The DDA-based budgeting model in Google Analytics uses Data-driven Attribution (DDA) results and historical performance data to help you plan and optimize media budgets. It estimates how budget changes may impact future conversions and revenue.
On this page
Before you begin
To use cross-channel budgeting, your property must meet certain eligibility criteria. If your property isn't eligible, you can view the required details that determine its status.
To use these reports, your property will need to meet the following criteria:
- Conversion data: Cross-channel budgeting requires at least one year of data for a conversion to be eligible for Projection or Scenario plans. Web conversions are supported, while app conversions aren’t yet supported. Key events aren’t supported. Budgeting models will use historical event data to backfill associated conversions when available if the one-year data requirement is met..
- Campaign data (cost): Cross-channel budgeting requires at least one year of campaign data from at least two channels, including Google and non-Google. Link to your Google products to import historical data for Google integrations. You can use data import to upload campaign data for non-Google integrations or other import sources. After you upload campaign data, it can take up to one day to view a status update on the budgeting eligibility page for your property. If you map your primary channel groupings using campaign data, ensure that you upload data for the "Campaign field” in your data imports.
After you meet the requirements, Google evaluates your models. The evaluation can take up to two days, which includes cost import correctness and model quality. You can create plans after the checks pass. A property may not pass model validation if there are anomalies or major gaps in its conversions or cost data.
Deleting or making changes to your product links, campaign data import, or channel grouping could impact your budgeting models.
How it works
Import sources, dimensions, and metrics
Cross-channel budgeting uses data from your Google Analytics property, including data-driven attribution, aggregated campaign information from linked products and data imports, and offline data when available. Only paid channels are included in budgeting results, organic channels are not. Key metrics include:
- Conversions: The number of times users triggered a conversion.
- Ads cost: The total amount paid for ads.
- Cost per conversion: Total cost divided by number of conversions.
- Total revenue: Revenue from purchases, subscriptions, and advertising.
- Return on ad spend: Revenue for selected conversions divided by total ad cost.
Model methodology
The DDA-based budgeting model in Google Analytics uses your property’s historical data driven attribution as the primary input to a Bayesian regression model to forecast the optimal total budget per channel over a defined time period. This regression model is based on your property’s attribution data.
Bayesian regression modeling is a statistical analysis technique that measures the impact of historical marketing activities to guide future budget planning decisions. The DDA-based budgeting Bayesian regression model combines your property’s DDA historical data with signals learned from imported data to estimate media effects. DDA-based Budgeting in Google Analytics fundamentally considers what would’ve happened if the marketing spend had been different based on data-driven Attribution priors (as an attribution input) and response curves (as a Bayesian regression output), ultimately recommending optimal budget allocation based on how marketing spend affects KPIs such as revenue and conversions.
Forecasting and response curves
A time-series model is used to forecast the spending patterns on a per-channel basis. Revenue and conversions are estimated from response curves learned from historical data in the Google Analytics property. Based on the response curves and forecasted spend, revenue and conversions are projected for future time periods. The ROI estimates are consistent with data driven attribution. Models are trained on 12 months of data to account for seasonality.
Model confidence
Budgeting models show results for paid channels with high model confidence based on the historical data availability. Model confidence levels are determined by a backtesting methodology for cost per channel. Models are tested on historical time periods to evaluate accuracy of your property’s budgeting model. Similarly, for conversions and revenue, historical time periods are used to assess how well the trained model's channel-level estimates match the historical data-driven attribution revenue and conversions. Low model confidence typically indicates gaps in campaign data, changes in conversion implementation, changes in your primary channel grouping, or deleted conversions or channels.
Using a DDA-based model with Projection and Scenario planner allows you to plan budgets based on historical performance data.
Reasons a property may not pass model checks
In some cases, even if your property meets the initial data eligibility requirements, a model might not be available for your property. This typically occurs because the underlying data lacks the necessary signals, or because the model's outputs failed our strict quality assurance tests.
When a model doesn’t pass, it’s generally due to one of these reasons:
- Data limitations: The model needs a rich, diverse dataset to learn accurately. It will fail if there is insufficient historical history, missing cost and impression data from third-party platforms, or a lack of cross-channel diversity in your media mix.
- Model convergence: Model convergence refers to the model being able to detect consistent and repeatable patterns in the data, which increases the confidence level that those patterns are accurate. The model tries to find a stable set of parameter estimates that explain the relationship between estimated channel contributions (marketing spend) and ROI (business outcomes). Models aren’t able to detect consistent patterns when they have incomplete data, missing data, or sales data that doesn’t match the conversion patterns.
- Misalignment with reality: A model will fail if it generates impossible scenarios, like negative baseline sales. This is a sign that there is a statistical error or misspecification in your model that needs to be addressed before it should be used for decision making.
- Model fit: The model’s predicted outputs significantly differed from historical, observed data for the property.
Refer to the Budgeting overview page for your property’s model quality check status and additional details.