ClearBrain's Causal Analytics platform enables you to track any conversion Goal of your choice, and in seconds analyze which actions are most likely to cause vs correlate to that Goal.
From the ClearBrain Insights page you will find a section titled Behaviors. This section lists all the user actions that occur prior to performing a selected conversion Goal.
ClearBrain's AI based platform in turn breaks down trends in your audience's behaviors, as well as which of these behaviors are most likely to cause conversion.
Analyzing Causal Lift of Each Behavior
A unique feature in ClearBrain is the ability to project the causal vs correlated impact of specific behaviors on your defined conversion goal. To help break down this analysis, we provide a set of key information about each user behavior:
- Behavior: The name of the user action being analyzed; can be any pageview, screen, or event
- Causal Lift: The predicted lift in conversion rate to your Goal, if a user performs this behavior; behaviors are by default ranked by their correlation error (a combination of the magnitude of the projected lift and how confident we are in its accuracy); if our algorithm is not confident enough in the projection to distinguish from correlation, the lift bar will appear grey
- Correlation Index: How much more likely this audience is to perform this behavior, relative to the average user
- Users with Action: The count of users who performed this behavior, and the percentage they represent of the total audience
- Average Frequency: The average number of times a user in this audience performs the behavior
Lastly, within this table you have a couple different filter options in the top right, to easily find your behavior of interest. The "Actions" toggle will filter the behaviors list to different categories (e.g. pageviews or screens). The "Time Period" toggle will allow you to analyze these behaviors performance over different time periods (e.g. last week, last 2 weeks, etc)
How Causal Projections Work
The foundation of projecting causal lift is in experimentation.
Traditionally, the only way to determine if a user behavior actually causes conversion would be to run an A/B test or feature flag experiment. By running a control-treatment experiment you can effectively isolate users who see one version of a page vs another set who don't, ensuring that the difference in versions is the cause of any difference in conversion lift.
ClearBrain's Causal Analytics algorithm works in a similar way, but off of simulated A/B experiments on forecasted conversions.
The algorithm works in two main steps: A) Eliminating Correlation Error through Confounding Variables, and B) Simulating Control-Treatment Effects on historical data.
In brief, the fundamental problem of causation vs correlation (that A/B experiments leap frog) is one of confounding variables. Confounding or lurking variables influence both the behavior of interest and the conversion goal, making it impossible to isolate the true causal effect of the behavior of interest.
ClearBrain in turn uses a statistical technique called Observational Studies to isolate these confounding variables in your data, and minimize their effect on your data.
Subsequently, ClearBrain's patent-pending algorithm will train a linear regression model (example below) on your historical data, combining your user behaviors and identified confounding variables into a single predictive model, the output of which is an average treatment effect.
ClearBrain then takes the difference in the average treatment effect (y) between when the behavior (B) in question is performed (treatment) vs not performed (control), which produces the projected causal lift.
For more info on how our Causal Algorithm works, see: