Raw ROAS figures from retail media platforms are not comparable. Amazon's 14-day last-click window, Walmart's 30-day view-inclusive window, and Criteo's first-click model each produce a different number for identical underlying consumer behavior. Normalized ROAS applies a correction factor to each platform's reported figure to arrive at what ROAS would have been under a single common measurement standard.
The Three Sources of Cross-Platform ROAS Distortion
1. Attribution window length (W)
A longer attribution window captures more conversions simply by extending the time period in which a sale can be credited to an ad. A consumer who clicked on Day 1 and purchased on Day 20 is counted under a 30-day window but not under a 14-day window. Longer windows always inflate ROAS relative to shorter windows for identical campaign performance.
2. View-through attribution (VT)
Some platforms credit a conversion to an ad impression even when the consumer never clicked — only viewed it. View-through attribution inflates ROAS by claiming credit for purchases that may have occurred entirely independently of the ad. Platforms that include it systematically report higher ROAS than platforms that count only click-driven conversions.
3. Attribution model type (M)
First-click attribution gives all credit to the first ad interaction; last-click gives all credit to the final interaction before purchase. For multi-touch journeys these models produce very different per-campaign ROAS figures even when total revenue is identical.
Platform Normalization Adjustments
| Platform | Default Window | View-Through | Model | Typical ROAS Inflation |
|---|---|---|---|---|
| Amazon Ads | 14-day | No | Last-click | Baseline (0%) |
| Walmart Connect | 30-day | Yes (14-day) | Last-click | +15–35% |
| Criteo | 30-day | Yes (1-day) | First-click | +15–40% |
| Instacart Ads | 14-day | Yes (1-day) | Last-click | +5–15% |
| Target Roundel | 14-day | No (SP) | Last-click | +0–10% |
| Kroger Precision | 14–30-day | Varies | Last-click | +10–25% |
An agency reporting Walmart at 6.2x and Amazon at 4.8x might conclude Walmart outperforms. After normalization, Walmart's figure reduces to approximately 4.5x — making Amazon the stronger performer. The allocation recommendation flips entirely.
Why 14-Day Last-Click Is the Standard Baseline
The 14-day last-click, no view-through baseline is Amazon's default methodology for Sponsored Products — the largest and most liquid retail media channel. Using it as the baseline means all corrections go downward, which is methodologically cleaner. It also reflects a reasonable approximation of purchase cycles for most CPG categories.
Normalization vs. Incrementality
Normalized ROAS corrects for measurement methodology differences. It does not measure whether advertising caused the purchase. A normalized 4x ROAS still includes organic purchases that would have occurred without the ad. Incrementality testing is the separate process for quantifying true causal impact. Normalize first to enable fair cross-platform comparison; apply incrementality analysis to validate true ROI.
FAQ
Generally yes — platforms with more generous attribution see their ROAS reduced. Amazon Sponsored Products, as the baseline, requires minimal adjustment. The goal isn't to make platforms look worse; it's to make them comparable. A platform that genuinely outperforms will still show that after normalization.
No. Platforms do not publish the revenue decay curves or view-through conversion rates needed for precise corrections. The factors used in normalization are estimated from industry research and third-party measurement studies — which is why normalization requires a maintained, versioned methodology rather than a simple formula.
No. Blended ROAS is total attributed revenue divided by total spend — a simple aggregate that doesn't correct for methodology differences. Normalized ROAS corrects each platform individually. You can calculate a blended normalized ROAS by aggregating individual normalized figures, but raw blended ROAS remains subject to the same distortions as individual platform ROAS.
Correction factors are estimates with uncertainty ranges, not exact figures. A typical confidence interval is ±5–8 percentage points. Decisions based on small normalized differences (3.1x vs. 3.4x) should be treated with appropriate uncertainty. Large differences (5x vs. 3x after normalization) are robust to correction uncertainty.
RetailNorm applies the W×VT×M correction model automatically. Each platform's ROAS is adjusted to the 14-day last-click baseline using versioned correction factors. Output includes a normalized ROAS figure per platform with confidence scoring based on data quality and correction uncertainty.