Glossary · Attribution Measurement
Lookback Window

Attribution Window

Definition

The period of time after an ad interaction — click or impression — during which a resulting purchase is credited to that ad. Sales within the window are attributed; sales outside it are not. Every major retail media platform sets its own window, making cross-platform ROAS comparisons misleading without normalization.

How It Works
Consumer clicks ad on Day 0

Consumer purchases on Day N

If N ≤ window length → sale credited to ad
If N > window length → sale NOT credited

Attribution windows determine which sales get counted in a platform's ROAS calculation. A consumer who clicks an Amazon ad on Monday and purchases the following Thursday (Day 3) is counted regardless of window length. A consumer who clicks and purchases 25 days later is counted by Walmart (30-day window) but not by Amazon (14-day window) — even though the ad interaction was identical.

This creates a systematic, structural difference in reported ROAS between platforms that has nothing to do with actual campaign performance.

Attribution Windows by Platform

PlatformClick WindowView WindowNotes
Amazon Sponsored Products14 daysNoneIndustry reference baseline
Amazon Sponsored Brands14 daysNoneSame as SP
Amazon DSP14 days14 daysView-through included
Walmart Connect30 days14 daysBoth click + view included by default
Criteo30 days1 dayFirst-click model; view-through included
Instacart Ads14 days1 dayView-through included
Target Roundel (SP)14 daysNoneClose to Amazon baseline
Kroger Precision Mktg14–30 daysVariesConfigurable by campaign type
Tesco Media28 days1 dayUK market
Carrefour Links30 days1 dayEU market

How Window Length Inflates ROAS

The inflation effect is driven by the shape of the purchase distribution after an ad click. In most CPG categories, the majority of post-click purchases happen within the first 3–7 days. The tail of the distribution — purchases on Day 8 through Day 30 — accounts for a smaller fraction of total attributed sales but still adds meaningful volume to a 30-day window that a 14-day window would miss.

Empirical estimates from third-party measurement studies suggest that for typical household goods and grocery categories, moving from a 14-day to a 30-day window inflates attributed revenue by approximately 12–22%, which translates directly to ROAS inflation of the same magnitude.

Worked example

An agency runs identical campaigns on Amazon and Walmart with $50,000 spend each. Amazon reports 4.2x ROAS ($210,000 attributed revenue). Walmart reports 5.8x ROAS ($290,000). The planner concludes Walmart significantly outperforms. After normalizing both to a 14-day last-click baseline, Walmart's figure reduces to approximately 4.0x — virtually identical to Amazon. The apparent performance gap was entirely a measurement artifact.

Window Type: Click vs. View-Through

The window length is only one dimension. The type of interaction that starts the window matters equally. A click window only begins timing when a consumer actively clicks an ad. A view-through window begins when a consumer merely sees an ad impression — even without clicking. Platforms that include view-through attribution claim credit for purchases that may have been entirely organic, inflating ROAS independently of the window length effect. See: View-Through Conversion.

Why Platforms Don't Standardize

Each platform has a structural incentive to use the most favorable attribution settings: longer windows and view-through inclusion produce higher reported ROAS, which makes the platform look more valuable to advertisers. Industry efforts to standardize attribution have consistently stalled because no major retailer wants to voluntarily adopt a methodology that reduces their reported performance metrics. The result is a permanently fragmented measurement landscape that agencies must navigate through normalization.

FAQ

Can I change the attribution window on retail media platforms?

Some platforms allow custom window configuration, but the default is what appears in standard reporting. Amazon allows some window customization at the campaign level for DSP. Walmart and Criteo offer limited configuration. In practice, most agency reporting uses platform defaults, which is why normalization is necessary for cross-platform comparison rather than relying on custom configurations.

Is a longer attribution window always better for advertisers?

No — it depends on what you're optimizing for. A longer window captures more attributed revenue and produces higher ROAS figures, which looks good in reporting. But it may credit the ad with purchases that would have occurred organically, overstating the campaign's true impact. For budget allocation decisions, a more conservative window combined with incrementality testing gives a more accurate picture of true ROI.

What's the difference between an attribution window and a conversion window?

The terms are used interchangeably in retail media. Attribution window, conversion window, and lookback window all refer to the same concept: the time period after an ad interaction during which a resulting purchase is credited to the ad. Different platforms use different terminology for the same setting.

How does attribution window length affect budget allocation decisions?

Without normalization, longer-window platforms appear to have higher ROAS, which causes budget to shift toward them — even when true performance is equivalent or worse. This systematic bias compounds over time: more budget flows to the platform with more generous attribution, which reports even higher ROAS (on a larger base), which attracts more budget. Normalization breaks this cycle by correcting for window length before using ROAS as an allocation signal.

RetailNorm corrects for attribution window differences as the first step in normalization. The window correction factor (W) adjusts each platform's reported ROAS to a 14-day equivalent using revenue decay curves calibrated by category. The result is a normalized ROAS figure that reflects performance differences, not window differences.

Normalize your platform data →