How to read this article based on your profile
- Creative agency that doesn't run paid media: read §1 and §2 to understand what changes, then §6 for the multi-touch pitfall that concerns you directly.
- Marketing manager / CMO: read §3 (decision matrix) and §5 (3 structures) to decide what to ask of your acquisition agency.
- Experienced acquisition agency: read §4 (cases where the classic funnel stays valid) and §5 (HS structures) to compare against your current method.
1 · AndroMeda in 2 minutes: it starts from the ad, not the audience
AndroMeda fully replaces the engine that delivers ads at Meta (the machine that has run Facebook ads from the start). The system works in 2 steps: it first picks candidates from tens of millions of eligible ads, then ranks them to decide which one wins the final auction (an engine called Meta Lattice handles that). Technical details on the Andromeda and GEM infrastructure.
The shift in logic: before, you built audiences by hand (interests, lookalikes, custom lists) and pushed creatives onto them. AndroMeda flips the logic. The algorithm reads the ads themselves (the message, the story, the visual signals) and decides who sees what. The creative now does the work the algorithm used to do on the targeting side.
Concretely, this means:
- Your hand-built audiences move to the background, and can even hurt if they block the algorithm's learning too much
- Variety is rewarded: 10 different variants work better than 3 that look alike
- Your ads wear out faster and it shows sooner, because the algorithm cycles more quickly
- Classic metrics like ROAS looked at on its own become misleading (see §7)
The impact on your business is measurable. According to eMarketer 2026 Meta Ads benchmarks: the average Meta click-through rate (clicks per 100 impressions) rose from 0.9% in 2024 to 2% in 2026, a doubling, driven by accounts that moved to AndroMeda with creative volume. But the cost per 1,000 ad impressions (CPM) rose 70% in one year (from $22.76 in January 2025 to $38.70 in January 2026, source eMarketer). The pressure on poorly structured accounts is growing, with numbers to back it up.
To dig into the technical definition of AndroMeda and the history of Meta updates, the view of an expert French-speaking agency: Meta Andromeda, the guide to survive and scale in 2026 by Danilo Duchesnes.
2 · What changes in your account structure
For a creative agency that has never run a Meta account, here are the 3 major changes:
Change 1: a sharp cut in the number of campaigns
Before (classic funnel): 3 to 8 campaigns per funnel stage, several ad sets per campaign, several ads per ad set. A complicated tree.
Now (AndroMeda): 1 test campaign, 1 scale campaign, 1 retargeting campaign (the last one optional depending on profile). That's it. If you see a Meta account with 15 active campaigns in 2026, there's an 80% chance the structure is outdated.
Change 2: the creative brief becomes strategic
No more one-shot hero campaigns. You produce 10 to 20 variants per campaign to give the algorithm something to work with. The art director's role becomes: producing genuinely different variants (not 3 versions of the same image), with different angles (social proof, product demo, customer testimonial, real-life setting, urgency, etc.).
This precise point is what transforms the creative agency's job. You no longer judge an ad around a table, you look at the real numbers in 7 to 14 days. See article 4: long-planned creative and UX are the exception, not the norm.
Change 3: the metrics to watch change
ROAS looked at on its own becomes misleading. The new metrics to watch first (see §7) are the creative engagement signals: hook rate at 3 seconds (how many people stay past the first 3 seconds), hold rate at 15 seconds, average view rate, click-through rate on a cold audience, cost per add-to-cart. Final ROAS is still measured, but further down the journey, not as a metric to decide a creative on the spot.
3 · The HeySquad 4-point decision table
Here are the field thresholds HeySquad applies on its Belgian missions 2023-2026 to choose between AndroMeda and the classic structure.
Point | HS field threshold | Verdict |
|---|---|---|
1. Monthly Meta Ads budget | ≥ €3k/month Belgium | AndroMeda holds up. Below that, the algorithm doesn't have enough to learn from |
2. Tracked conversion history | ≥ 30 days minimum | AndroMeda holds up. Below that, the brand is too young or the tracking too recent |
3. Pixel quality and tracked events | Complete set-up (events properly set, server-side if possible) | AndroMeda holds up. A limited or broken pixel means AndroMeda learns on junk data |
4. Business profile | B2C e-commerce, or B2B that generates a lot of leads | AndroMeda holds up. Niche high-ticket B2B, the classic funnel stays relevant |
Important note: the sales cycle length and catalog size are NOT decision points. AndroMeda works just as well on a single product as on a 500 SKU catalog. The cycle length is secondary, it's the quality of the pixel tracking that decides, not the number of days to close a sale.
If you tick all 4 points: switch to AndroMeda full. If you tick 3 out of 4: a hybrid structure is worth considering. If you tick 2 or fewer: stay on the classic funnel while you fill the missing points.
To bring your tracking up to the required level, see server-side tracking.
4 · The 4 cases where the classic funnel stays valid
The clear-cut opinion isn't absolute. 4 situations justify keeping the classic structure in 2026.
Case 1: high-ticket B2B, or a long sales cycle
Annual contracts above €50k, sales cycle over 90 days, decision made by several people. The AndroMeda algorithm needs a lot of conversions to learn. In high-ticket B2B with 5 to 10 qualified leads a month, it doesn't have enough to work with. The classic funnel, with audiences segmented by hand (industry interests, lookalikes of your customers, retargeting site visitors), stays more controllable.
Case 2: new brand without 30 days of conversion history
Without enough data in the pixel, AndroMeda learns on junk data. Mandatory first step: 30 to 60 days of simple campaigns (search ads on your brand, plus one image awareness campaign) to build up conversions. Only then consider AndroMeda.
Case 3: monthly Meta budget under €3k in Belgium
Below this threshold, the algorithm doesn't get enough to optimize. A classic structure run by hand, with precise targeting on a few custom and lookalike audiences, is better. AndroMeda needs budget to learn.
Case 4: limited pixel or tracked events in the account set-up
Many Meta accounts arrive on a HeySquad mission with a poorly configured pixel: missing events (add-to-cart, initiate checkout, purchase), no server-side tracking, and Meta's conversion counting left on its default, with no consistency across channels. Before switching to AndroMeda, we get the tracking clean. Otherwise the algorithm learns on junk data and results collapse within 7 to 14 days.
Field honesty: even when the classic funnel is technically the right choice in these 4 cases, it has become hard to control lately. Meta is pushing the algorithm into every layer of the product. Keeping a clean manual structure takes more vigilance than before. That's a reality to factor into the decision.
5 · The 3 account structures HeySquad uses on missions
Structure 1: AndroMeda full
Typical client profile: B2C e-commerce mostly, an e-shop doing at least €250k revenue a year, catalog size irrelevant (single product or 500 SKUs both fine).
Monthly budget: from €3k/month of Meta spend in Belgium.
Account structure: 1 test campaign, 1 scale campaign, 1 retargeting campaign. That's the absolute minimum. No more campaigns running at once, except in a very specific case (a separate test by region, or a bilingual FR/NL test to split out).
Creative volume: 10 to 20 variants per campaign minimum to give the algorithm something to work with. A variety of angles is mandatory (at least 5 distinct angles).
Budget split: 20-30% on the test, 60-70% on scale, 10-20% on retargeting depending on account maturity.
Structure 2: hybrid, AndroMeda plus a brand boost
Typical client profile: accounts that boost organic posts to their usual audience (creators, brands with an active LinkedIn or Instagram community, B2B service firms that publish content regularly).
Monthly budget: from €3-5k/month on Meta.
Account structure: 1 AndroMeda campaign (test and scale combined depending on maturity), plus 1 classic brand awareness campaign dedicated to boosted posts and sponsored organic content. The second campaign supports the editorial side without entering AndroMeda's logic.
Creative volume: 10-20 variants on the AndroMeda side. 5-10 boosted organic posts a month on the brand side.
Structure 3: classic three-stage funnel
Typical client profile: niche high-ticket B2B, new brand without history, accounts that don't meet the 4 points of the table in §3.
Monthly budget: generally below €3k/month on Meta.
Account structure: a classic three-stage funnel (top, middle, bottom) with audiences segmented by hand (industry interests, lookalikes of your customers, retargeting).
Creative volume: 3-6 variants per funnel stage. Less pressure on variety than AndroMeda.
Honest caveat: hard to control lately. Meta pushes Advantage+ recommendations into every layer. Keeping a clean classic structure takes more vigilance than before.
6 · The HeySquad pitfall: don't cut an ad on its numbers seen in isolation
Creative agencies discovering performance tend to reason ad by ad: "this ad doesn't convert, we cut it." That's the most frequent mistake we see in audits.
An ad not converting doesn't mean it isn't helping the rest of the buying journey.
An ad with no directly attributed conversion can be the ad that:
- Introduced your brand to a prospect who then clicked another ad and bought (a conversion several steps contributed to)
- Produced the 15-second hold rate needed to push the prospect into a retargeting audience that does convert
- Fed the algorithm signal that then helped it target the other variants better
Routinely cutting "underperforming" ads based on a 7-day ROAS seen on its own breaks this AndroMeda mechanic (multi-touch attribution credits every step of the journey, not just the last click). The system is built to combine signals from several ads across several touchpoints. You judge a campaign as a whole (multi-touch attribution, plus the surplus of conversions versus a control group that didn't see your ads), not ad by ad.
It's a pitfall every creative agency moving toward performance falls into, in their first 3 months. That's why this article is for them.
7 · The metrics to watch first
Final ROAS is still measured, but further down the journey. The metrics that help you decide a creative on the spot sit early in the journey:
Metric | Target threshold | Why |
|---|---|---|
Hook rate at 3 seconds | > 50% | Measures whether the creative grabs attention. The first signal that counts in AndroMeda. |
Hold rate at 15 seconds | > 25-30% | Measures whether the creative holds attention. A solid interest signal. |
Average view rate | > 30-40% on video | Confirms the chosen angle works. |
Click-through rate on a cold audience | > 1% | Raw interest signal, without the bias of an already-warm audience. |
Cost per add-to-cart (B2C e-com) | Per target cost per conversion | A signal that comes before the purchase, faster to measure (3-7 days vs 30 days for purchase ROAS). |
Purchase ROAS / cost per qualified lead | Per business cost per conversion | To measure across the whole journey, not ad by ad. A weighted reading. |
If the early-journey metrics (hook rate, hold rate, click-through rate) are good but the end-of-journey ROAS is poor, the problem is after the click (the landing page, the buying journey), not the creative. That's exactly what free Microsoft Clarity lets you dig into on the UX side.
