Google uses modeling to estimate online key events that can’t be observed directly. Modeling allows for accurate attribution without identifying users (for example, due to user privacy, technical limitations, or when users move between devices). Including modeled key events allows Google to offer more accurate reporting, optimize advertising campaigns, and improve automated bidding.
How modeled key events work
Google’s models look for trends between key events that were directly observed and those that weren’t. For example, if key events attributed on one browser are similar to unattributed key events from another browser, the machine learning model predicts overall attribution. Based on this prediction, key events are then aggregated to include both modeled and observed key events.
Google's key event modeling approach
Check for accuracy and communicate changes
Holdback validation (a machine learning best practice) maintains the accuracy of Google’s models. Modeled key events are compared to observed key events that were held back, 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
Modeled key events are only included when there is high confidence of quality. If there isn’t enough traffic to inform the model, then modeled key events aren't reported (or, in the case of Google Analytics, are attributed to the "Direct" channel). This approach allows Google to recover loss of observability while also preventing over-prediction.
Customize for your business
Google’s more general modeling algorithm is separately applied to reflect your unique business and customer behavior.
Don’t identify individual users
Google doesn’t allow fingerprint IDs or other attempts to identify individual users. Instead, Google aggregates data (such as historical key event rates, device type, time of day, geo, etc.) to predict the likelihood of key events.
Modeled key events in Google Analytics 4 properties
Your Google Analytics 4 property began including paid and organic channel modeled key events around the end of July 2021. Data from before that date isn't impacted.
Core reports (such as the Event, key events, and Attribution reports) and Explorations where you can select event-scoped dimensions will include modeled data. These reports automatically attribute key event events across channels based on a mix of observed data where possible and modeled data where necessary.
Examples of key event modeling
- Browsers that don't allow key events to be measured with third-party cookies will have key events modeled based on your websites’ traffic.
- Browsers that limit the time window for first-party cookies will have key events (beyond the window) modeled.
- Some countries require consent to use cookies for advertising activities. When advertisers use consent mode, key events are modeled for unconsented users.
- Apple’s App Tracking Transparency (ATT) policy requires developers to obtain permission to use certain information from other apps and websites. Google won’t use information (such as IDFA) that falls under the ATT policy. Key events whose ads originate on ATT impacted traffic are modeled.
- When the ad interaction and the key event happen on different devices, key events may be modeled.
- Key event modeling covers both click-based events and engaged views for YouTube, to help with attribution for engaged-view key events.
- Any Google Ads conversions created based on Google Analytics key events will include modeling.