[GA4] Behavioral modeling for consent mode

This article is for website or app owners who use a cookie consent banner, consent widget, or another consent management solution, and who are concerned about missing data from users who opt out. 

When you implement a consent banner for your website or app, Analytics will be missing data for users who decline consent. Behavioral modeling for consent mode uses machine learning to model the behavior of users who decline analytics cookies based on the behavior of similar users who accept analytics cookies. Modeled data allows you to gain useful insights from your Analytics reports while respecting your users’ privacy.  

For example, behavioral modeling estimates data based on user and session metrics, such as daily active users and key events rate, that may be unobservable when identifiers like cookies or user IDs are not fully available. It helps you answer important questions like:

  • How many Daily Active Users do I have?
  • How many new users did I acquire from my last campaign?
  • What is the user journey from landing on my website to actually making a purchase?
  • How many of my site visitors are based in Germany vs. the UK?
  • What is the difference in user behavior between mobile vs. web visitors?
Modeling in Google Analytics 4

Modeled data vs. observed data

When users visit your site and grant consent for Analytics cookies or when they don't opt out of personalization using advertising ID in Android Settings, Analytics associates user behavior with various identifiers to provide continuity in measurement. We refer to this kind of data as observable data because it comes from users who have given Analytics permission to observe their behavior.

When users don't grant consent, events are not associated with a persistent user identifier. For example, if Analytics collects 10 page view events, it can’t observe and report whether that’s 10 users or 1 user. Instead, Analytics applies machine learning to estimate the behavior of those users based on the behavior of similar users who do accept analytics cookies or equivalent app identifiers.

The training data used for modeling is based on the observed user data from the property where modeling is activated. 

Google's behavioral modeling approach

Google's behavioral modeling approach applies the following machine learning best practices.

Check for accuracy and communicate changes

Holdback validation maintains the accuracy of Google’s models. Estimated user data is compared to a portion of observed user data that was held back from model training, and the information is used to tune the models. Google will communicate changes that might have a large impact on your data.

Maintain rigorous reporting

Behavioral modeling is only included when there is high confidence of model quality. For example, if there isn’t enough consented traffic to inform the model, then events triggered by users who decline consent aren't reported. This helps ensure the accuracy of the data.

Customize for your business

Google’s more general modeling algorithm is separately applied to reflect your unique business and customer behavior.

Prerequisites

Because the model is trained on the observed data for your Google Analytics 4 property, your property must have enough data to train the model. To be eligible for behavioral modeling, your property must meet the following criteria:

  • Consent mode is enabled across all pages of your sites and/or all app screens of your apps.
  • Consent mode for web pages must be implemented so that tags are loaded before the consent dialog appears, and Google tags load in all cases, not only if the user consents (advanced implementation).
  • The property collects at least 1,000 events per day with analytics_storage='denied' for at least 7 days.
  • The property has at least 1,000 daily users sending events with analytics_storage='granted' for at least 7 of the previous 28 days. 
    • It may take more than 7 days of meeting the data threshold within those 28 days to train the model successfully; however it's possible that even the additional data won't be sufficient for Analytics to train the model.

Behavioral modeling starts from the date a given property becomes eligible.
 

Note: Meeting all the prerequisites for behavioral modeling listed above doesn't guarantee eligibility, as the quality of the underlying machine learning model also plays a role. We're continually working to improve the model and expand eligibility without sacrificing quality. Properties that are not eligible today due to model quality not being sufficient may become eligible in the future.

In the very rare event that a property no longer meets the prerequisites for behavioral modeling after previously meeting them, estimated data will no longer be available. If the property later meets the prerequisites again, estimated data will be available again. The estimated data will be available only from the date the property became eligible again.

Show or hide modeled data in reports

To see modeled data in your reports, choose the Blended reporting identity. You must be an Administrator to control this setting:

  1. In Admin, under Data display, click Reporting identity.
  2. Select Blended.
  3. Click Save.

To stop seeing modeled data, select another option. The option you choose does not affect data collection or processing. You can switch between the options at any time without making any permanent impact on data. Learn more about Reporting identity.

How behavioral modeling appears in Google Analytics

Analytics seamlessly integrates modeled data and observed data in your reports. When Analytics includes modeled data, you will probably see differences when compared to reports that include only observed data (for example, higher user counts in reports that include modeled data).

Use the data-quality icon (shown below) to see when modeled data is integrated.

The following table summarizes the messages you might see via the icon.

Data-quality icon status Description
Including estimated user data As of [modeling effective date], Analytics is estimating data that's missing due to factors such as cookie consent.
Including estimated user data

As of [modeling effective date], Analytics is estimating all possible data that's missing due to factors like cookie consent.

Including estimated user data

As of [modeling effective date], Analytics is estimating data that's missing due to factors such as cookie consent.

* Estimated data may not yet be available for yesterday.

Excluding estimated user data Your property's reporting identity setting doesn't allow Analytics to estimate data that's missing due to factors such as cookie consent. Unless you use the blended setting, your reports only include data available from users who consented to the use of identifiers.
Estimated user data unavailable The date range selected is prior to when this property became eligible for estimated data.
Estimated user data unavailable This report includes retention data or a segment which includes a sequence. As a result, it doesn't include estimated data.
Estimated user data unavailable Your property doesn't meet the eligibility criteria to use estimated data.

Some pages in the Analytics interface will also display a banner with information about the modeling status.

The following table summarizes the messages you might see via a banner.

Banner message Banner location
Most templates include only data from users who consented to the use of identifiers, except for the free-form and segment-overlap templates, which do include data from estimated users. Explorations home page
If an exploration has a segment with a sequence, it will show only data for users who consented to the use of identifiers. Exploration detail page
This [report/exploration/audience] includes only data from users who consented to the use of identifiers. Exploration detail page
If this segment includes a sequence, it will show only data for users who consented to the use of identifiers. Segment builder

Modeling in explorations

Modeling in path and funnel explorations is applied differently than in reports. In reports, modeling is applied to metrics like users, sessions, and new users counts. However, modeling is not applied to event counts such as page_view, first_visit, and session_start. If users do not grant consent for Analytics to associate a persistent user identifier with them, Analytics can't detect if the events are the action of the same user. That results in a higher number of first_visit and session_start events for those users, because the event is sent each time they load a page.

In the path or funnel exploration on the other hand, modeling is applied to the first_visit and session_start event. If users do not grant consent for Analytics to associate a persistent user identifier with them, Analytics estimates the true number of first_visit and session_start events. Thus, the first_visit and session_start event count is lower in path and funnel explorations than in reports.

Unsupported features

The following features don't support using modeled behavioral data:

  • Audiences
  • User explorer, cohort, and user lifetime explorations
  • Segments with a sequence
  • Retention reports
  • Predictive Metrics
  • Data export, for example, BigQuery export

Was this helpful?

How can we improve it?
Search
Clear search
Close search
Google apps
Main menu