Attribution Window

An attribution window is the period following an advertising interaction during which a subsequent conversion will be credited to that interaction. A thirty-day click window credits a purchase to an advertisement clicked up to thirty days earlier; a one-day view window credits a conversion to an impression seen within the previous day. The setting is a reporting convention rather than a measurement of causation.

Its significance is that it substantially determines reported performance without changing anything real. Extending a window increases the number of conversions attributed to a channel, improving its apparent return, while shortening it does the reverse. Two organizations running identical campaigns with different window settings will report materially different results, and comparisons between platforms with different defaults are not comparisons of performance.

Click and view windows behave very differently and deserve separate treatment. Click-through attribution requires a deliberate action and is therefore reasonably defensible, though it still credits the channel for demand it may not have created. View-through attribution credits an impression that was served, frequently not seen, and followed at some point by a conversion, which is a weak basis for causal inference and is where reported and incremental performance diverge most sharply.

Window length should reflect the actual purchase cycle rather than a platform default. A low-value impulse category resolves within hours, and a thirty-day window there mostly captures coincidence. A considered purchase with weeks of research genuinely involves touchpoints long before conversion, and a short window there systematically undercredits channels operating early in the journey. Examining the distribution of time between first touch and conversion in the organization's own data is the way to choose.

The choice creates predictable distortions in budget allocation, which is the practical consequence. Long windows favour channels that reach people early and channels with high impression volume, since both have more opportunity to be present within the window. Short windows favour channels that intercept people at the point of decision. Neither is neutral, and a mix optimized under one setting will look different under another.

Because settings differ between platforms and each counts its own conversions independently, the sum of platform-reported conversions routinely exceeds actual conversions, sometimes substantially. This double counting is not a defect in any individual platform's reporting but an inevitable consequence of each measuring in isolation, and it is why platform figures cannot simply be added together to describe total performance.

Cross-device and cross-session behaviour complicates the window further. A person who clicks on a phone and converts on a desktop several days later may or may not be recognized as the same individual, depending on whether they were identified in both contexts. Where that link is absent, the conversion falls outside attribution entirely regardless of the window setting, which means window length is only one of several reasons reported figures understate or misplace credit.

Consistency matters for trend analysis, since changing a window changes reported history. Analyses comparing periods across a settings change are comparing different measurement rules rather than different performance, and this is a common source of unexplained step changes in reporting. Where a window must change, restating history under the new rule is what preserves comparability.

Because the window is a convention rather than evidence of causation, decisions of consequence should rest on experimental measurement instead. In practice the settings are standardized and documented by data analytics so that channel comparisons are made on a common basis, campaign execution runs through marketing services, and where allocation decisions are material the attributed figures are validated against holdout testing before budget is moved.