Someone clicks your link on the 3rd, hesitates, comes back on the 26th, and pays. Did that link sell? The answer doesn’t depend on any truth — it depends on a number of days that someone set once, often without realizing it. Here’s how to choose it.
A touchpoint can't claim a sale indefinitely. There has to be a date after which you consider it irrelevant — otherwise an article read three years ago would keep claiming credit for your orders today.
That duration is called the attribution window. It's the most consequential setting in any measurement setup, and it's also the one you inherit from a tool without ever opening it.
It doesn't decide whether the sale exists. The sale exists regardless — it's collected, it's in your accounting.
It decides who gets credit for it. Inside the window, the click can be tied to the sale, and the placement that produced it appears in the report. Outside the window, the link is forgotten: the sale becomes unattributed revenue, and nobody knows where it came from.
The mechanics are simple. At the moment of the click, an identifier is stored for the visitor. When a conversion happens later, the system checks whether there's a recent attributable click — and "recent" is exactly what the window defines. The full vocabulary is in the lexicon entry on attribution windows.
Our default is thirty days, adjustable from seven to ninety, workspace by workspace. It's not an industry truth — it's a reasonable starting point, and it's worth knowing why.
Thirty days covers nearly all considered purchases without being so long that it credits anything and everything. Under a week, you're only measuring impulse buys. Beyond three months, a click becomes a coincidence: over a quarter, there's always some link somewhere in someone's past.
The two boundaries aren't arbitrary either. Seven days is the minimum below which a weekend is enough to make a journey disappear. Ninety days is the limit beyond which the credit given says more about the length of the window than about the effectiveness of the placement.
| Window too short | Window too long |
|---|---|
| Late sales fall to unattributed | An old click claims a sale it didn't make |
| Discovery channels look useless | Discovery channels look amazing |
| Budget shifts toward bottom of funnel | Budget shifts toward awareness |
| Total unattributed inflates without explanation | Everything looks attributed, and that's false |
Look at the third row. The window setting moves budget. Not indirectly, not marginally: a seven-day window on a product with a six-week buying cycle mechanically erases all the content work, and the next budget decision will be made against it. Nobody in the meeting will know the decision was made by an integer in a form field.
And the error is hard to spot, because it doesn't look like an error: the report is coherent, the totals add up, nothing is broken. It's just that part of the revenue is filed in the wrong place.
The right window isn't an opinion — it follows from your actual buying cycle. Three ways to get there, from simplest to most rigorous.
Ask your customers. For five or ten recent sales: "How did you find us, and how long before you bought?" It's rough, it's free, and it's often enough to choose between thirty and sixty days.
Look at the time between first contact and payment for the sales you already attribute. If most journeys complete in ten days, a thirty-day window is comfortable. If half take more than thirty days, your window is cutting them off silently.
Test the sensitivity. Recalculate the same month with two different windows. If the placement rankings don't move, the setting isn't your problem. If they move a lot, you've just discovered that your report was mostly a function of that number.
Some useful benchmarks, to verify for your situation rather than copy: a low-price impulse purchase is decided in days; a course, service, or subscription is decided in weeks; a service sold to a business is decided in months. The attribution marketing page shows how these journeys read once they've been reconstructed.
Two durations coexist in any measurement setup, and confusing them leads to wrong conclusions.
The attribution window says how long a click has the right to claim a sale. Thirty days with us by default.
The data retention period says how long raw events stay in the database: thirteen months for clicks, while statistical aggregates stay indefinitely because they no longer contain anyone. The field-by-field detail is on security and GDPR, and the article on tracking without surveillance gives the full table.
In other words: a click from eight months ago is still visible in history, but it won't attribute anything anymore. That's intentional.
It decides whether a click can claim a sale. It doesn't decide which click wins when there are several within the window: that's the role of the attribution model — first click, last click, or linear. The two settings combine and often get confused in discussions.
An example to make it concrete — this is an example, not a data point. Three touchpoints, at day 0, day 12, and day 28, a sale at day 30, a thirty-day window: all three are within the window, and the model decides how to split the credit. Shrink the window to fifteen days, and the first touchpoint disappears: the model has only two shares to distribute, and your placement rankings change without your distribution having moved an inch. The article on the three models details what each one answers.
"What attribution window do ad platforms use?" Each has its own, and they differ depending on whether it's a click or just an impression. That's one reason why adding up reports from several platforms always exceeds actual revenue, as explained in the multi-channel ROAS article.
"What happens exactly when the window expires?" The click stays in history but stops being a candidate for attribution. The sale that arrives afterward is recorded without an origin, and it shows up clearly on the unattributed line rather than being distributed at random.
"Does a longer window improve my numbers?" It increases the attributed share, which looks flattering and means nothing. The only real gain is a window that matches your cycle. Everything else is decoration.
"Can you have a different window for different conversion types?" The useful distinction is between a lead and a sale: the first comes quickly, the second can take time. Setting the window to the sale delay covers both, since the longer one contains the shorter one.
"How do you read the window from an external tool?" It's a parameter of the tracking site, exposed like any other on the technical documentation side, and it applies both to the tracking script and to conversions sent via the API.
The full mechanics, from click to collected sale, are on sales tracking. And the vocabulary — window, touchpoint, customer journey, attributed sale — is filed in the lexicon.
Read next
The same revenue, three different answers depending on the attribution model. The method, and what each model refuses to tell you.
Measuring a campaign doesn't require identifying anyone. What a tracked link actually records, and who is responsible for what under GDPR.
The monthly letter
What we published, what we learned building it, and the numbers we can show. No offers, no automated follow-ups.