Creative Volume for Meta Andromeda
Creative
A production system for Meta's emphasis on varied creative inputs.
Overview
Meta's Andromeda retrieval engine rewards accounts that feed it a steady stream of distinct creative inputs. Rather than betting on a single hero asset, this workflow builds a production system for supplying varied hooks, formats, angles and messaging combinations so the ad-selection model has room to find the right creative for each person. The outcome is a repeatable pipeline that keeps enough fresh, differentiated creative in market to prevent fatigue and give Andromeda the signal density it needs to optimise.
Working through OpenAds, you keep an AI assistant on the analysis and drafting work — auditing current creative diversity, proposing new variant batches, mapping them to ad sets and drafting launch structures — while every publish, budget change and status flip is held for your approval. You end each cycle with an approved batch of creatives live, a clear record of what changed, and a measurement plan for which variants to scale and which to retire.
When to use it
Reach for this workflow whenever creative supply, not budget, is the constraint on a Meta account. It is most valuable when performance has plateaued despite healthy spend, or when you are deliberately building for Andromeda's preference for volume and variety.
Your top ad sets are running on two or three creatives and CPMs are climbing from fatigue.
You are launching a new Advantage+ or manual campaign and want breadth from day one.
You have a backlog of raw assets — clips, statics, testimonials — that has not been turned into structured variants.
You want a recurring cadence (weekly or fortnightly) of fresh creative rather than ad-hoc uploads.
You are consolidating fragmented ad sets and need enough creative to fill broader targeting.
Example prompts
"Audit creative diversity across my Meta prospecting campaign and flag any ad set running fewer than four distinct creatives."
"Draft a batch of 12 creative variants across three hooks and two formats for our winter sale, mapped to the assets in the account."
"Propose a launch structure that adds these variants to my Advantage+ campaign with a ₹5,000 daily test budget."
"Which live creatives are fatiguing — rising frequency and CPM, falling CTR — and should be retired this week?"
"Generate five new primary-text angles for our best-performing video and prepare them for approval."
"Build me a fortnightly creative-refresh plan for this account with ₹30,000 reserved for new variants each cycle."
"Summarise last cycle's variant performance and recommend which three to scale."
What OpenAds prepares vs what you approve
OpenAds handles the analysis — auditing diversity, spotting fatigue and reading variant performance — and the proposal — drafting copy, building the variant matrix and structuring the launch with budgets. Nothing reaches Meta on its own. Every upload, ad-set change, budget setting and status flip is presented for your approval, and only the items you explicitly sign off move to execution, where OpenAds publishes exactly what was approved and records the change. Analysis and proposal are always safe to run; approval is the gate, and execution never happens without it.
What to look at
SignalWhy it mattersWhat good looks like Active creatives per ad setMeasures the diversity Andromeda has to work withFour or more distinct variants per active ad set FrequencyRising frequency signals fatigue and audience saturationStable or gently rising within the flight CTR (link)Early read on whether hooks are resonatingHolding or improving as new variants enter CPMClimbing CPM often reflects tired creativeFlat or falling after a refresh Cost per resultThe efficiency outcome the batch must protectAt or below target after the learning window Share of delivery by variantShows which creatives Andromeda is actually choosingSpread across several variants, not one
Common pitfalls & best practices
Cosmetic variants. Changing a word or a colour does not give Andromeda real signal — vary hooks, formats and angles that genuinely differ.
Refreshing too late. Wait until CPMs spike and you are already fighting fatigue; queue the next batch before the current one tires.
Fragmenting budget. Splitting spend across too many tiny ad sets starves each of the data it needs — consolidate and let the model prioritise.
Skipping the learning window. Judging variants after a day or two invites noisy decisions; give each batch time to exit learning before retiring anything.
Losing traceability. Launch without naming conventions and you cannot tell which matrix cell won — label every variant so results map back cleanly.
The workflow, step by step
Audit current creative diversity. Ask OpenAds to pull live ads by ad set and score them on format, hook, angle and age, flagging where diversity is thin or assets are fatiguing.
Define the variant matrix. Agree the dimensions to vary — hooks, formats (video, static, carousel), value propositions and audiences — so the batch covers meaningfully different combinations rather than cosmetic tweaks.
Draft the creative batch. Have the assistant generate primary text, headlines and descriptions for each cell of the matrix, mapped to the assets you have available.
Structure the launch. Propose how variants attach to ad sets — consolidated for Andromeda's learning or split for controlled testing — with naming conventions that keep reporting clean.
Set budgets and guardrails. Define the incremental spend for the new batch, for example ₹4,000 per day across the test ad sets, plus rules for when to pause underperformers.
Review and approve. Inspect the proposed batch, adjust copy or targeting, and approve the assets and structure you want to go live.
Publish and label. On approval, OpenAds executes the uploads and launch, tagging each variant so you can trace performance back to its matrix cell.
Measure and refresh. After the learning window, review which variants earned delivery, retire the weak ones and queue the next batch to keep supply ahead of fatigue.
Run this workflow through your AI assistant.