Why per-ad and per-asset comparisons mislead on Meta (historical ROAS doesn't predict future delivery; low-volume slices; day-1 delivery ≠ day-8), so judge ad sets or CBO campaigns in aggregate, launch new ads in batches of uniquely different angles every ~2 weeks, and don't chase the single winning headline or image.

The breakdown effect: judge in aggregate

TL;DR

Breakdown effect. Meta allocates on predicted marginal performance. An ad’s past average ROAS doesn’t predict what more spend on it would return. Per CTC, Meta’s documentation says the link is “not deterministic”.¹
Judge the ad set, or the campaign if CBO is on, not individual ads, headlines or images. A successful ad set has “lots of winning combinations” (Jon Loomer).²³
Don’t copy the “winning” text or creative into new ads. Look at which primary texts, headlines and formats drove conversions only to learn.⁴
Delivery on day 1 isn’t delivery on day 8. Don’t react to which ad Meta favours day to day.⁵
Next move when aggregate results disappoint: a new batch with a different psychological angle: pain point, solution or offer. Targeting and settings aren’t the lever.⁶

Details

Loomer’s batch workflow:

Launch 1–2 ads (one image, one video, 5 texts and 5 headlines each). Use the creative testing tool only if you need a true A/B test.
After a week, plus a week after any test, ask how the ad set performs. If it’s good, do nothing.
If it’s not good enough, publish a uniquely different batch into the same ad set. Review again after ~2 weeks.
With 10–15 ads and still poor aggregate results, try push delivery on a neglected ad, or turn off a dominant poor performer (both for peace of mind). Then look hard at the landing page: clear, few options, visible price, nothing mysterious.⁷⁸⁹

Why not hunt winners. Low-volume slices mislead; that’s the breakdown effect in Loomer’s sense.⁵ Ben Heath adds that personalised delivery means the ad with the best average result isn’t the best for each person. Split-testing your way to one perfect ad loses the benefit of a diverse set.¹⁰

Throttling test (Dara Denney). New advertisers ask why Meta doesn’t spend more on the ads with a cheaper CPA. Dara has tested forcing more spend onto those ads several times, and they usually performed worse. Meta had predicted that correctly. Her advice is to analyse what the dominant ad does creatively, not to rebalance spend.¹¹

CTC’s version. Since the ranking model uses signals you can’t see, the operator’s job is to trust allocation within cost controls and stop making ad-level decisions from backward-looking dashboards.¹²

Disagreements / caveats

Many practitioners still judge individual ads and concepts: Dara Denney’s per-concept ABO tests, Zach Stuck’s $2K spend-velocity signal (Scaling a Meta ad from signal to winner), creative analytics in Diagnosing why a Meta ad isn't working (metric-pattern cheatsheet). The consensus is narrower: don’t make kill or keep decisions on thin per-ad numbers. Per-ad diagnostics still guide what to make next.
Aggregate judgement assumes the ad set’s goal and attribution are right. Combine it with Meta's incremental attribution model — should you switch from standard?.

The breakdown effect — judge Meta results at the ad-set level, not per ad

The breakdown effect: judge in aggregate

TL;DR

Details

Disagreements / caveats

See also

The breakdown effect — judge Meta results at the ad-set level, not per ad

The breakdown effect: judge in aggregate

TL;DR

Details

Disagreements / caveats

See also

Footnotes