Is an attribution window the same thing as an attribution model?

No. The window is the time rule, deciding how far back a platform may look from a conversion to find a qualifying click or impression. The model is the credit rule, deciding how the eligible touchpoints share the conversion between them. Both change your reported numbers, which is exactly why you should never change them in the same week: if you do, you cannot tell which one moved the result.

Does a longer attribution window mean better performance?

No, it means more conversions are credited to the same activity. In the worked example above, moving from a one-day to a seven-day click window took the reported count from 60 to 90 while the spend, the ads and the actual orders stayed identical. Cost per purchase fell from 500 TL to 333.33 TL without a single business change. Only compare window to window, and label every figure with the window that produced it.

How long should I wait before judging a campaign day?

At minimum the full length of your click window, plus a day or two for reporting lag. With a seven-day click window, a Monday you inspect on Wednesday still has five days of its window left to run, and the conversions that arrive later are backdated into Monday's row. This is why ad sets get paused for underperformance that the finished data would not have supported.

Should I turn on a view-through window?

It depends entirely on how you report it. A view window can only increase your credited conversion count, so blending it into the same column you use to calculate cost per acquisition will make every campaign look cheaper than it is. If you want to understand upper-funnel contribution, keep the click-only and click-plus-view figures in separate columns and make budget decisions from the click-only column until you have a reason to trust the other one.

What is an attribution window? Timing guide

Ask what is an attribution window and most answers start talking about credit: which touchpoint deserves the sale. That is a different axis entirely. An attribution window is the time dimension of measurement. It is how far back a platform is allowed to look from the moment a conversion happens, hunting for an ad interaction worth crediting. Leave your campaign, budget and creative untouched, change only that number of days, and the report will hand you a different number of conversions.

What is an attribution window, mechanically

A conversion fires on your site at some timestamp. The platform takes that timestamp and looks backwards. If it finds a click on one of your ads inside the click window, the conversion is credited to that ad. If it finds only an impression, and you have a view window switched on, it may be credited on that basis instead. Nothing is counted forward from the ad. Everything is counted backwards from the sale.

That is why the setting is always written as a pair. A configuration described as seven-day click and one-day view means two separate lookbacks running at the same time: seven days for people who clicked, one day for people who only saw. The two are not interchangeable, and the combined figure should never be read without knowing the split.

Where you set it differs by platform. Google Ads attaches a conversion window to each conversion action, so a lead form and a purchase can carry different windows inside one account. Meta exposes the setting at the ad set level and applies it to reporting and to optimisation together. The available options and the maximums on both platforms have changed more than once, so check the current list in the platform's own official documentation rather than trusting a blog post, including this one.

Which touchpoint gets the credit, once the window has decided what qualifies at all, is a separate question handled by the attribution model. Window first, model second: the window decides what is eligible, the model divides what is left.

The same campaign, two conversion counts

Here is the arithmetic, with numbers chosen to be simple rather than realistic. A campaign spends 30,000 TL in a month. Read with a one-day click window, it reports 60 purchases. Read with a seven-day click window, same spend, same month, same ads, it reports 90.

One-day click: 30,000 divided by 60 gives 500 TL per purchase.
Seven-day click: 30,000 divided by 90 gives 333.33 TL per purchase, exactly a third lower.

Add an average order value of 750 TL and the return figures pull apart the same way. The one-day reading credits 60 times 750, which is 45,000 TL of revenue, so ROAS reads 45,000 divided by 30,000, or 1.5. The seven-day reading credits 90 times 750, which is 67,500 TL, so ROAS reads 67,500 divided by 30,000, or 2.25. The second ROAS is 1.5 times the first, for the simple reason that the conversion count is 1.5 times the first: 90 divided by 60 is 1.5.

Not one extra unit left the warehouse. The bank balance is identical under both readings. The seven-day window credited 30 orders that the one-day window declined to credit, because those buyers came back on day three or day five. Paste either figure into a ROAS calculator without recording which window produced it and you have manufactured a number nobody can reproduce next month. A ROAS or cost-per-acquisition figure with no stated window is not a measurement.

Why last Monday's numbers are still moving on Wednesday

This part catches experienced buyers, because it applies no matter which window you picked. Both Meta and Google date a conversion to the day of the ad interaction, not the day the purchase happened. A sale made today, by someone who clicked five days ago, is written into the row for five days ago.

So a campaign day does not finish when the day finishes. With a seven-day click window, Monday's row can keep collecting conversions until the following Monday. Open the report on Wednesday and five of those seven days have not happened yet: 7 minus 2 leaves 5. Every judgement you make about Monday at that moment rests on a partial number, and a partial number is always low, never high.

Two habits deal with it:

Match your review cadence to your window. On a seven-day click window, do not kill an ad set on three-day data. Wait for the window to close, then allow another day or two for reporting lag.
Compare like maturity with like maturity. Judge last week's settled data against an equally settled week from the previous period, never against the last three days. Fresh data loses every comparison it enters, whatever the campaign actually did.

Automated rules inherit this flaw and inherit it faster than you do. A rule that pauses anything above a cost threshold, or an automated scaling layer that raises budgets on yesterday's efficiency, is reading an open window by design. Build the maturity delay into the rule itself rather than into your good intentions.

The window is also the optimiser's training set

Reporting is the visible half. The same window also defines which conversions are handed back to the bidding system as examples of success.

Put a one-day click window on a product people research for a week and the algorithm only ever meets the impulse buyers. It will faithfully learn to find more of them, and the slower, often higher-value buyer becomes invisible to it, not because that person fails to convert but because they convert outside the frame. Widen the window and you gain signal volume, at the cost of crediting interactions whose real influence is more arguable.

View windows sit at the far end of that trade. Switching one on will always raise your credited conversion count, because it can only add rows, never remove them. Whether those people bought because of the impression, or would have bought regardless, is a causality question that attribution is structurally unable to answer. Report the click-only figure and the click-plus-view figure beside each other and treat the gap as an open question rather than a result.

Window length is also a policy variable outside your control. Meta changed its default attribution setting in the wake of Apple's App Tracking Transparency rollout, the fallout of which is covered in our note on iOS 14 and Facebook ads. The short version worth carrying here: a shortened window is a measurement event, not a performance event, and the dip it puts in your dashboard is not a dip in sales.

Choosing a window from your own lag data

There is no correct window, only one that matches your buying cycle. Public claims about typical lag are no substitute for your own distribution, which you already own.

Pull the lag distribution, not the average. Both platforms expose days-to-conversion style breakdowns inside their attribution reporting, and the report names move around, so locate the current one in the official documentation. The average is useless here because the distribution is lopsided.
Find the point where it flattens. If additional conversions per extra day of lag have collapsed to almost nothing by day three, a thirty-day window buys you noise plus a much longer wait for clean data.
Segment by product, not by account. A 200 TL consumable and a 20,000 TL sofa do not share a decision cycle. In Google Ads you can give them separate conversion actions. On Meta you can separate them by ad set.
Write the choice down and hold it. One window, one currency, one definition, for a full measurement cycle. Changing it mid-quarter destroys the comparison you were trying to make.

When you do change it, treat the change as a break in the series. Mark the date, keep old figures labelled with the old window, and never average across the boundary. The temptation will be to read the jump as performance. It is not performance, it is a new ruler.

What the setting can and cannot tell you

The window governs eligibility for credit and nothing beyond that. It cannot tell you whether the ad caused the purchase, it cannot reconcile your platform total with your accounting total, and it cannot make a low-volume account statistically comfortable. What it can do, once you have set it deliberately and documented it, is make your numbers reproducible: two people pulling the same report, with the same window and the same date maturity, should land on the same figure. That is a lower bar than truth, and it is the bar most accounts are quietly failing.