Previous article (part 3): Attribution modeling terminology
Attribution modeling involves analyzing the influence of a series of interactions on conversions in the period preceding a conversion.
How long the period subject to analysis should be is decided by the set conversion window, called in some reports the lookback window (conversion window, lookback window).
In the Google Analytics terminology introduced in 2024, the term “conversion” is reserved for conversions exported to and shared with Google Ads, while other important events are called key events. In this context, the more universal term will be lookback window.
Some illustrations in this article from before 2024 may point to key events as conversions.
The lookback window is the period of time preceding a conversion (key event) in which the clicks and impressions potentially affecting that event’s occurrence are taken into account.
An example online-store customer had the following interactions with advertising:
- on February 1 they clicked a Google Ads search ad and reviewed the offer;
- on February 20 they clicked a Google Ads ad again and considered buying;
- on March 1 they clicked a remarketing ad on Facebook, but didn’t complete a transaction.
On March 10, this user went to the site directly and made a purchase.
With a 30-day lookback window, only the second Google Ads click and the Facebook remarketing ad will be included in the analysis. If we extend the lookback window to 45 days, then the first Google search ad click will also be able to receive a share in attribution.
The phrase “will be able to” receive a share is not accidental. Whether a given click ultimately receives a share in attribution depends on the attribution model used.
Nevertheless, regardless of the model used, interactions outside the lookback window are not taken into account.
More information about lookback windows can be found in the Google Ads help article and in the article on attribution in Google Analytics.
Changing the lookback window
The lookback window is changed in the reporting system’s settings or in a given report’s parameters. Usually you’ll have several specific periods in the range of 1–90 days available, or the option to choose your own period in that range. If we don’t apply a chosen lookback window in a report, the system’s default settings are usually used.
When changing the lookback window in the system settings, it’s worth noting whether the changes will work retroactively, i.e. whether they’ll also apply to earlier reports. Such changes can be made without worry, because they’re reversible. This is the case when, for example, in a Facebook ads report we choose a different attribution method (in general, report parameters shouldn’t permanently affect the system).
In some systems, changes to the lookback window will only take effect from the moment they’re made (they won’t be retroactive). Such changes should be made carefully. Making them means reports before and after the change won’t be fully comparable, and the changes in the data will be irreversible. Even if we restore the earlier setting, it won’t apply to data from the period in which the lookback window was different (see also the article on attribution in Google Analytics).
Lookback windows in attribution model comparisons
This section (updated in January 2024) refers to solutions used in Universal Analytics (the previous version of Google Analytics). Attribution still works in an analogous way today in Campaign Manager (the ad-tracking tool of the Google Marketing Platform).
In the attribution model settings there’s an option to select a lookback window other than the default:
If a model doesn’t have its own lookback window set (the Lookback window option is turned off), then that model will use the default lookback window for the report. In attribution model comparison reports, an individually defined lookback window will be marked under the model name, and it will be applied in that model even if a different lookback window is defined in the report.
In the example below, the Last interaction and Last AdWords click (the former name for Google Ads) models don’t have their own lookback window. The Last AdWords click II model has its own lookback window set to 30 days. In the report, the lookback window is set to 90 days, so it will be applied in the Last interaction and Last AdWords click models, while the Last AdWords click II model will analyze the influence of interactions based on its own lookback window, i.e. over 30 days:
It’s worth remembering that for a report we also choose which conversions (key events) are to be analyzed. These can be both e-commerce transactions and completions of other goals — it’s worth making sure we’re really analyzing the events we mean to:
What information is tied to the lookback window?
The lookback window should be tailored individually to the length of the attribution path and user behavior. For products bought on impulse or where the decision time is very short (e.g. calling roadside assistance), you should rather use shorter lookback windows.
For products with a long decision process (e.g. buying real estate), it’s worth using longer lookback windows, to include as many past interactions as possible.
The ability to change the lookback window will come in handy if we want to reconcile reporting between systems with different lookback windows — it’s worth bringing them to a common denominator at least in this dimension.
Comparing data from different tracking systems (e.g. Google Ads and Analytics), advertisers often see significant discrepancies — one of the causes can be a different lookback window. For example, the standard lookback window in the Google Ads conversion tracking system might be 30 days, while in Analytics or an affiliate program’s reporting system a 90-day lookback window might be used.
The lookback window won’t affect the results shown by the last-interaction model — that will always stay the same. In turn, for models that assign value to the first interaction, this influence will be significant, and in fact, when we talk about the first interaction, it’s always worth establishing “within what lookback window”.
We’ll also see the effect of changing the lookback window with multi-touch models (which include more than one interaction) and with modified last-interaction models, e.g. those that assign value to the last ad click from a given source (such as last Google Ads click).
Comparing attribution in the same model but for different lookback windows, we can see how a given source’s influence on attribution changes over time.
Remarketing as the first interaction?
In some reports done with the first-interaction model, part of the value may be assigned to remarketing campaigns. But how is it possible for a remarketing campaign to be the first interaction on the path? Remarketing targets returning users, so a first visit from remarketing is obviously impossible. So where does such data come from?
The fact that we see transactions for which remarketing is the first interaction can result from the length of the lookback window. If it’s 90 days, and the original visit took place, say, 95 days before the transaction, after which, say, 15 days before the transaction there was a remarketing visit, then that visit will be considered the first interaction in this model.
You also have to remember that remarketing targeting can use signals about logged-in users that the tracking system won’t take into account. For example, a Facebook user may first encounter the site on a smartphone, then click a remarketing ad during a visit to Facebook on a computer (where Facebook recognizes them thanks to being logged in), which in Google Analytics or another system may be read as a first visit, if Google’s systems don’t know that the same user is using both devices.
Next article (part 5): Direct visits and non-direct models
Worth reading: A guide to attribution in Google Analytics
