When you split an ad set, each set gets fewer conversion signals than the learning threshold (50 per week). The machine learning stalls and auction costs go up. The finer you slice, the easier it looks to manage. But to the algorithm, you are dividing its learning material and starving each piece.
Does Splitting Ad Sets Work?
In most cases, it doesn’t. Meta’s machine learning operates at the ad set level. If 10 conversions a day are split across five sets, each set is left with only 2. At that level, no set gets out of the learning phase, and the algorithm keeps struggling to decide whom to show ads to. CPA then fluctuates, and operators misread this as a signal to “split the sets further and test.” It is a structure where the wrong prescription keeps repeating.
Why Do Costs Rise Under the VCG Bidding Structure?
Meta’s auction follows the VCG (Vickrey-Clarke-Groves) principle. The price you pay is set by the opportunity cost you impose on other competitors. If several sets target the same audience, they compete for the same impression opportunities. Impressions get blocked or unit costs rise. Viewed the other way, a larger budget does not guarantee a win under this structure. Creative with a high predicted user response lets a small brand beat a large company. The outcome is decided not by the number of sets, but by how precisely the creative hits the customer’s unmet need.
When ROAS Drops, What Should You Check First?
- Upper-funnel signals: If click-through rate and early watch metrics have weakened, suspect creative fatigue or a failed appeal.
- Structure: If sets are over-fragmented, switch to consolidated operation.
- Comparison method: Comparing creatives by 7-day simple ROAS creates an illusion due to differences in budget scale (Simpson’s Paradox).
Frequently Asked Questions
Does splitting Meta ad sets work?
If conversion data is not sufficient, it backfires. Each set lacks enough learning signal, so optimization stalls.
Are Meta automatic targeting ads better?
Generally, yes. Manual targeting lets the operator’s subjective judgment narrow the learning scope. Consolidated automatic targeting widens the data the algorithm can learn from.
Why does Meta ad ROAS drop?
The main causes are an overly fragmented set structure, creative fatigue, and wrong comparison metrics. Check them in order, starting from the upper funnel.
What is APS ROAS?
It is a metric that recalculates creative performance on the same budget basis. If a challenger creative is 1.30x or more of the top creative, it is judged REPLACE. At 1.05x to 1.29x, it is judged WARN.
Soulpapa Marketing first corrects STP, USP, and appeal design before any skill of splitting sets. Then we run creatives on a consolidated structure using APS criteria. With this approach, we have run multiple e-commerce brands and delivered ROAS of 300%. What we do is not ad agency work. We build ad spend into a brand asset. That is Hero Branding.
Original in-depth analysis: Meta Ad Optimization: Why Common Methods Ruin ROAS, Rebutted with the VCG Bidding Structure and APS
Insights from Soulpapa Marketing — Korea’s digital marketing agency.
Original Korean article: https://soulpapa.co.kr/2026/10/06/meta-ad-set-splitting-roas-drop-reason/
