A window you chose after seeing the results is not a measurement

Every attribution report rests on a number almost nobody argues about. Not the model, not the touchpoints, not the data source — the window. The number of days after an advert during which a sale still counts as caused by it.

It is usually inherited. Somebody accepted a platform default in the first week, the setting was never revisited, and three years of budget decisions have been made on top of it. Occasionally it is worse than inherited: it gets adjusted, quietly, until the channel somebody likes stops looking bad.

The attribution window decides your channel ranking before a single conversion is collected. It is the most consequential setting in measurement and the least defended.

What an attribution window actually is

An attribution window is the period after an advertising interaction during which a conversion is still credited to it. Inside the window, the sale belongs to the channel. Outside it, the same sale belongs to nothing.

That is a rule, not an observation. Nothing about a purchase changes at midnight on day thirty. The customer who bought on day thirty-one was influenced by exactly the same things as the one who bought on day twenty-nine; the only difference is which side of an arbitrary line they landed on.

This matters because the line is not neutral between channels. A short window systematically favours channels that catch people at the end of a decision — branded search, retargeting, a map listing somebody found while standing in the rain. A long window favours channels that reach people at the start of one. Neither is more honest than the other. But whichever you pick has already chosen the winner before the data arrives.

• Click windows count somebody who clicked and later converted.
• View-through windows count somebody who saw an advert, did not click, and converted anyway. These are shorter, and for good reason: the evidence is weaker.
• Engagement windows, on some platforms, count an interaction short of a click — a video watched, a form opened.
• Each is a different claim about causation, and pooling them into one 'conversions' column hides which claim is being made.

Platform defaults are not recommendations

Every advertising platform publishes its own window options and defaults, and they differ from each other. A default is a reasonable starting point for that platform's inventory, not a statement about how long your customers take to buy.

Two examples make the spread obvious, and both are published by the platforms themselves rather than inferred.

One platform tops out at 28 days. Another will go to 90. If you run both and read each one's self-reported conversions, you are comparing a channel measured over four weeks with a channel measured over three months, and then deciding which one to fund. That comparison is not close to fair, and nothing in either interface warns you.

Comparing two platforms' reported conversions is comparing two different rules applied to two different populations. The gap between them is not performance.

The window should come from your sales cycle, not from the interface

The right window is roughly the time it actually takes your customers to go from first contact to money received. That is a number you can measure in your own records, and almost nobody has.

The method is unglamorous and takes an afternoon. Export a year of closed sales with the date each one was paid. Export the leads that produced them with the date each arrived. Match the two on a phone number or an email address, and plot the interval.

What comes back is a distribution, not a number, and its shape tells you more than its average. A trade business usually sees a sharp spike in the first week and a long thin tail of planned work. A remodeler sees a hump at three to six months. A personal injury firm sees almost nothing for a year and then a slow rise.

A useful starting rule: take the median lag and roughly double it. That captures the bulk of the distribution without stretching so far that you are sweeping in customers who had forgotten the advert existed. Then check the tail you excluded and decide whether you are comfortable calling it unattributable, because that is exactly what you have just done.

When a different window per channel is honest

Channels reach people at different stages of the same decision, so a single window applied to all of them is not neutral — it is a window tuned for whichever channel happens to sit closest to the purchase. Per-channel windows are legitimate, under one condition.

Consider two sources for the same bathroom remodel. A search advert for an emergency leak reaches somebody who will buy this week. A visual planning platform reaches the same household nine months earlier, when they are collecting ideas and have no intention of buying anything yet. Judged over thirty days, the second produces almost nothing — not because it failed, but because the rule was borrowed from a channel with a different job.

So varying the window by channel is defensible. What makes it defensible is the sequence, and the sequence is the whole of this article.

Set the window before you look at the results, write it into the report, and apply it to every file. A window chosen after seeing which length flatters a channel is not a measurement of that channel — it is a description of what you wanted.

This is not a new idea, and it is not ours. It is pre-registration, borrowed from clinical research, where the reason is blunter: a study that decides its endpoint after seeing the data can produce a positive result from noise, reliably, every time. Marketing measurement has the same failure mode and almost none of the discipline.

1. Decide the window for each channel, in advance, from the lag in your own records.
2. Record it in the report itself, next to the number it produced, so a reader can see the rule that made it.
3. Apply it to every source in the same file, including the ones you expect to do badly.
4. If you change it, change it for everything, say why, and re-run the old periods so the series stays comparable.

Longer is not better, and the temptation runs one way

Extending a window always increases the number of attributed conversions, because it can only ever add sales and never remove them. That makes it the easiest number in marketing to improve without improving anything.

A ninety-day window will report more conversions than a thirty-day one for every channel you own, in every month, forever. Some of that increase is real work the short window was cutting off. Some of it is customers who would have bought regardless and happened to pass an advert on the way.

Nothing inside the platform distinguishes those two. The count goes up either way, and the person who extended the window gets to report an improvement they did not cause. This is why the direction of the temptation matters: nobody ever shortens a window to make a channel look better, so drift is always toward generosity.

The honest test is whether the sales the longer window added plausibly trace to the interaction — and, crucially, whether you would have accepted that argument before you saw how many there were.

A cheap check that does not require an experiment

Plot the added conversions by how many days after the interaction they arrived. A channel doing real long-cycle work produces a hump — a cluster of sales around the typical consideration period, which is what you would expect if the advert started something that took time.

A channel whose extra conversions are spread flat across the whole extension is a different story. Sales arriving at a steady trickle for ninety days look like background purchasing that the window happened to catch, not like an effect with a shape. That is not proof either way, and it is a great deal better than extending the window and admiring the total.

What the window cannot fix

A window decides which observed sales get credited. It does nothing about the sales nobody observed, and in most businesses that is the larger problem.

If a customer calls an untracked number, walks in, or is referred by a neighbour, no window length will attribute them, because there is no interaction to attribute from. Extending the window in that situation adds only the customers who happened to be traceable, which quietly shifts the ranking toward whichever channels are easiest to track rather than whichever work best.

• Untracked phone calls are the largest gap in most local businesses, and the cheapest to close.
• Referrals and word of mouth usually produce more revenue than any paid channel and appear in no platform report.
• Offline closes — showroom, site visit, invoice raised later — never reach the pixel that started the clock.
• None of these are window problems, and lengthening the window will not touch them.

Before tuning the window, find out what share of revenue has no traceable source at all. If that bucket is large, the window is the second problem.

Reporting the window is part of reporting the number

A conversion count without its window is an unfinished sentence. The same campaign, the same month and the same data produce materially different figures under different rules, so the rule belongs beside the figure.

This is a small discipline with an outsized effect on whether anybody believes you. A report that states its window invites the right argument — whether the rule is the right one — instead of the wrong one, which is whether the number is made up.

It also protects the work. When a channel's ranking changes six months later, the first question is always whether the method changed. If the window is recorded on both reports, that question takes ten seconds. If it is not, you will spend a week reconstructing what the setting was in March and you will probably get it wrong.

CloseRev matches your closed sales against your lead sources from two exported files, so the window is a decision you make and can see rather than a setting buried in an advertising account. The report records the window used per source, and anything it could not match is reported as Direct / Unknown rather than being absorbed by whichever channel had the longest window.

A short checklist

Most of the value here comes from four decisions, none of which requires new software.

1. Measure your own lag from lead to money before choosing anything. An afternoon with two exports beats any default.
2. Set the window per channel if the channels genuinely reach people at different stages, and set it in advance.
3. Write the window into the report, beside the number, every time.
4. Find out how much revenue has no traceable source before deciding the window is your problem.

None of this makes attribution exact. It makes it defensible, which is a different and more useful property. A number you can explain the rules behind will survive a finance director's questions. A larger number you cannot will not, and it should not.

Questions people actually ask

What is an attribution window?

The period after an advertising interaction during which a sale still counts as caused by it. A conversion that happens inside the window is credited to the channel; the same conversion a day later is credited to nothing, or to whatever the customer touched next. It is a rule you choose, not a fact you observe.

What is a good default attribution window?

There isn't one that works across a business. The window should match how long your purchase actually takes from first contact to money received — days for an emergency trade, months for a remodel, a year or more for a personal injury case. If you must start somewhere, measure the median lag between lead and sale in your own records and use roughly twice that.

Is it cheating to use a different window for different channels?

No, provided you set it before you look at the results and record it in the report. Channels genuinely catch people at different stages: a search advert reaches somebody at the moment of need and a planning-stage channel reaches them months earlier. What is not defensible is choosing the window after seeing which one flatters the channel you wanted to keep.

Why do platforms use different windows from each other?

Because each one is describing its own inventory and its own commercial interest. Windows are published and configurable, they differ by platform, and view-through windows are usually shorter than click windows. Comparing two platforms' self-reported conversions is comparing two different rules applied to two different populations.

Does a longer window always produce more conversions?

It produces more attributed conversions, which is not the same thing. A longer window sweeps in sales that would have happened anyway, so it raises the count and lowers the credibility. The test is whether extending the window adds sales that plausibly trace to the interaction, and that is a judgement you should make before you see how much it adds.

How do I know my window is too short?

Take a year of closed sales, match them back to the leads that produced them, and plot the interval between the two. If a meaningful share of your revenue sits beyond the window you are using, the window is cutting off real results and every channel with a long lag is being underpaid for its work.

Found this useful?

• Share on X
• Share on LinkedIn
• Share on Reddit
• Share on Email