What is Attribution window?

An attribution window is the period after an ad interaction during which a conversion is still credited to that ad.

Somebody clicks an ad on the first of the month and buys on the twentieth. Under a thirty-day window that purchase is credited to the ad. Under a seven-day window it is not, and the campaign shows one fewer conversion for the same delivery.

Nothing about the advertising differs between those two accounts. The window is a convention, chosen in a settings panel, and it moves every attributed figure the account reports.

That makes it one of the most consequential settings in advertising and one of the least deliberately chosen, because it usually arrives as a default and is inherited as a fact.

Most platforms run two windows at once and they exist for different reasons.

The click window covers conversions following a deliberate interaction. A click is meaningful evidence of influence, so a longer window is defensible — the person chose to engage and it is reasonable to suppose the engagement mattered for some time.

The view window covers conversions following an impression with no interaction. The evidence is much weaker: the person may not have registered the ad at all. Platforms default these shorter, and shorter still is usually the more honest choice.

The combination produces figures that are not comparable between accounts unless both windows are stated. A campaign reported on a long click window and a long view window will show substantially more conversions than the same campaign on short ones, and the difference is entirely definitional.

The window should match how long people actually take to buy, and that is measurable rather than a matter of opinion.

The distribution of time between first contact and purchase exists in the business's own data. In an impulse category most of it is within days. In a considered purchase — a service, a large item, anything involving other people's approval — the tail runs for weeks and a short window will systematically undercount.

Two failures follow from getting it wrong.

Too short undercounts, makes long-consideration channels look worse than they are, and pushes budget toward whatever converts quickly. It also starves an automated bid strategy of the conversions it needs to learn from, because conversions outside the window never reach the model.

Too long overcounts, credits advertising for purchases it had nothing to do with, and flatters upper-funnel activity in a way that cannot be distinguished from it working.

The defensible position is to pick the window that covers the bulk of the real distribution, write down why, and leave it alone.

When the window changes, every historical comparison across the change breaks.

Cost per acquisition falls and return on ad spend rises on a longer window, with no change in delivery. A report drawn across the boundary shows an improvement that did not happen, and somebody will be congratulated for it.

Where an automated bid strategy is running, the effect is larger than reporting. The strategy optimises against the conversions it is told about, so changing the window changes the target it is pursuing and puts the campaign back into a period of unstable delivery while the model rebuilds.

The practice that avoids the mess: change it rarely, change it deliberately, note the date prominently, and do not compare across it.

With the attribution model. A window and a model together define the measurement. Last-click on seven days and last-click on thirty days are different instruments, and comparing figures from two accounts requires both to match.

With bidding. A conversion that falls outside the window does not exist as far as the bidder is concerned, so the window partly determines what the campaign learns to find.

With reporting lag. Conversions arrive throughout the window, so recent days are always incomplete. A campaign judged on yesterday's figures under a thirty-day window is being judged on a fraction of what it will eventually be credited with, and the fraction is larger for fast-converting channels than slow ones — which biases every recent-period comparison toward the former.

That last point deserves emphasis because it drives real decisions. An account reviewing the past week under a long window is systematically underrepresenting whatever takes time to convert, and the campaigns that look worst in a recent-period report are frequently the ones whose conversions simply have not arrived yet.

State the window beside any attributed figure that leaves the advertising team.

Set the click and view windows separately and deliberately, with the view window short.

Derive both from the observed time-to-purchase rather than from a default.

Do not change them without a reason, and never mid-test.

And remember what the window is for. It bounds a convention about credit; it says nothing about cause, which is a different question requiring a different instrument.

The incompleteness of recent data is worth handling rather than tolerating, because it drives weekly decisions.

Under a long window, a conversion credited to an ad may arrive three weeks after the impression. Today's report of last week therefore contains a portion of last week's eventual conversions, and the portion grows for some days afterwards.

Two habits deal with it.

Hold back the reporting period. Judge performance on a period that has had time to mature — several days or more behind the window's length, depending on how the distribution falls — rather than on the most recent days.

Compare like-aged periods. Last week against the week before, both read today, are differently mature. Last week read today against the previous week read a week ago are comparable.

Neither is complicated and both are routinely skipped, which is why so many campaigns are paused in their second week for underperforming against figures that had not finished arriving.

There is a case for deliberately running shorter windows than the platform offers, and it is worth stating because the instinct runs the other way.

A short window credits only the conversions closest to the advertising, which are the ones most likely to have been caused by it. The figures are lower, more conservative and closer to what a controlled test would show. An account that manages to a short window and still clears its ceiling has genuine headroom rather than borrowed credit.

The cost is that slow-converting activity is undercounted and will be underfunded, which is a real loss in businesses with long consideration. Whether the conservatism is worth it depends on which error the business would rather make.