Andromeda reads creative variety itself as a targeting signal, and the practical consequence is that the number and diversity of assets inside an ad set affects delivery performance independent of each asset's individual quality. The lift from running 5 to 15 distinct assets per ad set is real — Advize sees 10-25 percent CPA improvements in client accounts crossing from under 5 to the 10-15 range Meta recommends. Above 15 assets, the lift flattens and often reverses. The reason is that composition matters more than count: 15 assets covering three distinct hook angles and five visual formats produces better signal than 15 near-duplicates of the same winning ad.
Creative variety is a targeting signal, not just a quality signal. Moving from under 5 to 10-15 assets lifts CPA 10-25 percent. Above 15, lift flattens or reverses. Composition — 3+ hook angles across 4+ visual formats — matters more than raw count.
What does it mean that Andromeda reads creative variety as a targeting signal?
Andromeda uses the composition of an ad set's creative library to infer the breadth of audience the ad set is attempting to reach. A library of fifteen assets covering three hook angles (price, outcome, status) across five visual formats (UGC, studio, animation, lifestyle, product-focus) signals an intent to reach multiple sub-segments with different motivations. The system then tests each asset against overlapping sub-audiences and lets the auction discover which asset-audience pairings work. A library of three near-duplicate assets signals narrow targeting intent, and the system optimises for a smaller set of winning audiences faster but caps the total addressable pool. This is why Meta recommends 10-15 distinct assets per Advantage+ campaign — the recommendation is not about avoiding fatigue, it is about giving the targeting signal enough shape to work with.
How much CPA improvement does moving from 5 to 15 assets actually produce?
The CPA improvement from moving from under 5 to the 10-15 asset range typically falls between 10 and 25 percent in Advize's experience across DTC accounts. The variance depends on starting conditions: accounts that were running 2-3 near-identical assets see the largest lift (often 20-30 percent) because the baseline was starving the targeting system of signal entirely. Accounts already running 5-6 assets with reasonable diversity see smaller lifts (8-12 percent) because the system already had enough composition to work with. The lift is also non-linear across the range — the move from 3 to 8 assets usually produces larger gains than the move from 10 to 15.
Why does the lift flatten above 15 assets?
The lift flattens above 15 assets because Andromeda's exploration budget per ad set is finite. The system allocates a bounded number of impressions to testing each asset against each sub-audience. Adding assets above the 15 mark divides that budget across more assets without expanding the total budget, which means each additional asset gets less exploration volume and generates less reliable signal. Above 20 assets, the exploration dilution starts producing worse CPA than the 10-15 configuration because the system cannot establish clear winners before each asset is burning through impressions on sub-audiences that may not convert. Andromeda penalises reskinned ads precisely because they add asset count without adding variety signal, and the same mechanism causes the above-15 flattening.
Why does composition matter more than count?
Composition matters more than count because Andromeda's variety signal reads angle diversity and format diversity separately from asset diversity. Fifteen assets that all pitch the same hook from different camera angles register as low variety even though the asset count is high. Six assets that pitch three distinct hooks (what the product does, why it is different, who it is for) across four formats (UGC testimonial, product demo, animated explainer, founder-led) register as high variety at a lower count. The practical implication is that an ad set with six well-composed assets often beats one with twelve near-duplicate assets, and asset count is only a useful proxy when the composition is also strong.
What is the right composition structure for a 10-15 asset library?
1) At least three distinct hook angles — pick from: price/offer, outcome/result, status/identity, problem/pain, mechanism/how-it-works. 2) At least four distinct visual formats — pick from: UGC, studio, animation, lifestyle, product-focus, founder-led, split-screen comparison. 3) At least two assets per hook angle so Andromeda can test variations of each angle rather than treating each as a one-off. 4) A healthy mix of static and video (don't stack 15 videos or 15 statics — Advantage+ reads mixed-format libraries as broader intent). 5) At least one asset per format using the brand's hero product so the catalog signal reaches the targeting layer. 6) Refresh one asset per week — the composition-first refresh pattern, not the volume-first one.
How do you audit whether your ad set's variety signal is actually diverse?
Open the ad set and list every active asset. For each asset, tag it with the hook angle it uses and the visual format it occupies. Count distinct hook angles — should be at least 3. Count distinct visual formats — should be at least 4. Count assets per hook — if any hook has only one asset, add a variant. Count assets per format — if any format has more than 5, consider pruning. If the diversity score is below 3 hooks × 4 formats, the ad set is likely under-signalling to Andromeda regardless of total asset count, and the fix is composition expansion rather than asset addition. This audit takes 15 minutes and catches most under-performing variety configurations before the CPA data makes the issue obvious.
Conclusion
Creative variety is a targeting signal in a specific sense: Andromeda uses the shape of the asset library to infer who the ad set is trying to reach, and a diverse library produces broader targeting signal than a narrow one. The practical consequence is that moving from under 5 to 10-15 assets usually produces real performance gains. Moving above 15 almost never does. The next step for any ad set below the floor is to add distinct hook angles and visual formats until the library covers at least three angles across five formats, then stop.