A decision framework for setting attribution windows

A decision framework for setting attribution windows

The question: click windows, view windows, lookback periods — what should they actually be? Most teams inherit platform defaults and never revisit them, which silently shapes every credit decision downstream.

The framework, step by step:

— Measure your real conversion lag first. Pull the distribution of time-from-first-touch to purchase. The 90th percentile of that lag is your floor for the click window — anything shorter truncates legitimate paths.
— Treat view-through windows with suspicion. A view-through conversion (someone who saw but didn't click an impression, then converted) is the weakest causal claim in the stack. Keep view windows short (often 1 day) and report them separately, never blended into click credit.
— Align windows across platforms before comparing channels. A 7-day-click Meta number against a 30-day Google number is apples to oranges.
— Document every window in one place. Undocumented windows are the single most common cause of irreproducible reports.

The nuance: a longer window captures more true paths but also more coincidental exposures — people who'd have bought regardless. Window length trades completeness against correlation-masquerading-as-cause. Only incrementality testing tells you where the real line sits.

Bottom line for practitioners: derive windows from your measured conversion lag, isolate view-through, harmonize across platforms, and document everything. Defaults are a starting guess, not a measurement.