Autumn 2026 vs autumn 2025
Sales are holding last year's record level. New customers are the gap, and we know why.
We compared 1 August to 27 October this year with the same weeks last year, across the shop, Meta, email and the ads themselves. Three things explain most of the gap, and each one has a fix we can make before Black Friday.
The three things behind it
- Meta stopped reaching warm shoppers. Last autumn about a quarter of Meta revenue came from people who already knew Sew Heart Felt: past customers and people who had engaged but not bought. The new Evergreen campaign shuts past customers out, and nothing replaced that warm audience.
- Our best-proven ads are getting almost no budget. The posts with thousands of reactions are sitting idle, and 2 in 5 pounds now goes on posts with under 500 reactions.
- The abandoned cart emails have nearly stopped sending. A settings clash between two cart flows means about 1 in 7 of last year's cart emails go out. That's roughly £400 a week, more in November and December.
1. The shop: flat, not falling
2025 was a breakout autumn, up 93% on 2024 for these weeks. 2026 has matched it. Orders are slightly down, but people are spending more per order, so revenue has held.
Weekly shop revenue
Direct-to-consumer sales after refunds. Week of 24 Oct not shown (only 4 days in).
New-customer orders per week
The soft spot: new customers are 4 to 6% behind most weeks since September.
What's really behind the flat line
- Fewer new visitors. Unique visitors are down 17% (69.5k vs 83.9k).
- The ones who arrive buy more. Orders per visitor are up from 4.4% to 5.1%.
- Returning customers are steady: 899 orders vs 902.
- The gap is top-of-funnel. Meta is bringing fewer new people to the site, and each one costs more.
2. Meta: the drop lines up with the new campaign
Spend is close to last year (£33.6k vs £35.4k, 1 Sep to 27 Oct). What changed is how hard each pound works. We compare on 7-day click so both years are measured the same way.
Meta ROAS per week (7-day click)
August 2026 was ahead of last year. It started slipping in September and fell further from late September, when spend moved into the new Evergreen campaign.
First week of August 2025 had a tracking gap, so it is left out.
What moved
People click the ads more than last year. Fewer of those clicks turn into orders, and each view costs more.
Last year, warm shoppers punched above their weight
Aug to Oct 2025. Engaged people and past customers took 14% of spend and returned 23% of Meta revenue, at 3.2x and 5.2x.
The same ads, in the new campaign, did worse
| Ad | Old campaign | New campaign |
|---|---|---|
| Knitting Nell catalogue | 2.20x | 1.65x |
| Sheply unboxing | 2.14x | 1.51x |
| All Tiny Animals catalogue | 2.27x | 1.72x |
The same creative lost about a quarter of its return when it moved. Even on cold audiences alone, the old campaign ran at 2.29x this summer against 1.62x for the new one. Our over-55 buyers, the biggest group, fell most (2.86x to 1.87x), which fits with past customers being shut out.
3. Halloween: the products sold, the ads aimed cold
Halloween wasn't a product miss. Sales of Halloween items nearly doubled. Two things held it back: the ads and the offer.
Halloween product sales
15 Sep to 27 Oct. Agatha bundle (£6.3k), Tiny Animals bundle (£2.5k), Lucky bag tag (£2.1k) and Pumpkin (£1.6k) led.
Who the Halloween ads were shown to
2025 put three quarters of Halloween spend on warm shoppers at 2.75x. 2026 was all cold audiences at 1.84x.
Two more things that changed
- The hero offer was retired. 2025's Witch Mouse plus a free Black Cat for £19.95 sold 294. In 2026 the nearest thing was a £29.68 bundle, which sold 65. A named free character beat a percentage off both years.
- Stock was tight early. The Black Cat was down to 11 on 20 Sep and Agatha on a Broom sold out on 10 Oct, just as the season peaked.
- It took sales from the rest of the range. Halloween added £9.7k while core tiny animals fell £8.4k, so the season replaced sales rather than adding to them.
What grew and what shrank
Product line revenue, 15 Sep to 27 Oct, change versus last year.
Rugby and cricket mice had a one-off lift in 2025 from the Women's Rugby World Cup in England (22 Aug to 27 Sep 2025).
4. The ads themselves: proof matters, and our proof is sitting idle
Ads on posts with more reactions and comments earn more in both years. This autumn, much more of the budget runs on new posts that haven't built that up yet.
Spend on low-proof posts
Share of Meta spend on posts with few reactions.
ROAS by social proof
Higher-proof posts earn more in both years. Part of this is simply that winners collect reactions.
Last year's proven winners, this autumn
| Ad | 2025 ROAS | Spend Sep to Oct 2026 |
|---|---|---|
| Heart & Flowers "Over 100" (4.2k reactions) | 5.08x | £26 |
| Bestsellers 2025 V1 | 3.55x | £350 |
| Heart & Flowers catalogue unboxing | 2.62x | £251 |
| Bestsellers V4 carousel, Heart Balloon, Birthday Balloon | 2.0x+ | £0 |
6 of the 8 strongest 2025 ads got £350 or less. When a proven video was copied onto a new post it lost its reactions (Sheply: 4,800 down to 177) and about 8 to 10% of its return. On creative variety: across the account we now run more different ads than last year (19 ads carry 80% of spend, against 8). The "low diversity" flag is about one ad set, where three box-opening videos take 82% of spend.
5. Email: steady overall, with two problems to fix
Email revenue is level with last year (£42.6k vs £41.5k over the same 8 weeks): more campaigns made up for weaker flows. Two flow problems need fixing before peak.
Abandoned cart emails sent per week
Shoppers are adding to cart at about the same rate as last year. The emails aren't going out.
Sep 1 to Oct 27 cart-email revenue: £896 this year vs £4,141 last year.
Why the cart emails stopped
Every time someone adds to cart, two tools record it: Shopify and Unveild, the visitor-recognition app. Each one starts its own cart flow.
- The main cart flow removes anyone with an Unveild add-to-cart. That event fires at the same moment, so almost everyone is removed straight away.
- The Unveild cart flow removes anyone who adds to cart again. Unveild often records each click twice, so most people are removed from this one too.
Fix: one flow does the job and the other only picks up people the first one can't. It's about 20 minutes in Klaviyo.
Bot checkouts on 20 and 21 October
About 10,900 fake checkouts in two days. The abandoned checkout email then sent to them, and about 9,500 emails bounced.
Most of these fake profiles are marked as subscribed, so they also receive campaigns. Left in, they hurt inbox placement for real customers going into peak.
The plan before Black Friday
Launch a separate warm campaign
Past customers and engaged non-buyers, with last year's proven ads, on its own budget so we control exactly what it spends. Main Evergreen keeps excluding customers, so new-customer spend stays clean.
| Ad set | Nov | Black Friday week | 2 to 20 Dec |
|---|---|---|---|
| Past customers | £50/day | £80/day | £40/day |
| Warm non-buyers (site visitors, add to cart, Instagram and video engagers) | £80/day | £120/day | £60/day |
| Total | £130/day | £200/day | £100/day |
Based on last year: in November 2025 these two groups got about £105 a day and returned 5.7x (warm) and 9.2x (past customers) on Meta's numbers. Each ad set gets its own budget so neither can take the other's money. Past customers also get our emails, so some of these sales would happen anyway. We'll hold 20% of the customer list out of ads and compare their orders after four weeks to see what the ads really add.
Put the proven winners back to work
A "Proven" ad set with a minimum spend, using the original posts so the reactions carry over: Heart & Flowers Over 100, Bestsellers 2025 V1, the original Sheply unboxing, H&F catalogue unboxing.
Fix the cart flow and clean out the bots
One primary cart flow, duplicate-event exits removed. Suppress the ~10.9k bot profiles created on 20 and 21 Oct, and turn on bot protection at checkout.
A cheap named-character Christmas offer
Last year's free Black Cat was the strongest offer of the season, and a free named character beat a percentage off both years. A Christmas version (a free small character with a hero product) gives warm shoppers a reason to buy now.
Judge Meta on the shop, not on Meta
We'll score Meta weekly on new customers and cost per new customer from the shop, plus 7-day click results, so changes in how Meta counts sales can't hide or exaggerate progress.
How we measured
- Shop revenue: direct-to-consumer Shopify orders (website and Shop app) after refunds, excluding Etsy, Amazon, Faire and wholesale.
- New vs returning: Triple Whale.
- Meta ROAS: 7-day click in both years. This year's new ad sets also count 1-day "engaged views", which would make 2026 look better than a fair comparison.
- Email revenue: Klaviyo's own attribution.
- Windows: 1 Aug to 27 Oct for weekly charts, 15 Sep to 27 Oct or 1 Sep to 27 Oct for headline comparisons, labelled on each chart. The last few days of October will move slightly as Meta's reporting settles.