand to have one, and Gen Z trusts a brand app (58%) above Amazon (52%).
Those last three are stated intent. If you're building the app business case, build it on your own cohort data.
B · Meta
Meta: the disagreement is the finding
Start with the audited number, because everything else in this category is downstream of it.
Meta's quarterly results disclose the year-over-year change in average price per ad and in ad impressions delivered. Those are the only Meta cost figures anybody publishes that have been through an audit. The seven-quarter series is in the chart above, and its shape is the single most useful thing on this page: price growth and supply growth trade off against each other, and Q4 is when supply wins.
Three notes on reading that chart.
- Meta's figure is revenue ÷ impressions across everything. All geographies, all placements, all objectives. A US DTC brand buying conversions in the Feed is a small slice of it.
- Tinuiti's panel went negative because it moved. A −7% CPM alongside +17% impression growth in the same panel is the signature of spend migrating into cheaper surfaces — Reels, Shorts-style placements, Advantage+ inventory expansion. That's an achievable result, not an ambient market condition.
- Q2 2026 is the quarter to watch. Panel impressions collapsed to +2% and panel CPM turned +10%. The cheap-supply trade appears to be exhausting. If that persists into Q4 2026, the 2025 pattern won't repeat — and this is the strongest single argument for pacing conservatively into peak this year.
On Meta's own audited basis, 2026 is running hotter on both axes than 2025 did: price growth +12.0% versus +9.5%, impression growth +16.5% versus +8.0%. The panel view agrees on direction for the first time in Q2 2026. Going into Q4 2026, the reasonable planning assumption is that the 2025 escape valve — cheap incremental supply — is narrower than it was.
What actually happened over the weekend
Triple Whale tracked $2.88bn of revenue and $607m of ad spend across 33,000 shops over the four-day weekend — roughly 19.7% of the $14.6bn GMV Shopify reported for the same window, which makes it the largest disclosed DTC sample available.
| Metric | BFCM 2025 | YoY | What it means |
|---|---|---|---|
| Meta CPM | $22.26 | +7.6% | Cost rose, modestly |
| Meta ROAS | 2.26× | +3.2% | Return rose slightly faster than cost |
| Meta CPA | $34.06 | — | — |
| Meta share of BFCM ad spend | 67.6% | — | Concentration risk, unchanged |
| Google CPA | $26.31 | +34.2% | The blow-out |
| Google ROAS | 3.62× | −21.1% | Still the best ROAS, falling fastest |
| TikTok CPM | — | −27.5% | Cheapest impressions of the weekend |
| TikTok ROAS | 2.26× | +28.4% | Matched Meta on return, at lower cost |
| Blended AOV | $79.18 | — | Health & Beauty ran below this at ~$65 |
| New-customer share of revenue | 48% | from 59.7% | The quarter's worst number |
| Sources 16, 17 · Triple Whale BFCM 2025 · panel | |||
Meta wasn't the problem in 2025. Google was.
Meta CPM +7.6% with ROAS +3.2% is a channel holding its ground. Google CPA +34.2% with ROAS −21.1% is a channel repricing underneath you. Most 2026 planning conversations still open with "Meta CPMs are out of control." The panel data says otherwise, and the misdiagnosis sends budget to the wrong place.
Placement economics, and the two surfaces that stopped behaving alike
Within the same panel, Reels carried CPMs 30–40% below Feed placements, while Facebook Feed returned 2.51× ROAS on $154m of spend and Instagram Stories returned 2.51× on $202m of conversion value. Meta Reels contributed $184m of the panel's revenue. Reels' share of Instagram impressions in Tinuiti's panel reached 35% in Q2 2026, up from 21% a year earlier.
On Advantage+: Tinuiti's retail panel put Advantage+ Sales campaigns at 27% of retail Meta spend in Q4 2025, then 20% in Q1 2026 — down from a 38% peak. That's the only disclosed-sample Advantage+ share figure available, and it is a fact about budget allocation, not about performance. No source with a disclosed sample publishes Advantage+ versus manual performance isolated to the BFCM window. Anyone telling you Advantage+ "outperformed by X% over Black Friday" is quoting a case study, not a panel.
Beauty versus supplements on Meta
No disclosed-methodology panel publishes a beauty or supplement Meta CPM, CTR, CVR or ROAS. Not Tinuiti, not Triple Whale, not Northbeam, not Gupta Media, not Varos — I checked each directly. The nearest published cuts are a combined Health & Beauty ROAS of 2.1× and AOV of about $65 over BFCM 2025, and Klaviyo's Health & Beauty revenue growth of +14%. Everything more granular than that, anywhere on the open web, came from a page with no disclosed sample.
What can be said structurally, because it's mechanism rather than benchmark:
- Supplements carry restricted-category friction that beauty doesn't. Meta's Personal Attributes policy bars copy implying knowledge of a user's condition; dietary, health, weight-loss and weight-gain products are restricted to 18+; custom and lookalike audiences can't be built on health information, and audience names implying a condition are auto-disabled. Each of those narrows the addressable audience — and a narrower audience is the most reliable way to raise your own CPM regardless of what the market is doing.
- Beauty's creative format advantage is real and format-linked. Beauty converts on demonstration — application, texture, before-and-after within policy limits. That's native to Reels, which is where the cheap impressions are. Supplements have far less to show and lean on testimonial and claim, which is precisely the content most constrained by policy.
- Audience breadth runs the other way for prospecting. Supplements have a broader functional-need audience; beauty has a narrower demographic one but far higher purchase frequency. Neither effect has been quantified in any published panel.
The Q4 auction curve
This is the chart most Q4 plans are built on and almost none of them cite.
Here's the only disclosed-methodology weekly CPM source I could find, with the interpolation made explicit so you can see exactly where measurement stops and modelling starts.
The method, in full. Gupta publishes four 2024 Meta anchors: Cyber Monday CPM $17.70, stated as 138% above the annual average, which gives a 2024 annual average of 17.70 ÷ 2.38 = $7.44. From that: Black Friday $16.85 = 2.27×; Black Friday week average $13.42 = 1.80×; Cyber Monday week average $12.53 = 1.68×. The baseline assumption, stated explicitly: the first week of October sits at the annual average, because Gupta's separately-published Q4 premium over Q1–Q3 is only +13–17%, which a 1.80× peak week can only produce if October is near the annual mean. Every week without an anchor is monotonic interpolation.
The Google line has no published weekly anchor anywhere. Its shape is derived from the quarterly series only — Search CPC and Shopping CPC both ran −1% YoY through the Black Friday quarter, so the search auction demonstrably doesn't reprice the way the social auction does. Amplitude is set at roughly half of Meta's. Treat the Meta curve as evidence and the Google curve as a shape, not a measurement.
Peak days are far worse than peak weeks
Peak week costs about 1.80× early October and about 1.59× early November. But Black Friday ran 2.27× and Cyber Monday 2.38× the annual average — so a plan built on the weekly figure will underestimate the two days it cares most about by roughly a third. All four of those are anchors; the weeks between them are modelled.
The other end matters as much. The last week of December sits at roughly 0.72× the October baseline — the cheapest attention of the entire year, arriving immediately after the period you spent everything in. Budget for it.
C · Google
Google: volume, not price
And a correction to a premise that gets repeated in every deck I see.
Alphabet does not disclose paid clicks or cost-per-click, and hasn't for years
I verified this directly against the Q3 2025 Form 10-Q and the 2024 Form 10-K. Neither contains a Monetization Metrics section or any percentage change in paid clicks, CPC, impressions or cost-per-impression. Both filings mention only that the company will continue to "review, refine, and update our methodologies for monitoring, gathering, and counting the number of paid clicks and impressions" — the methodology sentence, with no metric attached. The disclosure was wound down beginning in 2018.
The consequence is structural and worth sitting with. Google has no platform-reported cost spine. For Meta you can check any panel against an audited number. For Google you cannot. Panel data is the ceiling of what's knowable about Google ad costs, and any page presenting "Google's reported CPC trend by quarter" is presenting something that does not exist.
With that established, the best available Google cost series is Tinuiti's, and its message is consistent across four quarters including the Black Friday quarter.
Brand versus non-brand
The brand CPC series is the interesting one. Brand keyword CPC growth eased to +2% by Q4 2025 after a +19% jump in Q1 2025, then fell to −9% in Q1 2026 and −1% in Q2 2026. Non-brand ran −1% in Q4 2025 and +4% by Q2 2026.
That is the opposite of the folk model, which says competitors pile onto your brand terms over BFCM and your defensive CPC explodes. In the largest US retail panel available, brand CPC decelerated through the Black Friday quarter while non-brand carried the inflation. If your own brand CPC spiked over BFCM 2025, that's a specific competitor bidding on you — a named problem with a named remedy — not a market condition.
Branded and non-branded CVR spread in beauty or supplements isn't published by anyone. Tinuiti publishes brand and non-brand CPC but not conversion rate, and publishes neither by vertical. Measure it in your own account — it's one of the few benchmark gaps you can close yourself in an afternoon.
PMax versus standard Shopping
Performance Max took 62% of combined Shopping and PMax spend in Q4 2025 and generated 61% of sales; 67% / 68% in Q1 2026; 60% / 60% in Q2 2026, where Tinuiti states PMax delivered ROAS similar to standard Shopping.
Spend share and sales share tracking within a point of each other for three straight quarters is the strongest available evidence that PMax is neither the miracle nor the disaster it's sold as. The necessary caveat: this is reported ROAS. PMax reports conversions across Search, Shopping, YouTube, Display and Gmail without disclosing the split, so a PMax campaign harvesting brand search inside an opaque box will report beautifully. The finding is that PMax reports like Shopping, not that it's incremental like Shopping.
Restricted-category friction on Search
Google's unapproved-substances policy prohibits a named list of supplements, products containing ephedra, hCG marketed for weight loss, herbal and dietary supplements containing active pharmaceutical or dangerous ingredients, and any non-government-approved product marketed as safe or effective for treating disease. DHEA is prohibited outside US-targeted campaigns; CBD requires certification. That isn't a disclaimer — it's a direct constraint on how many non-brand queries a VMS advertiser can legally bid into, and a narrower eligible query set is the most reliable way to raise your own effective CPC regardless of the auction.
What AI Mode and AI Overviews did to the picture
Two studies with real methodology, and they broadly agree.
- Agarwal & Sen, pre-registered randomised field experiment (Indian School of Business / Carnegie Mellon): 1,065 US desktop Chrome users, randomly assigned via a browser extension across two weeks each, January–February 2026. AI Overviews cut organic clicks 38% on queries where they appeared; outbound clicks fell from 0.61 to 0.38 per search; zero-click searches rose from 54% to 72%. AI Overviews appeared on 42% of queries. Reported satisfaction was "nearly identical" with and without — the summaries diverted traffic without improving the experience.
- SparkToro, using Similarweb desktop and mobile panel data, US Google searches January–April 2026: zero-click rate 68.01%, up from 60.45% in 2024.
Displaced organic demand is being re-acquired through the auction, at stable prices — for now
A randomised experiment showing a 38% organic click reduction is, from a media-planning standpoint, a demand shift from free traffic to paid traffic on exactly the informational and comparison queries that sit at the top of a beauty or supplement funnel — "best retinol for sensitive skin", "creatine vs protein". Tinuiti's panel shows paid Search clicks growing 13–14% a quarter at flat CPC, which is what you'd expect if that's what's happening.
Paid inventory and performance inside AI Mode specifically is not published. Google has released no impression, click or conversion data for ads in AI surfaces, and no panel has isolated it. Any figure claiming AI Mode ad performance is invented.
D · everything else
The other channels, sized honestly
One paragraph, one benchmark, one verdict each. Where a vertical cut exists it's given; where it doesn't, I say so rather than filling it in.
TikTok Ads and TikTok Shop
TikTok was the best-value impression of BFCM 2025 by a distance: CPM down 27.5% with ROAS up 28.4% to 2.26× — matching Meta's return at materially lower cost. On the Shop side, it's the one dataset in this entire research pass where beauty and supplements are separable.
Supplement buyers there average 2.8 orders a year and $98 of annual spend, against $236 on Amazon.
Verdict — beauty: fund it. Cheap impressions, native format, proven category depth. Verdict — supplements: fund it harder than beauty. VMS is the single largest Health & Beauty category on the platform, and the low annual spend per buyer is a retention problem you already know how to solve, not a demand problem.
Amazon Ads and retail media
Pacvue's Q1 2026 retail media benchmark gives the one genuinely beauty-specific paid figure I found anywhere: Amazon Beauty CPC $1.39, up 4.5% year on year, the fifth-most expensive of Amazon's top 15 categories, with brand average daily spend up 7.8%, ROAS up only 1.6% and conversion rate effectively flat. Across Tinuiti's panel, Amazon Sponsored Products spend grew +38% in Q2 2026 on +37% clicks and +1% CPC, while Amazon DSP spend grew +67%.
Verdict — beauty: fund it, but note that spend is rising 7.8% for a 1.6% ROAS improvement on flat conversion — you're buying share, not efficiency. Verdict — supplements: Amazon's $236 average annual spend per supplement buyer is more than double TikTok Shop's, so this is where the replenishment money already is. No supplement-specific Amazon CPC is published.
Affiliate and creator
Adobe's season data puts affiliate and influencer at 20.4% of holiday online revenue, up from 17.6%, growing 15.9% year on year — second only to paid search among tracked channels and more than four times social's 4.6% share.
Verdict — both verticals: fund it. It's the largest tracked non-search channel in the season and the only one where cost is variable and paid on performance, which is worth a great deal in a quarter where fixed CPM commitments are the main way plans go wrong. Commission-rate benchmarks for these verticals aren't published — searches returned only aggregator listicles with no sample. Model it as a variable contribution cost in the calculator above rather than looking for a rate to match.
Tinuiti's panel: spend +37% with CPM +11% in Q4 2025, then spend +27% with CPM −8% in Q1 2026. Note the policy constraint that matters here: Pinterest prohibits weight-loss imagery and language entirely, except for approved US and Canadian healthcare advertisers promoting prescription GLP-1 medication, and bars targeting on sensitive health conditions.
Verdict — beauty: fund a test. Gift-set and routine content is native to the surface and Q1 CPMs fell while spend rose, which is a healthy signal. Verdict — supplements: only if your product has no weight-management angle at all. The policy surface is narrow and enforcement is blunt.
The clearest "do not" on the page. Tinuiti's panel: Q1 2026 spend +77% with CPM +71%; Q2 2026 spend +33% with CPM +83%. Almost all the spend growth is price growth — advertisers paying dramatically more for barely more inventory.
Verdict — both verticals: don't fund it as a Q4 performance channel. An 83% CPM increase is the market repricing under you in real time, and Q4 is the worst possible quarter to learn a new auction. Reddit's value to these categories is organic visibility and AI-surface citation, which doesn't belong in a performance budget.
E · measurement
Blended measurement, as arithmetic
Every number above is a cost. None of them tells you whether you made money.
MER versus platform ROAS, and why the gap is structural
MER is total revenue ÷ total ad spend. Platform ROAS is platform-attributed revenue ÷ that platform's spend. They diverge for three reasons.
- Double-counting. Meta claims a conversion on a 7-day click / 1-day view window; Google claims the same conversion on last click; your email flow claims it too. Sum the platform ROAS figures and you'll routinely find attributed revenue exceeding actual revenue. That's not a measurement error you can configure away — it's three vendors each answering "did I touch this?" when you asked "did I cause this?"
- Baseline capture. Brand search, direct traffic and returning-customer purchases get claimed by whichever campaign touched them last. PMax takes 60–67% of Shopping-plus-PMax spend in a format that can harvest brand search inside an opaque box.
- Off-platform effect runs the other way. Haus reports that for omnichannel brands, 32% of a channel's impact landed on non-DTC sales — real revenue the platform never claims. If you sell in Ulta or on Amazon as well, platform ROAS understates as well as overstates, and the two don't cancel neatly.
The practical rule: set targets in MER, diagnose in platform ROAS. MER is the number that has to clear break-even; platform ROAS is the number that tells you which campaign moved.
New-customer CAC versus blended CAC
In a normal month the gap is a nuisance. In Q4 it's the whole story, because returning customers arrive in disproportionate numbers and flatter the blended figure exactly when you're trying to judge whether acquisition worked.
How blended CAC lies in Q4
Modelled on the published mix. Spend $100,000. Get 2,000 orders. Blended CAC = $50. Now split it: if 48% of revenue is new-customer revenue and AOV is flat, roughly 960 of those orders are new, so new-customer CAC = $104. A year earlier at 59.7%, the same spend and orders gave 1,194 new customers and a new-customer CAC of $84.
Blended CAC was identical in both years. Acquisition cost rose 24% and the blended number showed nothing.
Worth noting who got this wrong and by how much. Triple Whale's own pre-season forecast, published in August 2025, projected new-customer revenue share falling from about 57% to about 52%. It landed at 48%. The vendor with the largest DTC panel in the category under-forecast the deterioration by four percentage points — which is a useful calibration on how much weight any 2026 pre-season forecast deserves, including the modelled curves on this page.
Payback windows, and why 90 days breaks in Q4
A 90-day payback assumption embeds an assumption you're not making consciously: that the customer acquired today behaves like the average customer acquired on an average day. Four things break that in Q4.
- The discount is in the cohort. A customer acquired at 30% off has already consumed most of the first order's contribution. The payback clock starts from a lower base, so the same repeat behaviour produces a longer payback.
- Gift purchases have no second order. Beauty is heavily gifted. A meaningful share of your November new customers are buying for someone else and will never repeat — they're not slow to repay, they're never going to.
- December has a demand cliff behind it. A cohort acquired on 30 November hits its first natural replenishment window in January, the weakest retail month of the year.
- The window straddles your fiscal year. 90 days from Cyber Monday 2026 is 28 February 2027. If the board judges Q4 on Q4, a 90-day payback is invisible to the people deciding next year's budget.
The supplement case is the clean version. A $45 starter at 30% off contributes $1.43; the full-price refill contributes $14.16. At a new-customer CAC of $40, payback needs roughly 2.7 full-price refills — call it three billing cycles, so 90 days only works on a monthly subscription with near-zero early churn.
Incrementality testing, and what it costs in time
| Method | What it answers | Time cost | Practical constraint |
|---|---|---|---|
| Meta Conversion Lift | Did Meta cause these conversions? | 7 days min | Needs ≥10% of audience per cell and enough conversion volume; Meta rejects under-powered tests |
| Geo holdout | What happens if we switch a market off? | 18.6 days typical | 34 days for upper-funnel. Needs enough comparable geos |
| Full-channel blackout | What is our true baseline? | 2–4 weeks | The most honest and the most expensive. Never run it in Q4 |
| Inflation factor from a prior test | Roughly how wrong is our reported ROAS? | 0 days | Reuses a past reading. What the calculator above defaults to at 1.43× |
If you haven't started a test by early October, you're running Q4 on last year's factor
An 18-day geo test started in November returns its answer after Cyber Monday. A 34-day upper-funnel test started on 1 October returns on 4 November. Running Q4 on last year's inflation factor is fine — and far better than running it on platform ROAS at face value. What you must not do is start a test in peak week and act on it mid-flight.
F · planning
The 2026 budget-planning view
Everything above resolves into three decisions: how much to spend each week, what ROAS clears break-even at each depth, and what to do when the auction misbehaves.
Why peak is more efficient than November, with the arithmetic
Adobe: Cyber Week 2025 was $44.2bn of a $257.8bn November–December online season — 17.15%, over 5 of 61 days. Daily demand share at peak = 17.15 ÷ 5 = 3.43%/day. Season average daily share = 1 ÷ 61 = 1.64%/day. Demand multiple at peak = 3.43 ÷ 1.64 = 2.09×. Modelled Meta CPM, peak week ÷ early November = 1.80 ÷ 1.13 = 1.59×. Efficiency at peak versus early November = 2.09 ÷ 1.59 = 1.31×.
Demand outruns cost by roughly 31% in peak week. The instinct to pull back when CPMs climb is, on this arithmetic, backwards — provided contribution per order is still positive, which is the condition the decision tree below tests first.
How the curve was built, and how I checked it. It allocates 23% of Q4 media to October, 40% to November and 36% to December. Excluding October, that splits November / December 52% / 48%. Adobe's actual 2025 online season split 1 Nov–1 Dec against 2–31 Dec was 53.3% / 46.7% — so the curve lands within a point of observed demand without having been fitted to it. Peak week takes 16.2% of the quarter's media against Cyber Week's 17.1% of Adobe's Nov–Dec online total.
Q4 budget pacer
Spreads a total Q4 budget across 13 weeks using the pacing curve above, reprices each week using the modelled CPM inflation index, and returns impressions, clicks, orders, revenue and MER per week. CTR and CVR are not benchmarks — nobody publishes them, for any vertical, including blended. Rather than invent a figure, the defaults are back-solved: 1.5% link CTR × 3.0% blended CVR is the pair that makes this plan reproduce the one published outcome for these verticals, a 2.1× Health & Beauty BFCM ROAS. That's a calibration, not a measurement. Replace both with your trailing-90-day figures before you plan against this.
| Week | Spend | CPM | Orders | Revenue | MER |
|---|
Break-even ROAS by discount depth, in one line each
For a $60 beauty set at the cost structure modelled above — every figure here is modelled from the inputs cited in the economics section:
- Full price → break-even MER 3.32×. Achievable.
- 20% off → 5.52×. Already above every platform ROAS recorded in the BFCM 2025 panel.
- 30% off → 10.47×. Not achievable on paid acquisition alone.
- 40% off → contribution is negative before advertising. No ROAS target exists.
When CPMs spike beyond plan
Four questions, in order. Most people skip to the last one.
1. Is your CPM actually above plan, or is your frequency above plan?
If frequency is up, this isn't an auction problem. You're showing the same people the same ad repeatedly and paying a fatigue premium. Ship new creative or widen the audience — adding budget here accelerates the decay. If frequency is flat, it's genuine auction movement. Continue.
2. How far above the modelled index is it?
Under 1.3× plan: do nothing. The curve is modelled from four anchor points and a 30% miss is inside its honest error bar. Reacting here is how plans get destroyed. Between 1.3× and 1.8×: shift surface before you cut spend — Reels carried CPMs 30–40% below Feed in the BFCM 2025 panel, and moving placement mix recovers most of a 1.5× overshoot without losing volume. Above 1.8×: continue.
3. Is contribution still positive at the new CPM?
If yes, keep spending. A high CPM with positive contribution is an expensive good deal, and peak week runs roughly 1.31× more efficient than early November on demand-per-unit-cost. Do not cut a profitable channel because a ratio looks ugly. If it's marginally negative, cut the discount, not the budget — on a $60 set, moving from 30% off to 20% off lifts contribution from $4.01 to $8.69 and drops break-even MER from 10.47× to 5.52×. That's a far larger lever than any bid adjustment. If it's badly negative, continue.
4. Where does the money go instead?
First, owned channels. Klaviyo's panel attributed 42% of BFCM revenue to email and SMS, 43% on peak days, at near-zero marginal cost. It's the only place a displaced dollar reliably still works during peak. Second, the cheaper auction — TikTok CPM ran 27.5% below prior year over BFCM 2025 with ROAS up 28.4%. Third, hold it for the last week of December, where the modelled index bottoms near 0.72× the October baseline. Never deepen the discount to rescue ROAS. It moves the ratio and destroys the contribution underneath it — the single most common Q4 error, and the reason both calculators on this page exist.
Reference
What is actually published, by vertical
Checked against Tinuiti, Triple Whale, Northbeam, Gupta Media, Pacvue, Klaviyo, NielsenIQ and the platforms themselves, September 2026.
| Metric | Beauty | Supplements / VMS | All ecommerce |
|---|---|---|---|
| CPM | not published | not published | Published — Triple Whale, Tinuiti, Gupta |
| CPC | Amazon only, $1.39 | not published | Published — Tinuiti |
| Link CTR | not published | not published | not published by anyone |
| Conversion rate | not published | not published | not published by anyone |
| CPA | not published | not published | Published — Triple Whale |
| ROAS | Combined H&B only, 2.1× | Combined H&B only, 2.1× | Published — Triple Whale |
| AOV | Combined H&B only, ~$65 | Combined H&B only, ~$65 | Published, $79.18 |
| Return rate | not published | not published | Published, 19.3% — NRF |
| Discount depth | not published | not published | Published — Adobe, Salesforce |
| Fulfilment cost | Triangulated band only | Triangulated band only | Triangulated band only |
| Coverage audit conducted for this page, September 2026 | |||
Every "not published" above means the figure was searched for against the named publishers and does not exist in a form with a disclosed sample. It does not mean zero, and it does not mean unknowable. Most of these you can measure in your own account faster than you can find someone else's version.
Self-defence
Do not trust these numbers
This category has an unusually large ecosystem of pages publishing precise-looking ad benchmarks with no measurement behind them. They rank well, they get cited in decks, and then the figures circulate as fact.
Six tells, each one observed on a real page during the research for this article.
Tell 1 — The metric could not have come from the stated source
The clearest example I found: a page publishing beauty Meta benchmarks including CPA and ROAS by sub-vertical, sourced to "thousands of beauty-industry ads observed" in a public ad library. A public ad library contains creative and delivery metadata. It structurally cannot contain an advertiser's cost or revenue. CPA and ROAS are not observable from it at any sample size. The numbers are precise — skincare CPA $55–110, blended ROAS 1.8–3.2× — and they cannot have been measured the way the page says they were.
Your check: ask what instrument would produce this number. Cost and revenue metrics require access to an ad account or a store's order data. If the stated source is scraped public data, a survey, or "industry analysis", the cost figures are constructed.
Tell 2 — No sample size, no date range, no geography
Applying that rule eliminated the large majority of search results for every vertical query I ran. A legitimate panel says "33,000 shops, $607m spend, 27–30 November 2025" or "over $4bn in annual ad spend under management." An illegitimate one says "based on our analysis of industry data."
Tell 3 — Benchmarks that are secretly one agency's client list
Not fraud, but frequently mis-read. One page I do cite for CAC by vertical discloses, to its credit, that its figures come from a "client portfolio, ~40 DTC brands, $2m–$130m revenue" and are estimates rather than measured medians. That disclosure is exactly what makes it usable — and it also tells you that 40 brands is a different kind of object from 33,000. I quote it with that caveat attached, and it should never be quoted without it.
Tell 4 — A triangulation presented as a measurement
The fulfilment-cost figures in the calculator come from a page that states plainly: "the all-in per-order ranges are planning benchmarks, not measured medians; there is no public NRF or Shopify dataset that publishes median fulfilment cost per order by vertical." Honest, and therefore usable as a labelled planning input. The same numbers, republished elsewhere without that sentence, would be a fabricated benchmark. The figure didn't change. The disclosure did.
Tell 5 — The vertical cut that is too good to exist
If a table gives you CPM, CPC, CTR, CVR, CPA and ROAS broken out for skincare, haircare, fragrance and supplements separately, it's fabricated. No entity holds that data. Triple Whale, with 33,000 stores and $18.4bn of tracked annual ad spend, publishes a single combined "Health & Beauty" ROAS and AOV. Tinuiti, with over $4bn under management, publishes no vertical cuts at all. A page with more granularity than either has less data than both.
Tell 6 — A precise number for a thing nobody measures
Four of these will circulate this October and none of them exists: "beauty discounted X% over BFCM" (Adobe tracks eight categories and cosmetics is in none of them); "beauty ecommerce return rate is X%" (no body publishes one — even the aggregators assembling estimates say so); "Google's reported CPC rose X%" (Alphabet reports no CPC); "Advantage+ outperformed manual by X% over Black Friday" (no disclosed-sample source isolates Advantage+ to the BFCM window).
What I threw out, and why
Beauty cohort benchmarks — a press release publishing first-order payback of 84%, LTV:CAC of 1.4×, repurchase 35% and a 78% GWP lifetime-value lift, with no brand count, customer count, geography or period. Directionally interesting, unusable as a benchmark.
First-purchase-discount effect on LTV — the most-cited page attributes its central claim to a third-party study with no link, no sample, no period. The claim may well be true. It isn't evidenced.
App versus mobile web premiums — every source found was a vendor selling app software, gated behind a lead form, with no disclosed panel.
Free-shipping threshold benchmarks, affiliate commission rates — no source with a stated sample; aggregator listicles only.
Varos and Common Thread Collective — both named in my research brief. No public benchmark report with a stated sample could be located for the first; the second's 2025 results are podcast-only. Their absence is recorded rather than papered over.
Northbeam — partially admitted. Publishes period and source but not panel size. Used in the narrative for direction only, never for a headline figure or a calculator default.
Honesty
Where the data does not exist
Fourteen gaps hit during this build. Each is stated so that nothing fills it with something plausible later.
| Gap | Nearest published substitute | What to do |
|---|---|---|
| Beauty or supplement CPM, CPC, CTR, CVR or CPA on Meta | Combined Health & Beauty ROAS 2.1× and AOV ~$65 | Measure your own. Largest gap on the page and it is not closing. |
| Link CTR and conversion rate — any vertical, including blended | Nothing. No disclosed panel publishes either. | Both calculators take them as inputs, deliberately. |
| Beauty or personal-care discount depth | Adobe's eight categories, none of which is cosmetics; Salesforce 28% all-category | Use Klaviyo's Shopify-weighted direction (−10% YoY) as the closest signal. |
| Supplements in any holiday measurement | Shopify's "fitness & nutrition" top-category mention; NIQ TikTok Shop $784m | Treat VMS as unmeasured in season data. Do not infer from "health & beauty". |
| Beauty or supplement return rates | NRF 19.3% all-online | Exposed as a calculator input; the single most sensitive assumption in the margin model. |
| First-order-discount cohort retention | Northbeam's Nov–Dec first-time MER deterioration | Run the cohort split yourself. Widely asserted, nowhere evidenced. |
| Free-shipping threshold values and AOV effect | Unexpected costs = 46% of cart abandonment, intent data | Set the threshold from your contribution model, not a benchmark. |
| App vs mobile web conversion and AOV | App sessions growing ~1.3× web visits; intent data only | Vendor claims only. Build the case on your own cohorts. |
| Advantage+ vs manual, isolated to BFCM | Advantage+ share of retail Meta spend: 27% → 20% | That is an allocation fact, not a performance one. Do not read it as either. |
| Weekly Google CPC series | Quarterly only: Search and Shopping CPC both −1% through Q4 2025 | The Google line on the curve is labelled a shape, not a measurement. |
| Paid performance inside Google AI Mode / AI Overviews | Organic side only: 38% click reduction, RCT, n=1,065 | No impression, click or conversion data exists for ads in AI surfaces. |
| Gross margin by vertical | Nothing with a disclosed sample. | An input in the calculator, never a default presented as a benchmark. |
| Cost of a peak-season stockout | Stockout worry rose 33% → 42%, intent | Model as contribution per order × lost units, plus the CAC already spent on the visit. |
| Affiliate commission benchmarks | Affiliate = 20.4% of season revenue | Channel size is published; rate cards are not. |
The absence of an industry benchmark is not a reason to guess. It's a reason to instrument.
Link CTR, conversion rate, return rate by category, branded versus non-branded CVR, first-order-discount cohort retention — all of them are sitting in your own data right now. Which brings this back to where the page started: the numbers above tell you which way the market moved. Only your own store tells you where you are.
What I'd actually do with this
First, re-point the budget conversation at Google. If your 2026 planning meeting opens with Meta CPMs, you're solving last year's problem with the wrong channel. Google CPA rose 34% and ROAS fell 21% in the panel; Meta held its ground.
Second, set the discount from the contribution model, not the calendar. Run your own numbers through the calculator above before anyone agrees a sitewide percentage. If 40% off puts you underwater before advertising — and on a plausible beauty cost structure it does — that's a fact about arithmetic, not a matter of opinion, and it's much easier to win that argument in October than on 26 November.
Third, pace into peak rather than away from it. Demand outruns cost by roughly 31% in peak week. The condition is positive contribution per order, which is the first thing to check and the last thing most dashboards show.
Fourth, start the incrementality test now or accept last year's factor. Either is defensible. Running Q4 on platform-reported ROAS at face value is not.
Fifth, and this is the one that actually matters: take an afternoon and measure the five things nobody publishes. Your link CTR. Your conversion rate. Your return rate by category. Your branded versus non-branded CVR. Your first-order-discount cohort retention. Every one of those is in your own data, and every one of them is more accurate than anything on this page, because it describes you rather than 33,000 stores that aren't you.
If something here contradicts what you're seeing in your own account, trust your account. That's not modesty — it's the whole method.
I do this for a living, mostly for beauty brands
I'm a fractional growth lead. I build the data layer your growth team is missing, then run media, creative and email against it. If you want a second pair of eyes on a Q4 budget before it's locked — or someone to run the contribution model with you — that's a conversation I'm happy to have for free.
Say hello →Sources & method
43 sources. Each entry states what the sample actually is, because that's the part that decides whether a benchmark is usable. Every URL was opened during this build.
Audited quarterly results. Ad impressions +14% YoY, average price per ad +12% YoY, advertising revenue $59.363bn. Population = every advertiser on Meta, worldwide, all placements.
Audited. Impressions +19%, price per ad +12%, ad revenue $55.024bn.
Audited. The BFCM quarter: impressions +18%, price per ad +6% (full year +12% / +9%), ad revenue $58.137bn.
Audited. Impressions +14%, price per ad +10%, ad revenue $50.082bn.
Audited. Impressions +11%, price per ad +9%, ad revenue $46.563bn.
Audited. Impressions +5%, price per ad +10%, ad revenue $41.392bn.
Audited. Q4 2024: impressions +6%, price per ad +14% — the exact inversion of Q4 2025.
SEC filing. Checked directly for a Monetization Metrics section: there is none. No paid clicks, cost-per-click, impressions or cost-per-impression change is disclosed. Google Search & other revenue $56,567m vs $49,385m a year earlier.
SEC filing. Same check, same result: no quantified monetization metrics. The filing references only its methodology for "monitoring, gathering, and counting the number of paid clicks and impressions".
Trade press documenting when Alphabet began winding down paid-click and CPC disclosure. Cited for the timeline only; the absence itself was verified in the filings above.
Anonymised performance data from advertising programmes under Tinuiti management, over $4bn in annual ad spend under management, US, skewed to large retail advertisers.
Detailed extraction of the Q2 2026 figures used here: Meta CPM +10%, impressions +2%; Search CPC +1% on +13% clicks; Shopping CPC −1%; PMax 60% of spend and 60% of sales; Reddit CPM +83%. Same panel as 11.
Meta CPM −3% on +17% impressions; Search CPC flat; brand CPC −9%; PMax 67%/68%; Advantage+ 20% of retail Meta spend, down from a 38% peak; Pinterest CPM −8%; Reddit CPM +71%.
The BFCM quarter. Meta CPM −7% on +17% impressions; Facebook CPM −13%; Instagram CPM +8%; Search CPC −1% on +13% clicks; brand CPC +2%, non-brand −1%; PMax 62%/61%; Advantage+ 27% of retail Meta spend.
States the methodology explicitly: "anonymized performance data from advertising programs under Tinuiti management, with annual digital ad spend under management totaling over $4 billion." Meta CPM −2%, Search CPC −1%.
33,000 shops, $2.88bn revenue, 26m orders, $607m ad spend, four-day BFCM weekend 2025. AOV $79.18; new-customer revenue 48% of total, from 59.71% in 2024; Meta 67.6% of BFCM ad spend; mobile ~70% of transactions.
Same BFCM 2025 panel, platform detail: Meta CPM $22.26 (+7.6%), Meta ROAS 2.26× (+3.2%); Google CPA $26.31 (+34.2%), Google ROAS 3.62× (−21.1%); TikTok CPM −27.5%, TikTok ROAS 2.26× (+28.4%); Health & Beauty ROAS 2.1×, AOV ~$65; apparel 3.1× / ~$88. Represented 19.7% of Shopify BFCM GMV.
Same 33,000-shop panel, placement detail: Meta CPA $34.06; Facebook Feed 2.51× ROAS on $154m spend; Instagram Stories 2.51× on $202m conversion value; Reels CPMs 30–40% below Feed.
$18.4bn tracked ad spend, 33,000+ brands, calendar 2025. Cited here for panel scale and for the fact that a panel this large publishes no beauty or supplement cut.
Pre-season forecast from the same vendor: new-customer revenue share projected to fall from ~57% to ~52%. Cited only to contrast forecast with the 48% outcome.
Social Media CPM Tracker, 14-day trailing average across "tens of billions of ad impressions", monthly from Sept 2022 and weekly for seasonal peaks. The only disclosed-methodology weekly/daily Q4 CPM source located. Meta 2024 anchors: Cyber Monday $17.70 (stated 138% above annual average), Black Friday $16.85, week 48 $13.42, week 49 $12.53. Advertiser count not disclosed.
10,000 stable Klaviyo brands, five-day BFCM 2025, Shopify-weighted. $3.8bn attributed value (+27%); email + SMS = 42% of revenue (43% peak days); discount rate −10% YoY, daily depth never above 30%, smallest-discount brands grew +14%; Health & Beauty +14% YoY, apparel +11%. SEC-filed.
Merchant actuals: $14.6bn GMV (+27%), 81m+ buyers, AOV $114.70, peak $5.1m/minute, cross-border 16% of orders, top categories cosmetics / tops & pants / activewear / fitness & nutrition.
Published US rate card. Online card processing: Basic 2.9% + 30¢, Grow 2.7% + 30¢, Advanced 2.5% + 30¢, Plus 2.25% + 30¢; +1% on international cards. Used for the payment-fee default in the margin calculator.
Adobe Analytics, US online, 1 Nov–31 Dec 2025. $257.8bn (+6.8%). Peak discount depth for eight categories — electronics 30.9% down to furniture 18.8% — and no cosmetics, personal-care or supplements line in either 2024 or 2025. Affiliate/influencer 20.4% of revenue (+15.9%); social 4.6%; mobile 56.4%. Cyber Week $44.2bn.
Over 1.5bn shoppers across 89+ countries. Global Cyber Week $336.6bn (+7%); US $79.6bn (+5%). Average selling price +6% while order volume rose only 2% globally and 1% in the US — the season grew on price, not baskets. Mobile 70% of orders.
Trade press reporting Salesforce's Black Friday reading: average discount rate peaked at 28% in the US and 27% globally, flat against 2024; US online order volume −1% YoY; units per transaction −2%; ASP +7%.
Total 2025 US retail returns $849.9bn; 19.3% of online sales returned; 9% of all returns fraudulent; 82% of consumers say free returns matter. No beauty or supplement cut. Consumer and retailer sample sizes are not stated on the public page.
Pre-registered randomised field experiment (AEA RCT Registry). n = 1,065 US desktop Chrome users, randomly assigned via a browser extension, two weeks each, Jan–Feb 2026. Organic clicks −38%; outbound clicks 0.61 → 0.38 per search; zero-click 54% → 72%; AI Overviews on 42% of queries. The highest-quality causal evidence on this page.
Similarweb desktop and mobile web panel, US Google searches, January–April 2026. Zero-click 68.01%, from 60.45% in 2024. Excludes the Google mobile app, where the author notes zero-click is likely higher.
NIQ ecommerce sales data, 52 weeks ending February 2026, plus an NIQ Omnibus survey. Vitamins & supplements $784m (+11.0%); facial skin care $637m; fragrance $315m; H&B total $4.4bn+, reaching ~10% of US households. Supplement buyers: 2.8 orders/yr, $98/yr vs $236 on Amazon. NIQ panel size not stated in the article.
From Pacvue's Q1 2026 Retail Media Benchmark Report. The only beauty-specific paid cost figure located in this entire research pass: Amazon Beauty CPC $1.39, +4.5% YoY, fifth-dearest of Amazon's top 15 categories; brand average daily spend +7.8%; ROAS +1.6%; conversion rate flat. Advertiser count not disclosed.
Primary policy documentation. Prohibits a named list of unapproved pharmaceuticals and supplements, ephedra, hCG for weight loss, anabolic steroids, herbal and dietary supplements containing active pharmaceutical or dangerous ingredients, and non-approved products marketed as safe or effective for treating disease. DHEA prohibited outside US-targeted campaigns; CBD requires certification.
Practitioner compilation of current platform policy, cited for policy content only, not for any benchmark. Meta Personal Attributes policy; 18+ targeting for dietary/health/weight products; custom and lookalike audiences may not be built on health information; Pinterest bans weight-loss content except approved US/CA GLP-1 advertisers.
Analysis of 640 experiments. Meta Conversion Lift requires a 7-day minimum and ≥10% of audience per cell; geo tests average 18.6 days, upper-funnel 34 days. Worked example: 4.0× reported → 2.8× incremental (a 1.43× inflation factor). For omnichannel brands, 32% of a channel's impact landed on non-DTC sales.
Discloses its own limits, which is why it is usable: "client portfolio, ~40 DTC brands, $2m–$130m revenue, 2024–2026", described as estimates. Figures are fully-loaded — ad spend plus creative, tools, agency fees and allocated team cost — not media cost. Beauty & personal care $90–130; health & wellness $80–130; blended $68–84. Small panel: directional only.
States plainly that these are "planning benchmarks, not measured medians" and that no public dataset publishes median fulfilment cost per order by vertical. Triangulated from 3PL rate cards and carrier rates at 5,000–50,000 orders/month. Beauty and supplements both $11–13 all-in. Used as a labelled calculator default.
Cited for its disclosure, not its numbers: "Beauty's return rate is a blended estimate, not a single reported figure. No major body publishes a standalone beauty ecommerce only return rate." That statement is the evidence for the gap recorded here.
Northbeam customer dataset, calendar 2025. Median first-time CAC up ~9%; median MER down just over 2 percentage points; first-time MER and first-time revenue deteriorated sharply in November and December. Panel size not disclosed — used for direction only, never for a headline figure or a calculator default.
Covers Beauty & Personal Care and Health & Wellness among eight verticals, but publishes only year-over-year percentage changes — no absolute values, and no panel size. Cited as evidence that even the MTA vendors do not publish absolute vertical benchmarks.
Digital intelligence panel plus proprietary consumer research (July 2026). Paid + Display now 33.8% of all visits, overtaking direct traffic; app sessions growing ~1.3× the rate of web visits; 86.5% of US 16–64s primarily shop on a smartphone or tablet; gen-AI referrals under 1% of visits but growing 200–280% YoY.
Survey of n = 1,205 mobile shoppers; geography not stated. Intent, not outcome, and mobile-skewed by construction. Used for cart abandonment causes (unexpected costs 46%), stockout worry 33% → 42%, app preference and trust, and the partner-cited figure that subscribers generate ~2.3× the 12-month value of one-time buyers.
Shared evidence base for this series, traced to named primary publishers during the Consumer & Market build. Used here for the 2025 season actuals referenced in the pacing validation.
Three sources I went looking for and couldn't use
Varos — no public benchmark report with a stated sample could be located in September 2026. Common Thread Collective — 2025 BFCM results discussed on podcast only; no published dataset with a stated sample. WordStream / LocaliQ — no current benchmark release with a stated sample for these verticals. Their absence is recorded rather than papered over.
Everything above is context. Your store is the source of truth.
Cost benchmarks are volatile and panel-dependent. Re-verify anything cost-related before you commit budget, and treat every figure here as a direction of travel rather than a target to hit. The numbers that should actually drive your Q4 are the ones in your own account.