Attribution Windows and the Arguments They Cause

An attribution window is the period after a touchpoint during which a conversion is credited to it. Seven days post-click, thirty days, ninety, with or without view-through. Time-window rules also appear outside attribution; this explanation covers the seven-minute payroll rule and its practical implications.

It is a choice, and it changes every number in the report. Widening the window increases attributed conversions without anything happening in the business, which is why arguments about windows are frequently arguments about whose numbers look better.

What the window actually does

A longer window credits more conversions to the touchpoint. It does not create sales; it reassigns them.

Windows differ by platform, so the same conversion is claimed by several parties with different rules. This is a large part of why platform-reported conversions sum to more than reality. See when two systems disagree.

The window interacts with cookie lifetime, and this is the part most often missed. If the identifier survives seven days and the window is set to thirty, the configuration says thirty and the measurement is seven. The report does not mention the discrepancy. See third-party cookies.

Check what actually survives before arguing about what should be configured.

Choosing one defensibly

Start from the purchase cycle, not from a convention.

Look at your own data: the distribution of time between first touch and purchase, and between last touch and purchase. This is a straightforward query and almost nobody runs it before setting a window.

Set the window to cover the bulk of that distribution — the point at which additional days add few conversions.

If the distribution has a long tail, the window is a judgement about where to cut, and it should be documented as such rather than inherited.

Different products need different windows. A £15 repeat purchase and a £4,000 considered purchase have completely different distributions, and applying one window to both misrepresents at least one.

View-through, and why it deserves scepticism

View-through attribution credits a conversion to an impression that was served without a click.

It always measured "an ad was served" rather than "a person saw an ad", and it was never a strong signal. A one-pixel ad below the fold counts identically to a full-screen one.

With cookie restrictions it has largely stopped working anyway.

If a platform reports view-through conversions, treat them separately from click-through, never summed. A blended number combining the two is not interpretable.

Why the argument is usually about something else

Window disputes rarely resolve on methodology, because the substance is elsewhere.

Someone's performance improves with a longer window. A channel that participates early in the journey benefits from a long window; a channel that closes benefits from a short one. Each will advocate for the window that flatters it, with sincere methodological arguments.

Nobody agreed the objective. If the question is "which channel should get more budget", the window is the wrong lever. The answer needs an experiment, and the window argument is a proxy for a question attribution cannot answer. See what attribution actually measures.

The resolution is usually to stop arguing about the window and run a holdout on the disputed channel. That converts an unwinnable methodological debate into a two-week test with a number at the end. See incrementality.

Practical rules

Pick one window per business question and hold it. Changing it changes every historical comparison, and a chart with a window change and no annotation is misleading.

Document the window with the metric definition. See documenting metric definitions.

If you change it, mark the date on every affected chart, and decide explicitly whether to restate history.

Never compare across platforms with different windows. It is not the same measurement, and a table placing them side by side implies it is.

Report click-through and view-through separately, always.

Check the cookie lifetime against the configured window once, and note the discrepancy where it exists.

What to do when someone wants it widened

The request usually arrives as a methodological suggestion and is usually a performance problem.

Ask what decision the change would inform. If the answer is that the channel's numbers look low, the window is not the issue.

Offer to run the distribution. Showing the actual time-to-conversion curve moves the conversation from opinion to data, and it frequently settles it — either the tail justifies a longer window or it plainly does not.

Offer the test instead. "A longer window will attribute more conversions to this channel; it will not tell us whether the channel caused them. A two-week holdout will."

And if the window genuinely should change, change it deliberately, document it, annotate the charts, and restate or note the discontinuity.

The summary

The window reassigns conversions; it does not create them. Widening it improves a channel's numbers without anything happening.

Set it from your own time-to-conversion distribution, not from a default.

Check what the cookie actually survives — the configured window may be fiction.

Report view-through separately, never blended, and never compare across platforms with different rules.

And when the argument recurs, it is usually a proxy for a causal question. Run the holdout instead. A platform example is provided in the Google Ads conversion-window documentation.