For most of the last decade, Meta advertising rewarded the brand with the sharpest targeting. You built interest stacks, layered lookalikes, split audiences by age and tested one ad in five colours. That playbook was fading for years before Meta's ad retrieval system, Andromeda, finished it off. Your creative now does much of the work detailed targeting used to do.
Q4 is where this bites. Auctions are crowded, competitors are all discounting, and Meta's own results show the average price per ad up 12% year on year in Q2 2026. Walk into Black Friday with ten near-identical versions of one ad and you are bidding with one ad. Meta reads them as the same idea and shows them to the same narrow slice of buyers.
Below we cover what Andromeda changed (using Meta's engineering posts and earnings calls as the record), what counts as a different concept, and how to build, structure and refresh a Q4 creative plan across the US, UK, Canada, Australia and Asia.
What did Meta's Andromeda update change?
Meta's ad system works in stages. Before any ad is ranked or bid on, a retrieval step picks a few thousand candidates out of tens of millions of live ads for each person scrolling. Andromeda is that retrieval step. Meta's engineering team announced it in December 2024, describing a 10,000x increase in model capacity, a 6% gain in recall and an 8% gain in ads quality on the segments it tested.
Meta's stated reason for building it: more than a million advertisers had used its generative AI tools to make more than 15 million ads in a single month. The old retrieval system could not sort that volume of creative, let alone read what each ad was about.
Meta has kept building on it. On the Q3 2025 earnings call, the company said it had combined retrieval and early-stage ranking into a single model, driving a 14% increase in ads quality on Facebook. Meta's ranking model, GEM, sits further down the pipeline. Meta says GEM lifted ad conversions 5% on Instagram and 3% on Facebook Feed in Q2 2025.
In plain terms, Meta reads the content of your ads and uses it to decide who sees them. Your creative brief is now your targeting brief too.
Some things did not change. The system still learns from the conversion events you send it, so a clean pixel and Conversions API setup with accurate purchase values matters as much as ever. Andromeda decides which ads are worth considering for a person. Your signal quality still decides how well Meta can tell which of those people buy.
What Meta says its newer ad models delivered
Meta's ad system works in stages.
Why did making more variations of one ad stop working?
Meta's system can now tell they are the same ad. A new headline over the same product shot, a different background colour or a swapped first frame does not give retrieval anything new to match. Those ads end up competing with each other for the same buyers while the rest of your potential market never sees you.
Meta's own training material spells it out. A 2026 Meta Blueprint workshop on creative diversification tells advertisers it is "no longer just about volume, it's about number of concepts" rather than number of assets, and notes that "the vast majority of ad sets remain under-diversified". Meta spokesperson Alisha Swinteck told Marketing Brew that the company's guidance is "deliberately outcome-based, like 'diversify your creative, test, and let the system optimize'", and declined to explain how its ranking models represent individual assets.
Ads Manager now has a creative diversity column (Columns, then Customize columns) that grades each ad set Low, Medium or High. Meta labels it as in development. When Search Engine Land tested it on client accounts, several ad sets the author considered well diversified still came back Low, so use it as a rough check and not a verdict. Popular vendor rules of thumb, such as running a fixed number of concepts per campaign or keeping a similarity score under a set threshold, are agency guidance and not Meta rules.
What counts as a different creative concept?
Think of a concept as a distinct reason to buy, shown in a distinct way. Six levers do most of the work. Change two or three of them together and you have a new concept. Change one cosmetic detail and you have a variation, which is still useful for testing within a concept, just not as a substitute for new ideas.
| Lever | What changes | Counts as new | Does not count |
|---|---|---|---|
| Format | How the ad is built | Static vs 30-second demo video vs carousel vs creator Reel | Same video cropped to 4:5 and 9:16 |
| Hook | The first two seconds or the headline idea | Problem callout vs bold claim vs reaction shot | Same hook with new wording |
| Angle | The reason to buy | Gifting vs self-reward vs value vs ingredients | Same angle, new adjective |
| Person | Who is on screen or speaking | Founder vs customer vs creator vs no person | Same creator in a different shirt |
| Setting | Where and when it happens | Bathroom routine vs office desk vs festive table | Same set, new lighting |
| Offer | What the deal is | Gift with purchase vs bundle vs free shipping vs no offer | 20% off vs 25% off |
What does a Q4 concept matrix look like for a Shopify brand?
Take a mid-sized skincare brand going into peak. Ten concepts, each built for a different buyer and moment. Every row changes at least three levers, and each concept can carry two or three internal variations of its own.
| # | Concept | Format | Person | Offer | Best window |
|---|---|---|---|---|---|
| 1 | Gift for her, under US$60 | Carousel of gift sets | No person, product flat lays | Free gift wrap | Early Nov to mid Dec |
| 2 | My 3-step winter routine | Creator Reel, talking to camera | Creator, 30s | None | All of Q4 |
| 3 | Founder explains the formula | Video, single take | Founder | None | October build phase |
| 4 | Real results at week 4 | Before and after static | Customer | Starter kit | October and November |
| 5 | Dermatologist reacts | Split-screen video | Expert | None | All of Q4 |
| 6 | Black Friday bundle value | Static with price stack | No person | Bundle saving | BFCM week only |
| 7 | Unboxing the advent calendar | UGC video | Customer | Limited stock | Mid Nov to early Dec |
| 8 | Thousands of five-star reviews | Review collage static | Many customers | Free shipping | All of Q4 |
| 9 | Last day for Christmas delivery | Short countdown video | No person | Express shipping | Mid December |
| 10 | Treat yourself after the gift run | Lifestyle video | Creator, 20s | Gift with purchase | Late Dec, Boxing Day |
How can a lean team produce ten concepts before Black Friday?
Many teams assume diversification means a bigger production budget. Mostly it needs a different brief. One shoot day can produce several concepts if you plan for it, because the person, setting and hook can change between set-ups while the crew and products stay the same.
Meta's own tools help at the variation level. Meta reported on its Q3 2025 call that the number of advertisers using at least one of its video generation features rose 20% on the previous quarter, and its engineering team says advertisers using Advantage+ creative saw a 22% increase in ROAS. Those are Meta's figures. Use the tools to stretch a concept into new formats and sizes. The thinking behind the concept still has to come from you.
- Write the concept list first, then the shot list. Each concept gets a one-line buyer, reason and hook before anyone picks up a camera.
- Plan two or three set-ups per shoot day: a founder piece, a routine in a real home, and a product-only table shot.
- Ask customers and creators for raw footage now. Seeding product in October gives you real reaction content by mid November.
- Mine your reviews and support tickets for angles. The objections buyers raise are ready-made hooks.
- Keep a few statics in every set. They are quick to make, quick to swap when an offer changes, and often carry offer concepts well.
How should you structure Meta campaigns for peak season?
Fewer, larger campaigns give the system more creative to choose from and more conversions to learn from. Meta reported on its Q3 2025 call that sales, app and lead campaigns now start with end-to-end Advantage+ automation turned on, letting the system choose audiences and placements. Advantage+ shopping campaigns were renamed Advantage+ sales campaigns in 2025.
A peak-season structure that suits most Shopify brands:
- One main Advantage+ sales campaign for prospecting, broad targeting, with your full set of distinct concepts. Use audience suggestions rather than hard interest limits.
- A separate campaign or ad set only where the goal or budget differs, such as a market with its own currency and shipping, or a short BFCM offer you want to switch off on a set date.
- Existing-customer controls set correctly, so you can see how much spend goes to people who already buy from you.
- No duplicated ad sets running the same ads against the same audience. Duplicates split your learning and push you into bidding against yourself.
- Budget raised in steps through late October and November, rather than one large jump on Black Friday morning.
When do Q4 peaks land in each market?
Peak is not one weekend everywhere. If you sell into several markets, your concept calendar needs to follow local dates, and so do your budget steps. Black Friday falls on 27 November 2026 and Cyber Monday on 30 November.
| Market | Key dates | Creative implication |
|---|---|---|
| US | Thanksgiving 26 Nov, Black Friday 27 Nov, Cyber Monday 30 Nov, shipping cut-offs mid Dec | Heaviest BFCM competition. Have offer concepts ready before Thanksgiving week. |
| UK | Black Friday 27 Nov, Boxing Day 26 Dec | Two peaks. Plan a post-Christmas concept set for Boxing Day sales. |
| Canada | Black Friday 27 Nov, Boxing Day 26 Dec | Boxing Day week is a major sale period. Run CA$ pricing in creative. |
| Australia | Click Frenzy in mid Nov (2026 dates not yet announced), Black Friday 27 Nov, Boxing Day 26 Dec | Black Friday now sits alongside Click Frenzy. Summer settings in creative, not snow. |
| China and SEA | Singles Day 11 Nov, 12.12 on 12 Dec | Marketplace-led peaks. For Meta, mostly relevant in SEA markets such as Singapore, Malaysia and the Philippines. |
| India | Diwali 8 Nov | Festive buying peaks before Black Friday. Gifting concepts need to be live in October. |
How often should you refresh creative through BFCM and December?
Refresh when the numbers tell you to, rather than on a fixed calendar. Peak speeds everything up: frequency climbs faster and offers change weekly. Watch each concept's cost per purchase and click-through rate against its own first week. When both drift the wrong way while frequency rises, replace the concept. Tweaking the headline rarely buys more than a few days.
- October: launch your evergreen concepts and let them gather data. This is where you find out which angles carry.
- Early November: add gifting and early-offer concepts. Retire anything that never got spend.
- BFCM week: switch in offer concepts with fixed end dates. Keep evergreen winners running alongside them.
- December: rotate to shipping-deadline, last-minute gifting and gift card concepts as cut-offs pass.
- Late December: Boxing Day and self-purchase concepts for the UK, Canada and Australia.
Your next step
Build your Q4 concept list from your own store
Enter your store URL and get a starting Meta ads strategy with audiences, creative angles and concept briefs built around what you actually sell.
What should you measure to know if diversification is working?
Ad-level ROAS is a weak guide here, because Meta shows different concepts to different people by design. A concept with modest ROAS may be the one opening a new segment. Judge the campaign and the business instead.
For context on costs, Triple Whale's data across its ecommerce brands put Meta CPMs at US$22.26 over BFCM 2025, up 7.6% on the year before. Meta's own results show average price per ad up 6% year on year in Q4 2025 and 12% in Q2 2026. Plan peak budgets from those trends. Your September CPMs will understate what you pay in late November.
- Spend distribution: how many concepts are getting meaningful spend. If one ad takes nearly all of it, you probably lack real diversity.
- New customer share: the proportion of purchases from first-time buyers, from Shopify order data.
- Blended cost per new customer across Meta and other channels, week on week through peak.
- Frequency and CPM by campaign, to spot fatigue early.
- Incrementality: a conversion lift study or a geo holdout if your spend allows it.
Key takeaways
- Andromeda reads your ad content and uses it to choose who sees each ad, so creative now carries much of the targeting load.
- Near-identical versions of one ad compete for the same buyers. Build distinct concepts by changing format, hook, angle, person, setting or offer together.
- Run fewer, broader Advantage+ sales campaigns with all your concepts inside, and avoid duplicated ad sets.
- Match your concept calendar to local peaks: BFCM in the US, Boxing Day in the UK, Canada and Australia, Singles Day and 12.12 in SEA, Diwali in India.
- Judge success on new customers and blended cost per acquisition, not on which single ad shows the best ROAS.
- 1.Meta Engineering: Meta Andromeda, next-gen personalised ads retrieval engine (Dec 2024)
- 2.Meta Engineering: Meta's Generative Ads Model (GEM) (Nov 2025)
- 3.Meta Q3 2025 earnings call transcript
- 4.Meta Q4 and full year 2025 results (SEC filing)
- 5.Meta Reports Second Quarter 2026 Results
- 6.Meta Blueprint: Creative Shop workshop on creative diversification
- 7.Marketing Brew: how advertisers approach Meta creative diversification
- 8.Relevant Audience: Meta tells advertisers near-identical creative counts as the same ad (citing Marketing Brew)
- 9.Search Engine Land: Meta's creative diversity metric, tested
- 10.Common Thread Collective: Meta's Creative Diversity score in Ads Manager
- 11.Search Engine Land: Meta simplifies Advantage+ campaign setup, renames shopping to sales campaigns
- 12.Triple Whale: BFCM 2025 data release
- 13.Tom's Guide: Click Frenzy Main Event 2025 dates
- 14.Diwali 2026 date