Most Q4 plans I see are built on the wrong numbers

The short version

Ten findings, each one a number, a source and a decision. If you only read one section, read this one — everything after it is the evidence.

1. Sentiment is at a modern low and your category is still growing. U. Michigan sentiment fell to 47.8 in September, 43% below its long-run average of 83.7, while US beauty grew 7% in both prestige and mass in H1 and supplements grew 5.5–7.1% in 2025. Don't let a macro-driven board impose a defensive discount your category data doesn't justify. Bring both numbers to the meeting.
2. AI traffic went from worst-converting to best-converting in twelve months. Adobe measured AI-referred traffic converting 22% worse than non-AI in May 2025 and 54% better by May 2026, on volume up 138%. Fund AI findability — structured data, complete attributes, machine-readable facts. Not agentic checkout.
3. Agentic checkout went backwards while AI referral went forwards. ChatGPT Instant Checkout was scaled back during 2026 on low conversion and accuracy, and Perplexity's shopping agent is under injunction. This is the thing I got wrong last year. Delete it from the Q4 roadmap; re-verify in January.
4. The weekend is 17% of the season. December is 47%. Cyber Week was $44.2B of a $257.8B US online season; 2–31 December was $120.4B. Hold gift-set inventory and budget past Cyber Monday instead of liquidating into the larger half of the season.
5. Four in ten holiday baskets aren't gifts. Self-purchase rose from 31% to 40% of holiday buying in three years, and it's 53% for Gen Z against 27% for Boomers. Build a "for you" route with full sizes and restock framing alongside the gift guide.
6. There is no beauty discount benchmark, and no supplement holiday benchmark at all. Adobe publishes peak discount depth for eight categories; cosmetics isn't one of them, and none of its 18 tracked categories is VMS. Benchmark against your own contribution margin and the 20–25% shoppers say they need.
7. Growth moved from colour cosmetics to skin, hair and fragrance. Makeup dollars rose while units fell in both prestige and mass. Prestige skincare grew 9% on double-digit units, sun care 19%, hair serums nearly 60%, mass fragrance 15%. Merchandise peak around the lines growing on units.
8. The biggest cart leak is a cost the shopper didn't see coming. 46% abandon over unexpected shipping or tax — ahead of changing their mind (35%) and price being too high (32%). All-in pricing before the final step costs a decision, not a budget.
9. Canada is a different season, not a different postcode. About 8.6 million Canadians shop Boxing Week, 83% expect to shop in stores against a 77% average, and only 11–12% turn to AI first. De minimis is gone, with $80–$200 per-item postal duties. Plan a second Canadian peak and quote duty-inclusive landed cost.
10. Sustainability is a tiebreaker; certification is a driver. 85% say they prefer sustainable brands while ~25% of CPG products sold are sustainability-marketed — a 60-point say-do gap. Keep green claims on the PDP, not in the peak ad. For supplements, lead with third-party testing.

Six numbers, clearly labelled

Three of these are measured outcomes. Three are stated intent. They are not the same kind of thing and I'm not going to let them sit in the same sentence without a label.

A · 2025 actuals

What actually happened at BFCM 2025

Cyber Week grew faster than the season around it, and almost all of that growth was price, not units.

Salesforce measured average selling price up 6% globally against order volume up just 2%. That's a weekend that looks like a record in dollars and like flat demand in baskets — and if you're selling a consumable, the difference between those two readings is the difference between a good quarter and a discount you didn't need to give.

The same weekend, measured four ways

These four numbers get quoted interchangeably in every deck I see, and they are not the same thing. Different geographies, different date ranges, different definitions of "retail". Use the one that matches your question and never average them.

Baskets, devices and payment

Shopify's global average cart was $114.70, or $112.29 on a constant-currency basis. The gap between those two is pure FX, and it's the cleanest illustration I've seen of why a Canadian merchant reading US benchmarks needs to check the currency basis before anything else. Peak throughput hit $5.1M per minute at 12:01pm EST on Black Friday for Shopify, and $16M per minute across Adobe's US panel between 8 and 10pm on Cyber Monday.

Mobile took 57.5% of Cyber Monday dollars in Adobe's US panel and 56.4% across the whole season, up from 54.5% in 2024. Salesforce, measuring orders rather than dollars and globally rather than US-only, put mobile at 70%. Both are correct — orders skew more mobile than dollars because mobile baskets are smaller. BNPL took $1.03B on Cyber Monday alone and $20B across the season, up 9.8%, with 82.2% of those transactions placed on a smartphone.

Where beauty and supplements actually landed

Adobe's season category table puts cosmetics at $8.4B, up 9.3% year on year — growing faster than electronics (+8.2%), apparel (+7.4%) and the 6.8% all-category average, on a base roughly one-seventh of electronics'. That's the single most useful beauty number from the season: the category is small in absolute online holiday terms and it's compounding faster than the things sitting next to it on the shelf.

Supplements are the harder case, and the honest answer is uncomfortable. Adobe's 18 tracked categories don't include a vitamins, supplements or nutrition line at all — so there is no Adobe read on VMS holiday performance. Shopify, however, names "fitness & nutrition" among its top-selling BFCM 2025 categories, alongside cosmetics. Read those two together: supplements register strongly on a DTC-heavy platform and disappear inside a panel weighted to the top 100 US retailers. If your board benchmarks you against "the market" using Adobe, the market it describes is not the one you sell in.

Channels, cross-border and returns

Two channel shares moved enough to matter. Affiliate and influencer traffic drove 20.4% of US online holiday revenue, up from 17.6%, and social media drove 4.6%, up from 3.3% — a 40.3% jump off a small base. Adobe named paid search and email as reliable drivers but didn't publish their revenue shares in the 2025 release, so a complete channel mix doesn't exist. That's a genuine gap, not an omission on my part. Salesforce, measuring traffic rather than revenue, put social at 15% of global and 16% of US Cyber Week site traffic, with TikTok alone at 9% of social traffic and up 55%. The two aren't in conflict: social sends a lot of traffic and converts it into a much smaller share of revenue.

16% of all Shopify global orders were cross-border during BFCM 2025, with the US, UK, Australia, Germany and Canada the top five merchant markets. Shopify didn't publish US↔Canada corridor volumes this year, so that question has no primary answer for 2025.

Returns ran at 15.8% of US retail sales across 2025 — $849.9B — rising to 19.3% for online sales and 17% for holiday purchases, with 9% of returns judged fraudulent. Adobe measured holiday returns down 1.2% year on year, with 38.8% initiated on mobile. No published source gives a beauty-specific or supplement-specific return rate for the season, which means your own number is the only one that exists.

B · 2025 actuals

The season is much bigger than the weekend

Cyber Week is 17% of the season. Planning against it alone means planning against a sixth of the opportunity.

US online spending across November and December reached $257.8B, up 6.8% — beating Adobe's own pre-season forecast of $253.4B and +5.3%. Worth sitting with for a second: the best-resourced forecaster in the category ran a point and a half light on its own market. If they can't call it, neither of us should be planning to two decimal places.

Two things follow from that shape. First, December after Cyber Monday is worth almost as much as everything before it — 46.7% against 53.3%. Second, October isn't a rounding error: $88.7B, up 8.2%, growing faster than the season it precedes, helped by an October Prime Day that took $9.1B in two days at an average 18% off. Demand genuinely moved earlier. Adobe recorded 25 days above $4B in 2025 against 18 in 2024, which is the cleanest single measure of the season flattening out.

Klaviyo's platform read corroborates the concentration from another angle: it projected nearly half of holiday revenue landing in just 16 days, with discounts roughly doubling during BFCM versus the rest of the year and close to half of Americans starting before Thanksgiving. Those were forward projections published before the season, and Klaviyo doesn't disclose its panel size — so treat them as a directional cross-check on Adobe's arithmetic, not as a second measurement.

The gifting picture, and the 40% that isn't gifting

NRF's October survey of 8,247 US adults put planned spend at $890.49 per person, down 1.3% from the record $901.99 — split $627.93 on gifts and $262.56 on seasonal items. Deloitte, surveying 4,270 US consumers a month earlier with a wider basket that includes experiences, got $1,595, down 10%. Both are stated intent, both collected before the season, and the realised season came in up 4.1%. That's the annual lesson: survey intent systematically understates what people actually spend.

This is the chart I'd put in front of a merchandising team. Self-buying rose from 31% to 40% of holiday purchases in three years and reaches 53% among Gen Z — nearly double Boomers. A plan that is all gift sets and "for her" navigation is mis-serving four baskets in ten.

C · 2026 actuals

Your category in 2026

Ecommerce is taking share faster than at any point since the pandemic, and both verticals are growing ahead of retail as a whole.

US online sales grew 12.2% in Q2 2026 against 6.7% for total retail, lifting ecommerce to 17.1% of US retail. This is the most encouraging section on the page and it sits directly next to the most discouraging one, which is roughly how Q4 2026 feels from where I'm sitting.

One caution before the Canadian number: StatCan measures retail ecommerce by Canadian retailers only, so anything a Canadian buys from a US or overseas seller is excluded. That's why Canada looks like 7.7% against the US 17.1%. Those two figures should never appear in the same sentence as a comparison, and you should be ready to correct anyone who tries.

Beauty in the first half of 2026

Circana's H1 read is the most important market number for a beauty brand this year: prestige +7% to $17.1B and mass +7% to $39.2B. Prestige and mass growing at the same rate is unusual, and it kills the simple trade-down story. What's actually happening sits one level down.

Read the unit column, not the dollar column. Makeup dollars are up in both tiers while makeup units fall in both tiers — that's price doing the work, and it's the one part of beauty behaving like a category in trouble. Everything adjacent to skin and hair health is growing on units: prestige skincare posted double-digit unit growth, body care +16%, sun care +19%, hair serums nearly +60%.

Fragrance has quietly become structural. McKinsey now puts it as the third-largest global beauty category, ahead of colour cosmetics, and mass fragrance at +15% is the single fastest line in the US table. Two more numbers frame the channel fight: ecommerce is now over 25% of US beauty sales, and Amazon, Ulta, Sephora and TikTok together take just under half the US market, with TikTok beauty growing around 260% a year since 2023 toward a projected $4B in 2026. K-beauty holds 3% of prestige skincare and 6% of mass skincare in the US — a smaller share than the discourse implies, but growing against a backdrop where Korea exported $11.4B of cosmetics in 2025, $2.2B of it to the US, its largest destination for the first time.

Supplements, where two vintages of the same number disagree

NBJ's March 2026 presentation put the 2025 US supplement market at $72.88B, +5.5%. Its June 2026 Supplement Business Report put it at $74.15B, +7.1%. Same publisher, same year, restated three months later. That's a normal revision as channel data completes, and a good reason to treat any single supplement market number as provisional. I've kept both.

Channel mix is the number to plan against: ecommerce is 24.3% of US supplement sales, behind mass (27.1%) and natural/specialty (25.4%). TikTok Shop alone is 3% of supplement sales, growing 71% — and 13% of everything sold on TikTok Shop is vitamins and supplements. No other category comes close to that concentration, and it's the clearest signal of where incremental supplement demand is being created.

Four structural forces, and what each one does to you

Tariffs and the end of de minimis

The $800 exemption is gone. FDA screening of all regulated imports regardless of value began 9 July 2025; tariff application to de minimis parcels began 29 August 2025, with postal parcels facing specific duties of $80–$200 per item depending on origin. For scale: de minimis shipments went from 134 million a year in 2015 to over 1.36 billion in 2024.

For beauty this is direct and severe if you're landing small parcels from Korea or China, or shipping DTC into Canada. For supplements the FDA screening change is the bigger one — every regulated import now needs correct data elements at entry, so a mis-declared ingredient is a refused entry rather than a wave-through. Interesting wrinkle: tariffs didn't show up as the thing shoppers blame. Among the minority planning to cut back, 40% cite prices and inflation and only about 9% cite tariffs. The cost is real; the consumer narrative about it isn't.

GLP-1 spillover

23% of US households now report GLP-1 use, skewing Gen X and incomes above $100K, with new patient prescriptions up 2.9 million (+16%) in the year to September 2025. McKinsey estimates GLP-1 households spend around 30% more on beauty products than non-GLP-1 households, driven by facial volume loss and hair concerns.

This is the rare force that's a tailwind for both verticals. For beauty it points at volumising skincare, body care and hair. For supplements it points at protein and muscle preservation — sports nutrition is the fastest-growing NBJ category at +8% and creatine posted +72% in MULO scan data. One caution: I found no published source quantifying GLP-1 households' supplement spend specifically. The beauty figure is published; the supplement one is inference from category growth, and I'm not dressing it up as more.

K-beauty and dupe culture

Korea is now the world's second-largest cosmetics exporter, having passed the United States, with a record $10.1B trade surplus. Yet measured share inside the US is still modest. The pricing pressure is real but it's showing up as masstige rather than as prestige collapse: Circana finds masstige outgrowing prestige by 6 points and mass by 9. Combined with makeup units falling while dollars rise, the honest reading is that prestige pricing power is intact in skincare and hair, and eroding in colour cosmetics.

Channel shift and retail media

Beauty ecommerce is over 25% of US category sales; supplements ecommerce is 24.3%. In both, the DTC site is now competing against four concentrated destinations rather than a long tail, and department stores have declined 2% a year since 2019. On retail media, the only figure I can cite from a primary forecaster without a subscription is that Amazon's retail-media revenues will exceed $75B by 2028, more than $65B ahead of the next-largest network. The detailed 2026 US retail-media spend forecast sits behind EMARKETER's paywall, so I'm not quoting a 2026 dollar figure I can't verify.

Demand measured in traffic rather than till receipts

Circana counts dollars at retail. Similarweb counts behaviour on the open web. The two corroborate each other in a way that should settle any internal argument about whether beauty is really holding up: US beauty site visits grew 32.5% year on year, from 3.23bn to 4.27bn, the second-fastest growth of any top-25 market after India. Canada grew 8.1% on visits and 8.6% on unique visitors — slower, but notably one of the few Canadian ecommerce subsectors where visitor numbers rose at all.

The product-view data underneath is emphatic on the same point. Fragrance product views grew 43.5% in the US, six times the 7.1% category average, while makeup managed 2.2% and shave and hair removal fell 4.2%. Two independent measurement methods — retail dollar sales and web product views — now say the same thing: the growth in beauty is in fragrance, treatment and self-care occasions, and colour cosmetics is the laggard.

Amazon is where the supplements signal finally appears

Adobe doesn't track VMS and no holiday survey measures it. Similarweb's Amazon unit-sales data gets closer than anything else here: Health & Household is Amazon US's largest category by units at 3.0bn, 14.6% of the platform, growing 12.2%. That's below Amazon's 14.4% platform growth, but it's a real, sizeable, growing number for the category closest to supplements — and it dwarfs Beauty & Personal Care's 2.0bn units.

Caveat before you use it: "Health & Household" on Amazon includes household cleaning, paper goods and OTC medicines alongside vitamins and supplements. It's a proxy, and I'm labelling it as one rather than quietly presenting it as your category.

Outside the US the beauty picture on Amazon is dramatically stronger. Amazon UK beauty units grew 66.2% — 2.37× a platform already growing 27.9% — and Amazon Germany's Kosmetik grew 38.9%, 3.21× its platform. In the US, beauty merely kept pace while fashion pulled ahead.

D · 2026

The state of ecommerce, honestly sized

Traffic is growing modestly, apps are growing faster than the web, and AI referral is growing explosively from a base so small it's still a rounding error in most stores' analytics.

Getting the order of magnitude right here is what separates a useful 2026 plan from a fashionable one. So, plainly: worldwide ecommerce web visits grew 6.8% while unique visitors grew 5.8% — visits outpacing visitors, which means the same people are browsing more. App sessions grew 8.6%, roughly 1.3× the rate of web. In the US both rose together. In the UK, Germany, France and Canada, unique visitors actually fell while visits held up.

Channel mix, and the number that is small and the number that isn't

This is the most operationally useful table in the whole Similarweb report, and for beauty the two halves tell opposite stories.

Generative AI is 0.3% of beauty traffic and growing 271.4%. Both halves of that matter. Anyone telling you AI referral is already a meaningful acquisition channel for beauty is wrong by two orders of magnitude. Anyone telling you to ignore it is missing the fastest-growing channel in the mix by a factor of roughly five over the next-fastest.

On the revenue side, Statista's model puts global B2C ecommerce revenue at $4.9 trillion by 2030, up over 27% on 2026, with the US moving from $1,215.9bn in 2026 to $1,507.0bn by 2030. It further estimates that AI's implementation in the shopping journey contributed about 1.94% of worldwide ecommerce revenue growth between 2024 and 2025 — the only attempt at an attributed revenue figure for AI I found anywhere. Canada isn't in the Statista sample, so there's no equivalent Canadian forecast.

E · 2026 actuals

Sentiment is at a modern low

This is the number that should change your plan — and the number most likely to make your board change it in the wrong direction.

The University of Michigan Index of Consumer Sentiment fell to 47.8 in its September 2026 preliminary reading — down from 51.7 in August and 55.1 a year earlier, and 43% below the 1978-present average of 83.7. Consumers are entering this Q4 feeling worse than they have in the modern history of the series, while spending more than they say they will.

The sub-indices say where the pain is. Current conditions read 50.9 and expectations 45.8; year-ahead inflation expectations jumped from 4.0% to 4.6%, the highest since June, with long-run expectations at 3.4%. The Conference Board's index, which weights labour-market perceptions more heavily, is far less alarming at 89.4 in August, with present situation at 121.2 but expectations at 68.2.

Both indices agree that consumers feel fine about now and bad about next year. That is the sentiment signature of a season where people buy — but buy defensively, and late.

Trade-down is real, and it isn't what you think

McKinsey's 2026 consumer survey finds 82% of consumers using items longer, 69% repairing rather than replacing, 30% buying apparel secondhand and nearly half doing services themselves that they used to pay for. Deloitte's pre-season wave found 89% searching for deals, 77% trading down to lower-tier retailers and brands, 77% expecting higher prices and 57% expecting the economy to weaken — the most negative reading since Deloitte began tracking in 1997.

But when McKinsey asked what consumers would splurge on with a spare $200, vacations led and health and wellness was cited by only about 8%. Beauty doesn't appear as a named splurge category in the published cut at all. That's itself a finding, and a caution against the comfortable story that beauty is where discretionary money runs to.

Is the lipstick effect breaking?

The case that it holds. US beauty is a $154B market growing 9% on a rolling twelve months, with prestige and mass both up 7% in H1 2026 — growth delivered while sentiment sat near record lows. Prestige skincare grew 9% on double-digit unit growth, which is real demand, not price. Cosmetics outgrew the all-category online holiday rate in 2025. On the evidence, beauty is still absorbing money being withheld from bigger purchases.

The case that it's breaking. The effect is supposed to be about small indulgent purchases — classically colour cosmetics. Colour cosmetics is the one part of beauty where units are falling in both tiers while dollars rise on price. Prestige makeup grew just 3%, the weakest line in the table. Meanwhile masstige is outgrowing prestige by 6 points and 77% of consumers say they're trading down.

My reading. The lipstick effect hasn't broken; it's moved. The indulgence money that used to go to a lipstick now goes to a serum, a fragrance or a body product — categories growing on units at +9%, +15% and +16%. If your Q4 plan assumes colour cosmetics is your resilient impulse line, you're running the 2019 version of this category.

Is VMS discretionary or a fixed cost?

The honest answer is that no published source settles it. The strongest indirect evidence points to fixed-cost behaviour: the US supplement market grew 5.5–7.1% in 2025 through the same sentiment collapse, and NRF card data shows health and personal care growing 2.5% in December — slow, but positive and remarkably stable. Against that, McKinsey finds health and wellness named as a top splurge by only around 8%, which suggests it isn't where discretionary money goes either. A category that neither surges nor collapses is behaving like a fixed cost — but that's my inference from three indirect readings, labelled as such, not a published finding.

F · 2026 intent

What shoppers say they'll do

Demand intent is the most stable number in this whole report — and it flatly contradicts how people say they feel.

99% of surveyed shoppers say they're likely to shop BFCM 2026, and roughly seven in ten have been "highly likely" for four straight years. 42% expect to spend more than last year; only 13% expect to cut back. Set that against September sentiment of 47.8 and the gap between how people feel and what they intend to spend is the single most important thing to plan around this quarter.

The bottom tier has lost eight points since 2024 and the top two tiers have gained. That's the practical case against a blanket sitewide discount: the money is moving up the curve, and a flat percentage off spends your margin hardest on the shoppers who were already going to convert.

Only 17% of shoppers begin during BFCM week itself. The weekend is where decisions are executed, not where they're formed. A partner dataset inside the same report puts it more sharply: 71% plan to start before Black Friday, up 20% year on year, while 84% expect deals to arrive in November or later. Demand arrives weeks before shoppers expect a discount, and that gap is the only period in the season when you can acquire without discounting.

G · 2026

AI: the finding that changed my mind

Not the adoption rate. The conversion inversion — and what it means for where the Q4 budget goes.

In May 2025, Adobe measured AI-referred traffic converting 22% worse than every other source. By May 2026 it converted 54% better. Inside twelve months, the worst-converting channel in retail became the best-converting one.

The engagement metrics behind that inversion matter as much as the headline. Shoppers arriving from an AI referral in May 2026 spent 53% more time on retail sites, browsed 23% more pages per visit, and were 15% more engaged overall. Volume grew 138% year on year and 1,324% since October 2024, after a season in which AI traffic to retail rose 693.4%.

How far AI has actually penetrated the journey

Adobe measures referrals — the last click before arrival. Similarweb measures the whole session, which is a different and arguably more honest question: did AI appear anywhere on the path to a transactional retail site? For June 2026 the answer is 11.4% of US desktop shopper journeys, up from 4.5% in July 2024.

Last-click attribution understates AI's real involvement in the purchase decision by roughly thirty-fold. Generative AI delivers 0.3–0.4% of visits as a last-click referral; it appears somewhere in 11.4% of journeys. Those two facts together are the most important thing on this page about AI.

And notice the shape of that line. It isn't a clean climb. It peaked at 10.9% in October 2025, fell to 10.4% in November and 9.9% in December, then recovered. That's a real, measurable Q4 effect: AI is a research-phase tool, and the research phase compresses during peak. Plan your AI-visibility work to pay off in September and October, when shoppers are still deciding. By Cyber Week they've stopped asking and started buying.

Two findings that kill the "AI versus search" framing

First: among journeys that use AI, 89% also involve search. Only 11% use AI alone. Second: the journeys that use both are the most complex and the highest-converting.

Agentic checkout: what's actually live

This is the part where I was wrong last year. I told clients agentic checkout was the thing to prepare for. It went backwards while the unglamorous version — people asking an assistant what to buy, then clicking through to your site — grew 138% and now converts 54% better than everything else.

Spend your Q4 AI budget on being findable and citable, not on agentic checkout integrations. The referral channel is real and working; the checkout channel is not, yet. Re-verify in January before committing 2027 budget.

Beauty and AI: the most actionable data in this whole piece

Similarweb ran topic analysis on actual ChatGPT beauty prompts and cross-referenced which beauty domains over-perform on generative-AI referrals relative to their overall traffic rank. It's the clearest evidence I've found of what kind of beauty brand wins in AI.

Hair care is the top beauty prompt topic, followed by skin care, then makeup, then fragrance at just 19. That's close to the inverse of where retail dollar growth is. Fragrance is the fastest-growing beauty category in both Circana dollars and Similarweb product views, yet it's the least asked-about topic in AI. People don't ask a chatbot what to smell like. They ask it how to fix their hair and skin.

AI's influence also outlasts the session, and there's beauty-specific proof. Similarweb tracked whether users visited an AI-recommended brand versus a named competitor within seven days: Sephora, when recommended, drew 7.9% of users against the competitor's 3.3%; Ulta drew 7.6% against 4.6%. Being the recommended name is worth roughly 1.7× to 2.4× the follow-through of not being it, in beauty specifically, a full week later.

Inside the marketplaces: the fastest-growing AI surface of all

The AI surface growing fastest isn't ChatGPT. It's Amazon's own assistant. Traffic from Rufus / Alexa for Shopping to Amazon product detail pages grew 537% year on year and 26% month on month in June 2026. That's faster than external generative-AI referral growth to the whole of ecommerce.

And it operates on rules that have nothing to do with SEO. Similarweb finds that native marketplace assistants apply a two-tier filter privileging retail-media spend and past purchase history — recommendation carousels integrate sponsored placements alongside previous orders ("one you recently purchased"), creating a retention flywheel for incumbents and raising acquisition costs for challengers. Brands skipping Sponsored Products in conversational interfaces face immediate exclusion.

Who's ready for this, and where the trust stops

27% of US consumers use AI frequently in their shopping journey, while 36% say they never do. Usefulness is roughly 2× higher for Millennials (67%) than Boomers (31%). Men are far ahead of women — 68% have used AI in a shopping journey against 43%.

But willingness drops sharply as AI moves from advising to acting. 38% trust AI to recommend products that fit their needs and 35% rely on it to narrow choices, but only 24% would let it make the final decision — and 42% say outright they wouldn't trust AI to make purchases on their behalf, rising to 57% of Gen X and 70% of Boomers.

One more framing number before we move on, because it settles the "is this still early?" argument: 49.1% of US consumers agree AI has already changed how they browse the internet, staying above 40% in every age band measured, and 43% used ChatGPT in the past month. This is no longer an early-adopter story in any demographic sense. What remains early is the commercial plumbing around it.

Worth remembering where AI doesn't win. Asked which channel makes them feel most in control, shoppers said a major marketplace (37%), a brand's mobile app (26%), the brand's website (21%), social media (9%) — and an AI tool or chatbot: 1%. AI shapes the shortlist and hands off. Salesforce's widely quoted "20% of Cyber Week orders were AI-influenced, worth $67B" is an attribution model covering AI-powered merchandising, search and service as well as assistants. It is not a count of purchases made inside a chatbot, and it's misquoted as one constantly.

H · evidence review

Sustainability: tiebreaker, not driver

Yes it moves the purchase — much less than stated preference implies, and almost never as the deciding factor at a discounted price point.

The gap between what consumers say and what scanner data records is the largest of any behaviour in this piece. 85% say they're more likely to purchase from brands engaged in sustainable practices. Sustainability-marketed products account for about 25% of CPG products sold. That's roughly 60 percentage points between stated preference and revealed share — and the revealed side is scanner data, which is what makes the comparison usable at all.

The commercially interesting half is what happens within that 25%. NYU Stern's Center for Sustainable Business, working with Circana, finds sustainability-marketed CPG products grew 4.9× faster than conventional products and carry a 27% average price premium. So sustainability isn't moving the average shopper, but it is defining a fast-growing, premium-priced minority segment. Meanwhile brand outreach on the topic has fallen — Globescan recorded brands communicating about sustainability dropping from 50% in 2022 to 31%.

Which green claims have evidence behind them?

In beauty, honestly: almost none, at the level of conversion evidence. I looked for published data tying refill formats, waterless formulations, packaging reduction or "clean"/free-from claims to measured conversion lift, and found no primary source with a disclosed methodology. The one adjacent finding with real data behind it is the 27% premium and 4.9× growth rate on sustainability-marketed CPG generally, which includes personal care but isn't broken out for it.

What the beauty data does show is that growth is concentrated in efficacy-led categories. Those are performance claims, not environmental ones. My inference, labelled as inference: in beauty, sustainability is packaging on top of an efficacy argument, and it does not substitute for one.

For supplements, the question is whether third-party testing outranks environmental claims. I can neither prove nor disprove it with published evidence — no survey puts them head-to-head as purchase drivers for VMS buyers. What can be argued from structure: supplements are an ingested, unregulated-before-market product where the buyer's dominant risk is "is this actually what it says it is". NSF, Informed Sport and USP resolve that specific risk; a recycled carton does not. The category's growth data is consistent with this — the fastest-growing segments are efficacy and outcome-led rather than values-led. Treat that as a well-reasoned hypothesis to A/B test on your own PDPs. It's cheap to test: certification badge above the fold versus sustainability badge above the fold, measured on add-to-cart.

I · behaviour

Nine shifts, and what each costs you

Where a widely-repeated claim doesn't survive the data, I've said so rather than quietly leaving it out.

1. The season has flattened, and scarcity is replacing urgency

Around 80% begin before Thanksgiving week and only 17% begin during it; Adobe recorded 25 days above $4B against 18 in 2024. The fastest-rising reason for shopping early is "worried it will go out of stock", up from 33% to 42% — rising faster than price motivations. Do this: real low-stock counts and back-in-stock alerts, not fake countdown timers. Scarcity is becoming a legitimate motivator precisely because it's verifiable, and a fabricated one destroys the trust the real one depends on.

2. Discount scepticism is shallower than you think

The modal minimum discount is 20–25% off; only around three in ten need 30%+; and just 5% won't buy without a discount from a brand they love. BOGO is the single most-preferred promotion, ranked first by 30% of shoppers, ahead of percentage and dollar discounts, with bundles and gift-with-purchase lower. Do this: test the minimum incentive by segment and reserve 30%+ for genuine clearance. For a consumable, BOGO on a second unit is close to a subscription start disguised as a discount.

3. Mobile is the whole game, and the leaks are self-inflicted

Mobile was 56.4% of season dollars and 70% of Cyber Week orders; 86.5% of US consumers shop primarily on a phone; 91% say a great mobile experience matters, with the "extremely important" share up from 41% to 47%. The top cart-killer is unexpected costs at 46%, ahead of changing their mind (35%) and price being too high (32%). Do this: all-in pricing before the final step. It's the largest controllable leak in the funnel and it costs nothing but a decision.

4. Wallets are an expectation now, not a convenience

Expected payment methods: saved card 69%, PayPal 58%, Apple/Google Pay 49%, BNPL 35%, Venmo or Cash App 30%. Among Gen Z, 64% expect Apple or Google Pay — above the 62% who expect a card. BNPL took $20B across the season, up 9.8%, with 82.2% of it on smartphones. Do this: every missing method is a segment of abandoned carts, and for your highest-growth generation the wallet now outranks the card.

5. Trust is moving to owned surfaces. Habit isn't.

Most-trusted place to buy: marketplaces fell 66% → 61%, brand website held at 49%, brand app rose 43% → 47%. Among Gen Z, a brand's own app (58%) now out-trusts Amazon (52%). But on "where do you feel most in control", marketplaces still lead at 37% against 26% for brand apps. Do this: trust is winnable now, habit isn't. Compete on control — effortless checkout, visible order status, easy returns — because that's what converts trust into habit.

6. Loyalty is the one lever with near-universal reach

Loyalty matters to about nine in ten shoppers, with only 10–11% saying it doesn't; 60% of luxury shoppers and 61% of "love-to-shop" shoppers call it extremely important. Do this: almost every other tactic in this piece works for one cohort. This one works for nearly everybody, and it's the cheapest way to make a BFCM discount buy a relationship rather than a transaction.

7. Returns are a margin line, not an ops line

Returns ran at 15.8% of US retail, 19.3% of online sales and 17% of holiday purchases, with 9% judged fraudulent. Do this: at a 19.3% online rate, a 25% discount plus a return is a deeply negative-margin transaction. Reducing mismatch-driven returns is worth more than most conversion work — and since no beauty- or supplement-specific return rate exists, measure your own and treat 19.3% as the ceiling to beat.

8. Push notifications are a depreciating asset

Push attention is slipping year on year and over-notifying accelerates the decline. The notifications that work give a reason to act now: limited-time sales or drops (42%), order and shipping updates (38%), loyalty updates (30%), back-in-stock alerts (25%) — generic product recommendations sit near the bottom at 10%. For genuinely urgent messages, push (31%) beats SMS (19%) and email (8%) among those with a preference. Do this: treat push as scarce inventory. Every low-signal send spends attention you'll want on 27 November.

9. "Arrive to confirm, not discover" — not supported

Covered in full above. In short: the thesis is plausible and the behavioural data contradicts it. Do this: don't strip discovery merchandising out of your landing experience on the strength of a narrative the measurement doesn't support.

Beauty vs supplements, side by side

These two verticals share a customer profile and almost nothing else operationally. Where a dimension has no published data for one of them, I've said so rather than filling it in.

The US and Canada are different seasons

Canada is not a smaller US market with a different currency. The season is longer, the shopper is more defensive and more physical, AI adoption is about half, and both the tax and the consent regimes differ.

The 2026 calendar, with the money on it

Canonical dates with the share of season demand that historically lands in each window. Put your planned Nov–Dec online revenue in and it splits it for you.

The shares are my arithmetic on Adobe's published 2025 US online totals, shown so you can check it, and the three November–December windows sum to 100%. October sits outside the Nov–Dec season definition and is expressed as an equivalent share.

Season-share arithmetic: Cyber Week $44.2B ÷ $257.8B = 17.1%; 1 Nov–1 Dec $137.4B less Cyber Week = $93.2B = 36.2%; December $257.8B − $137.4B = $120.4B = 46.7%; October $88.7B ÷ $257.8B = 34.4% equivalent.

What I'd actually do with this

If I had one afternoon with your Q4 plan and this research open next to it, here's the order I'd go in.

First, separate the macro from your category. Print the sentiment number and your category growth number on the same slide. 47.8 and +7%. Whatever the board decides, it should decide with both in view, because a defensive discount taken on macro anxiety is the most expensive mistake available to you this quarter.

Second, move budget earlier and hold budget later. Thirty percent of shoppers start in September or October and 84% don't expect a deal until November. That window is the only place in the season where you can acquire without discounting. At the other end, December after Cyber Monday is 46.7% of the season — don't liquidate into it.

Third, fix the two leaks that don't cost margin. All-in pricing before the final step, and wallet coverage. 46% of abandons are an unexpected cost; that's a decision, not a budget line.

Fourth, spend the AI money on being findable, not on agentic checkout. Structured product data, ingredient and dose detail as crawlable text, routine and comparison content. That's the work that shows up in a 54%-better-converting channel, and it's also just good SEO, which is beauty's largest acquisition channel anyway. It pays whether or not the AI story continues.

Fifth, be honest about the gaps. You don't have a beauty discount benchmark. You don't have a supplement holiday benchmark. You don't have a category return rate. Those aren't excuses — they're an argument for measuring your own and treating your prior-year cohort data as the only benchmark that describes you.

If something here contradicts what you're seeing in your own account data, trust your account. Panels and surveys describe a market; your store describes your customers. I'd just want to know which of these numbers you can't reproduce, because that's usually where the interesting conversation starts.

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 plan before the budget is locked, that's a conversation I'm happy to have for free.

Say hello →

Sources & method

45 sources. Each entry states what the sample actually is, because a benchmark without a method is an anecdote with a decimal point.