At this step, you need to choose which channel grouping representing a custom grouping should be used for the specific attribution model configuration. You can also apply additional settings for each channel within the channel grouping. These settings will be unique to the specific attribution model you are currently setting up.
1. Select a channel grouping
Select a channel grouping from the dropdown menu.
Channel grouping represents a custom grouping for marketing channels and is used to group data into logical categories to support better decision-making. It is important to note that channel grouping needs to be defined before you start setting up an attribution model. You can learn more about how to set up a custom grouping here.
There is a certain dependency between a channel grouping and an attribution model. Let’s look at the example:
1. The channel grouping is greyed out, meaning you can see its name but cannot select it. This is because the data sources used within the channel grouping and the data sources selected for the attribution model are different and do not match each other.
In the screenshot above, the channel grouping Funnel Grouping (Precis starter report) is greyed out because it uses the Meta data source, which is not used in the attribution model.
2. You can select a channel grouping here, but not all data sources match between the channel grouping and the attribution model configuration. Non-matched data sources will be greyed out and will not be included in the attribution modeling.
In the screenshot above, the channel groupings Market Grouping (Precis starter report) and Channel Grouping (Precis starter report) can be selected, but they only have three data sources that match between the channel grouping and the attribution model: Google Analytics 4, Google Adwords, and TikTok.
3. You can select a custom grouping here, and all data sources match between the channel grouping and the attribution model configuration.
In the screenshot above, the channel grouping Copy of Channel Grouping can be selected, and all data sources match between the channel grouping and the attribution model: Google Analytics 4 and Google Adwords.
2. Advanced settings for the channels within the selected channel grouping
In the advanced settings, you have the option to either adjust the expected ROI deviation or calibrate your attribution model with incrementality experiment data. These options are mutually exclusive, meaning that enabling one will automatically disable the other.
2.1. Set expected ROI deviation
This is the default configuration you will find when accessing the advanced settings for each channel within the channel grouping. It allows you to indicate how much you expect a channel’s Return on Investment (ROI) to differ from the norm, using a seven-step scale ranging from "No expected return" to "Significantly above normal".
If you expect a channel to perform below average, set a lower value. If you expect it to perform better than average, set a higher value. By default, all channels are expected to perform "At normal levels", which is the default configuration for every channel.
Adjusting the expected ROI deviation can be especially helpful if you are aware of tracking issues in your data or have conducted tests showing that the default attribution does not fully capture a channel’s true performance. This feature allows you to influence the model results based on your own insights, knowledge, or test findings.
To adjust the expected ROI deviation:
In the Attribution configuration flow, under Step 3 - Set up your channel grouping, go to Advanced settings.
For new Attribution configurations, simply select the channel for which you want to change the Expected ROI deviation. When editing a configuration, you will first need to click Edit for each channel you want to modify before proceeding.
By default, the Expected ROI deviation is always set to "At normal levels".
Use the slider to adjust the Expected ROI deviation.
When editing a configuration, press Save within the channel configurations. For new Attribution configurations, saving the Attribution model will apply the settings and recalibrate the model, taking the updated Expected ROI deviation into consideration.
2.2. Calibrate with experiment data
Attribution model priors define the model's initial assumptions about the contribution of each marketing channel. By default, the attribution model uses priors derived from your GA4 property to estimate the contribution of each marketing channel. However, you can replace these priors with measured results derived from your own incrementality experiments, allowing the model to incorporate causal evidence specific to your business.
To calibrate your attribution model with experiment data:
In the Attribution configuration flow, under Step 3 - Set up your channel grouping, go to Advanced settings.
For new Attribution configurations, simply select the channel you want to calibrate. When editing a configuration, you will first need to click Edit for each channel you want to modify before proceeding.
Toggle the option Calibrate with experiment data.
A form prompting you to provide experiment information will automatically open. alternatively, click Add new experiment.
5. Start by giving your experiment a name. This will help identify it if more than one experiment has been conducted for the same channel.
Next, select the KPI that was measured during your experiment: Revenue or Conversions.
If you select Revenue, you will be prompted to provide the incremental ROAS (iROAS) obtained from your test.
If you select Conversions, you will be prompted to provide the incremental CPA (iCPA) obtained from your test, as well as the Spend Currency used for the treatment group.
Next, select the method you want to use to provide the uncertainty values from your experiment. For more details on uncertainty, go to the next section.
Finally, select the start and end date of your experiment.
Note that when editing existing configurations, you will need to click Save within the channel configurations to apply the changes. For new Attribution model configurations, saving the entire configuration will apply the calibration with experiment data settings.
2.3. About 'Uncertainty' in experiment data
Incrementality tests estimate the additional impact generated by a marketing activity, such as additional conversions or revenue caused by a campaign. However, every experiment result has a level of uncertainty because the result is based on a sample of data rather than the entire population of customers.
Uncertainty helps indicate how reliable the estimated incremental impact is. A result with low uncertainty provides stronger evidence that the measured lift is close to the true impact, while a result with high uncertainty indicates more variation and less confidence in the exact lift value.
When calibrating your model with experiment data, you are required to provide the uncertainty of the experiment.
Alvie allows three different methods to provide uncertainty:
95% Confidence interval
Standard error
Relative uncertainty
1. 95% Confidence interval
This is the default option. Confidence intervals are a commonly used way to communicate uncertainty in incrementality test results and are often included in experiment outputs from many measurement platforms. They show the range within which the true incremental impact is expected to fall based on the experiment results.
For example, if an experiment reports an incremental ROAS of 2.0 with a 95% confidence interval between 1.2 and 2.8, this means that the true incremental ROAS is likely to fall within that range. A narrower interval indicates a more precise result, while a wider interval indicates more uncertainty.
In Alvie, when selecting 95% Confidence interval, you will be prompted to provide the upper and lower bounds from your experiment, as per the example above.
2. Standard error
When selecting Standard error, you will be prompted to enter a unique percentage value greater than zero, which measures how much the estimated incremental impact could vary if the same experiment were repeated with different groups of customers.
In incrementality testing, a smaller standard error means that the experiment result is more precise and the estimated lift is more reliable. A larger standard error means there is more uncertainty around the measured impact, and the true incremental effect could be further away from the reported result.
3. Relative uncertainty
When selecting Relative uncertainty, you will be prompted to indicate how confident you are in the incremental result from your experiment. This option is useful when the exact statistical uncertainty of an experiment result is not available and an estimated level of uncertainty needs to be applied.
Your confidence level determines the expected uncertainty applied to the result. Higher confidence indicates a more reliable estimate and lower uncertainty, while lower confidence indicates greater uncertainty around the reported result.