Filtering to only brands with custom incrementality factors (from lift tests, not the 1.2x benchmark), we find the within-brand CPMr → iROAS pattern holds in 71% of brands. But the cross-brand signal strengthens: brands with lower CPMr have significantly higher incrementality factors (ρ = −0.62, p = 0.009).
Platform-attributed ROAS includes revenue that would have occurred without ads. iROAS adjusts for this by applying each brand's incrementality factor. This analysis includes only brands with custom factors derived from lift tests — the default 1.2x benchmark brands are excluded.
Blended iROAS = (new_customer_rev × ACQ_factor + returning_customer_rev × RTN_factor) / spend. Each campaign's revenue is split into new vs. returning customer portions, and the appropriate incrementality factor is applied to each. This gives every campaign a unique effective incrementality factor based on its customer mix — campaigns with more returning revenue are penalized more heavily.
Unlike flat iROAS (ROAS × one factor), blended iROAS creates campaign-level variation in incrementality. A campaign with 95% new customer revenue gets a higher effective factor than one with 80% new customers. This lets us test whether CPMr is correlated with the effective incrementality factor itself — not just with performance.
Each bar is one brand's Spearman ρ between CPMr and iROAS. Cyan = lower CPMr predicts better iROAS. Coral = higher CPMr predicts better iROAS. The factors annotate each brand.
Among brands with custom incrementality factors, higher CPMr campaigns deliver better incremental ROAS in 71% of brands (12 of 17). The factors range from 0.55x to 1.84x — some brands see Meta as under-reporting, others as over-reporting — but the within-brand CPMr-to-performance relationship is directionally consistent regardless of factor magnitude.
Within each brand, campaigns split at median CPMr. Values shown are iROAS (ROAS × factor).
For 8 brands with 16+ campaigns, within-brand CPMr quartiles. Values are incrementality-adjusted.
| Quartile | Avg CPMr | Blended iROAS | Eff. Factor | Δ iROAS vs Q1 |
|---|---|---|---|---|
| Q1 (Lowest CPMr) | — | 2.07x | — | — |
| Q2 | — | 2.18x | — | +0.11x |
| Q3 | — | 2.00x | — | -0.07x |
| Q4 (Highest CPMr) | — | 2.19x | — | +0.12x |
With only custom-factor brands, the quartile dose-response is flat: Q4 (2.19x) and Q1 (2.07x) are close, and Q3 actually dips below Q1. With 8 brands and smaller sample sizes, the pattern is noisy. The within-brand signal is weaker when restricted to brands with measured incrementality.
A different question: across the portfolio, do brands with lower CPMr tend to have higher or lower incrementality factors? Each dot is one brand — its median campaign CPMr vs. its account-level incrementality factor.
The cross-brand correlation is ρ = −0.62 (p = 0.009) — highly significant. By filtering to only brands with custom incrementality factors, the signal gets stronger. Low-CPMr brands (below $87 median) average a 1.36x factor, while high-CPMr brands average 1.01x. This suggests that brands reaching broad audiences cheaply may be driving more truly incremental conversions at the account level.
These incrementality factors are calibrated at the account level, not the campaign level. We cannot yet determine whether individual high-CPMr campaigns within a brand are more or less incremental than low-CPMr campaigns in the same account. It remains possible that incrementality varies across campaigns within a brand. The next step is campaign-level lift testing — running holdout experiments at the campaign tier to measure whether CPMr predicts incrementality one level deeper than the account.
The within-brand iROAS analysis (Sections 02–04) shows that high-CPMr campaigns deliver better attributed and factor-adjusted ROAS. But the cross-brand signal here raises a question: if low-CPMr brands are more incremental, could the same pattern hold within brands? We don't have the data to answer that yet. The iROAS finding is robust to account-level incrementality adjustment, but campaign-level incrementality testing is the frontier.
Four anonymized examples showing CPMr vs. iROAS. Bubble size = campaign spend. Inc. factor noted for each brand.
Diagnostic correlations including the new incrementality proxy.
CPMr ↔ Frequency (avg ρ = +0.86) is still the dominant within-brand signal. CPMr ↔ blended iROAS (+0.18) is moderate. The cross-brand analysis (Section 05) shows the strongest signal: brands with lower median CPMr have higher incrementality factors (ρ = −0.62, p = 0.009). Campaign-level lift testing is needed to reconcile these signals.
Results under different spend thresholds. iROAS conclusions are identical to ROAS (same ranks).
| Threshold | Brands | ρ < 0 (Lower CPMr Wins) | Win Rate | Conclusion |
|---|---|---|---|---|
| Spend ≥ $500, ≥ 6 campaigns | 23 | 7 | 30% | Higher CPMr wins |
| Spend ≥ $1K, ≥ 8 campaigns | 17 | 5 | 29% | Higher CPMr wins |
| Spend ≥ $2.5K, ≥ 8 campaigns | 15 | 7 | 47% | Higher CPMr wins |
| Spend ≥ $5K, ≥ 8 campaigns | 14 | 6 | 46% | Higher CPMr wins |
What the iROAS analysis reveals about CPMr.
Among 17 brands with custom incrementality factors, 12 show a positive relationship between CPMr and iROAS. Directionally consistent but not statistically significant at this sample size.
Cross-brand ρ(median CPMr, factor) = −0.62 (p = 0.009). The signal gets stronger when filtering to brands with measured incrementality. Low-CPMr brands average 1.36x vs 1.01x for high-CPMr brands.
Custom factors range from 0.55x (Meta over-reports by nearly half) to 1.84x (Meta under-reports). These are real, measured differences — not default benchmarks.
The cross-brand signal is clear: lower CPMr correlates with higher incrementality at the account level. The frontier is testing whether this holds at the campaign level within a brand.
When we restrict to brands with measured incrementality factors (not the 1.2x benchmark), two things happen: (1) the within-brand finding weakens — 71% directionally positive but not statistically significant with n=17, and (2) the cross-brand signal strengthens to ρ = −0.62 (p = 0.009). Brands with lower CPMr are measurably more incremental. This creates an urgent question: does the same pattern hold at the campaign level? Campaign-tier lift testing is the frontier — measuring whether low-CPMr campaigns within a brand drive more truly incremental conversions than high-CPMr campaigns in the same account.