What does attribution window mean?

It's the period in which the platform credits a sale to the ad, such as one-day click, 7-day click, or a view without a click. The wider the window, the more the number gets inflated by sales that didn't all come from the ad.

Which window is closest to the truth?

The narrow window (one-day click) is closest to the ad's direct performance, because views without a click inflate the picture. The final judge is always the money that actually reached the store, from the backend.

Why is the ROAS in Ads Manager higher than the real one?

Because the platform credits itself with sales that might have happened anyway, or where the customer saw more than one channel. When you compare the Ads Manager number with the backend number and find a big gap, the backend is right.

The ROAS in Ads Manager isn't always the real one: the attribution window problem

There's a question I ask myself before I get excited about any ROAS: how was this number calculated? The exact same campaign can give you two completely different numbers depending on the attribution window, and most bad decisions get made because someone compares a number on one window with a number on another without noticing.

Same campaign, different numbers

The platform credits a sale to the ad based on a time window you choose: one-day click, 7-day click, or even just a view with no click. Each window gives you a bigger number than the one before it. A wide window credits the ad with sales it wasn't the only reason for, so ROAS looks better. Performance didn't improve. The lens just got wider.

What an attribution window actually means

The platform is saying: "Anyone who saw or clicked the ad during this period and then bought, I brought them." The problem is that a lot of those people were going to buy anyway, or saw the ad alongside other things. The wider the window, the more the number gets inflated by sales that didn't all come from the ad.

The problem: comparing apples to oranges

When you compare a campaign on a 7-day window with a campaign on a one-day window, or a platform that counts views with a platform that counts clicks only, you aren't comparing performance. You're comparing two ways of calculating. And the decision you make from that (who to stop, who to fund) is built on an illusion.

How I read the right number

I fix the window before any comparison. All campaigns get compared on the same window, otherwise the comparison means nothing.
I look at the narrow window first (one-day click). It's the closest to the ad's real, direct performance, and views without a click inflate the picture.
I reconcile against actual figures. The money that really reached the store (from the backend) is the judge. If Ads Manager says one number and the backend says a much lower one, the backend is right.
I look at conversion time rather than click time, so I know when the sale actually happened.

The real number is the one in your bank account

In the case where we cut spend by 16% and ROAS came out at 3.73, what made the decisions right wasn't the number in Ads Manager. It was daily follow-up and reconciling figures against one fixed source. When you read the right number, you stop what's really losing and spend on what's really winning, instead of what looks good in a wide-window report.

This is part of my work as a Performance Marketer and Media Buyer. Details of the work are here and the results, with their numbers, are in the case studies.