Meta · Facebook · Instagram
Information about the new “Andromeda” algorithm
The Andromeda algorithm is a new generation of ad delivery in the Meta ecosystem. Below is a structured overview of materials, parameters and target metrics.
General information about the algorithm
The Andromeda algorithm is a new generation of ad delivery in the Meta ecosystem (Facebook, Instagram and others).
Key features:
- It replaces the old targeting logic (demographics, interests) and focuses on behavioral signals, creative assets and automation.
- The system now answers not so much “who should see the ad” as “which ad should be shown to which user”.
- It requires more creative variety and simpler campaign structures: broad audiences and fewer small ad sets.
- Technically, it is based on deep neural networks, specialized hardware (for example, NVIDIA Grace Hopper Superchip) and hierarchical creative indexing.
- It is not a magic button: results depend on creative quality, conversion data and the right strategy.
Previously, advertisers focused on audience segments and many settings. Now it is more important to give the algorithm room to work (broad audience + many creatives) and let it optimize delivery. Precise interests and many small ad sets become less effective.
Materials for Andromeda
What materials are needed for effective work
For Andromeda to learn and optimize delivery, it needs a varied set of creatives, correctly configured tracking and a simple campaign structure.
The set of materials determines how flexibly the algorithm can select ads for different audience segments. The broader the audience and the more varied the creatives, the faster Andromeda finds working combinations.
Prepare several ad concepts, headline and copy variants, a landing page with a clear conversion and correct event tracking through Pixel/Conversions API. Simplify campaign structure so the signal is not diluted.
All materials should be aligned with business goals: which actions count as conversions, which audiences matter first and which KPIs are expected at the test stage.
A typical Andromeda launch set includes:
- 15-30 second video creatives with different messages and formats.
- Static banners and carousels with visually different concepts.
- Several headline and body copy variants for different pains/benefits.
- A landing page or website with a clear target action and UTM tagging.
- Configured Pixel and Conversions API with events for key conversions.
- A campaign with broad audience, automatic placements and basic exclusions.
- A plan to refresh creatives every 7-14 days.
Set
Start
| Element | Quantity / duration | Format / resolution |
|---|---|---|
| Main video | 1 video 20-30s | 16:9 — 1920×1080 |
| Short versions | 2 videos: 6s and 15s | 16:9 / 9:16 — 1080×1920 |
| Vertical formats | 1-2 additional videos | 9:16 — 1080×1920 |
| Carousel | 5-7 cards | 1080x1080 or 1080x1350 |
| Previews / covers | 3-5 variants | 1280x720 or 1920x1080 |
Practical example
What a starter set can look like
This set gives Andromeda enough signals: short videos work for reach and hook testing, a longer video explains the offer, a carousel details the product, and covers improve CTR.
In practice, each audience segment should have its own “video + carousel” combination, but even a basic set of 3-4 videos and 5-7 carousel cards already lets the algorithm optimize delivery consistently.
Parameters of these materials
Text description
When preparing materials, define parameters that help Andromeda work optimally: creative variety, broad audience, regular refreshes, correct tracking and a simple campaign structure.
| Parameter | Recommended value | Notes |
|---|---|---|
| Number of creatives | 8-15+ per ad set | Quality variety is what matters |
| Audience | Broad, without many small segments | Gives the algorithm more data for learning |
| Formats | Video + images + carousel | Allows testing different presentation formats |
| Creative refresh | Every 7-14 days | Helps avoid ad fatigue |
| Conversion tracking | Pixel + Conversions API on key events | Without signals, performance drops |
| Campaign structure | 1 campaign → 1 ad set (or minimal structure) | A simplified structure works better for the algorithm |
| Campaign budget optimization (CBO) | Enabled | The algorithm allocates budget itself |
| Exclusions | Exclude existing customers/non-target traffic | Cleans the signal and improves learning |
Metric focus
Key indicators
The chart shows which metrics to use when evaluating campaigns: return on ad spend, conversion cost, click-through rate, conversion rate and regular creative refreshes.
Possible performance benchmarks
How to calculate results in Andromeda
After launch, track ROAS, cost per conversion (CPA), CTR, conversion rate (CVR), creative refresh frequency and budget spend. Below are approximate target values for these metrics.
| Metric | Target value* | Comment |
|---|---|---|
| ROAS | 3–5× | $3-5 return per $1 spent |
| CPA | Depends on the niche (for example, $10-50) | Depends on average order value and competition |
| CTR | 1.0–3.0 % | Working range for social networks |
| CVR | 2.0–5.0 % | Conversion from click to target action |
| Refresh frequency | Creatives every 7-14 days | Required for stability |
| Creatives per campaign | 8–15+ | Minimum for testing hypotheses |
*Values are approximate and depend on the niche, market and offer.
Metric calculation with a $5 per day budget
Suppose we launch a campaign with a $5 per day budget and these target metrics: ROAS = 4×, CPA = $20, CTR = 2%, CVR = 3%. Below is an approximate calculation by day, week and month.
Budget dynamics
$5 per day → $150 per month
Spend grows linearly: $5 per day → $35 per week → $150 per month. With a target ROAS of 4×, expected return is about $600 if CPA, CTR and CVR stay stable.
Calculation table
| Period | Budget | Expected ROAS | Expected revenue | Approximate conversions |
|---|---|---|---|---|
| 1 day | 5 $ | 4× | ≈20 $ | ~0–1 |
| 7 days | 35 $ | 4× | ≈140 $ | ≈2 |
| 30 days | 150 $ | 4× | ≈600 $ | ≈7–8 |
If ROAS rises to 5×, revenue will be higher, while a higher CPA will reduce the number of conversions. It is important to test creatives regularly, refresh combinations and give the algorithm time to learn.