Attribution gives you the ability to customize how dimension items get credit for success events.
In some reports, you might want the order attributed to Paid search. In other reports, you might want the order attributed to Social. Attribution lets you control this aspect of reporting.
You can set a default attribution model for a given metric by updating the metric’s setting in the data view. Doing so overrides the metric’s attribution model any time it’s used in Analysis Workspace.
To update a component’s default attribution model:
Go to the data view that contains the component whose default attribution model you want to update.
Select the component, then expand the Attribution section on the right side of the screen.
Select Set attribution, then select the attribution model in the Attribution Model drop-down menu.
See Attribution models to learn about each attribution model.
Select Save and continue.
If your organization requires that a metric has multiple attribution settings, you can do one of the following:
Copy the metric in the data view with each desired attribution setting. You can include the same metric multiple times in a data view, giving each metric a different setting. Make sure that you label each metric appropriately so that analysts understand the difference between these metrics when generating reports.
Override the metric in Analysis Workspace. In a metric’s Column settings, select Use non-default attribution model to change the metric’s attribution model and lookback window for that specific report.
An attribution model determines which dimension items get credit for a metric when multiple values are seen within a metric’s lookback window. Attribution models only apply when there are multiple dimension items set within the lookback window. If only a single dimension item is set, that dimension item gets 100% credit regardless of attribution model used.
|Last Touch||Gives 100% credit to the touch point occurring most recently before conversion. This attribution model is typically the default value for any metric where an attribution model is not otherwise specified. Organizations typically use this model where the time to conversion is relatively short, such as with analyzing internal search keywords.|
|First Touch||Gives 100% credit to the touch point first seen within the attribution lookback window. Organizations typically use this model to understand brand awareness or customer acquisition.|
|Linear||Gives equal credit to every touch point seen leading up to a conversion. It is useful where conversion cycles are longer or require more frequent customer engagement. Organizations typically use this attribution model measuring mobile app notification effectiveness or with subscription-based products.|
|Participation||Gives 100% credit to all unique touch points. Since every touch point receives 100% credit, metric data typically adds up to more than 100%. If a dimension item appears multiple separate times leading up to a conversion, values are deduplicated to 100%. This attribution model is ideal in situations where you want to understand which touch points customers are exposed to the most. Media organizations typically use this model to calculate content velocity. Retail organizations typically use this model to understand which parts of their site are critical to conversion.|
|Same Touch||Gives 100% credit to the same event where the conversion occurred. If a touch point does not happen on the same event as a conversion, It is bucketed under “None”. This attribution model is sometimes equated to having no attribution model at all. It is valuable in scenarios where you do not want values from other events affecting how a metric gives credit to dimension items. Product or design teams can use this model to assess the effectiveness of a page where conversion happens.|
|U Shaped||Gives 40% credit to the first interaction, 40% credit to the last interaction, and divides the remaining 20% to any touch points in between. For conversions with a single touch point, 100% credit is given. For conversions with two touch points, 50% credit is given to both. This attribution model is best used in scenarios where you value the first and last interactions the most, but don’t want to entirely dismiss additional interactions in between.|
|J Curve||Gives 60% credit to the last interaction, 20% credit to the first interaction, and divides the remaining 20% to any touch points in between. For conversions with a single touch point, 100% credit is given. For conversions with two touch points, 75% credit is given to the last interaction, and 25% credit is given to the first. Similar to U-Shaped, this attribution model favors the first and last interactions, but more heavily favors the last interaction.|
|Inverse J||Gives 60% credit to the first touch point, 20% credit to the last touch point, and divides the remaining 20% to any touch points in between. For conversions with a single touch point, 100% credit is given. For conversions with two touch points, 75% credit is given to the first interaction, and 25% credit is given to the last. Similar to J-Shaped, this attribution model favors the first and last interactions, but more heavily favors the first interaction.|
|Time Decay||Follows an exponential decay with a custom half-life parameter, where the default is 7 days. The weight of each channel depends on the amount of time that passed between the touch point initiation and the eventual conversion. The formula used to determine credit is
|Custom||Allows you to specify the weights that you want to give to first touch point, last touch point, and any touch points in between. Values specified are normalized to 100% even if the custom numbers entered do not add to 100. For conversions with a single touch point, 100% credit is given. For interactions with two touch points, the middle parameter is ignored. The first and last touch points are then normalized to 100%, and credit is assigned accordingly. This model is ideal for analysts who want full control over their attribution model and have specific needs that other attribution models do not fulfill.|
|Algorithmic||Uses statistical techniques to dynamically determine the optimal allocation of credit for the selected metric. The algorithm used for attribution is based on the Harsanyi Dividend from cooperative game theory. The Harsanyi dividend is a generalization of the Shapley value solution (named after Lloyd Shapley, a Nobel Laureate economist) to distributing credit among players in a game with unequal contributions to the outcome.
At a high level, attribution is calculated as a coalition of players to which a surplus must be equitably distributed. Each coalition’s surplus distribution is determined according to the surplus that was previously created by each subcoalition (or previously participating dimension items) recursively. For more details, see John Harsanyi’s and Lloyd Shapley’s original papers:
Shapley, Lloyd S. (1953). A value for n-person games. Contributions to the Theory of Games, 2(28), 307-317.
Harsanyi, John C. (1963). A simplified bargaining model for the n-person cooperative game. International Economic Review 4(2), 194-220.
A lookback window is the amount of time a conversion should look back to include touch points. If a dimension item is set outside of the lookback window, the value is not included in any attribution calculations.
Consider the following example:
Depending on your lookback window and attribution model, channels receive different credit. The following are some notable examples:
2^(-0/7) = 1
2^(-0/7) = 1
2^(-6/7) = 0.552
2^(-9/7) = 0.41
Conversion events that typically have whole numbers are divided if credit belongs to more than one channel. For example, if two channels contribute to an order using a Linear attribution model, both channels get 0.5 of that order. These partial metrics are summed across all people then rounded to the nearest integer for reporting.