This field record asks: Do ad sets where several ads each hold material spend share achieve lower conversion cost than ad sets dominated by one ad? It is part of a seven-study Darwin IQ Labs series on early signals for newly introduced Meta creatives (Nos. 113–119).

In a 16-week panel of Meta conversion- and value-optimized ad sets from clients sampled of various industries (58 accounts; 202 ad sets; 2627 ad-set-weeks), each extra ad holding ≥10% spend share is associated with -0.9% next-week CPA relative to the account (95% CI -2.9% to +1.0%) — short of a −5% success bar. Spending ads’ age×gender audiences barely differ.

Sample

Dataset EMR-2026-10B: 58 Meta accounts, 202 ad sets, 2627 ad-set-weeks. Clients sampled of various industries.

Method

Breadth is the count of ads with ≥10% of the ad set’s weekly spend (also 5% and 20%). Outcomes are CPA relative to the same account-week. The primary test uses ad-set fixed effects with breadth this week predicting CPA next week, controlling for spend and active-ad count. Holdout is ~30% of accounts. Observational panel — not an RCT.

Findings

No meaningful CPA gain from breadth. Not at the size the theory needs. Within the same ad set, each extra ad holding 10%+ of spend goes with -0.9% CPA vs the account the following week (95% CI -2.9% to +1.0%; holdout accounts -1.4%, CI -5.1% to +2.2%). The bar was 5% lower CPA per extra ad.

Extra ads holding share do not cut next-week CPA enough to matter

Within-ad-set fixed-effects estimate: % CPA change per extra ad at ≥10% share

-5.0% barNext-week (primary)-90%Holdout accounts-140%Same week (no lead)130%

Source: Darwin IQ Labs, dataset EMR-2026-10B · Success bar −5%

Mechanism check: median Jensen–Shannon distance across age×gender among ads holding share is low (~0.11), so spending ads do not look demographically distinct.

What this record does not claim

  • No account list is released with EMR-2026-10B.
  • Breadth is not randomly assigned; fixed effects reduce but do not eliminate confounding.
  • Age×gender is a coarse audience lens.