How we use causal impact analysis to validate campaign success - Part and Sum

In this tutorial, Wesley, a Marketing Technologist at Part and Sum (https://partandsum.com/) shows us how to run an analysis on website traffic to see how a specific campaign, event, or update as impacted traffic.

Have you ever wanted to know how much an advertising campaign affected your website traffic? Or maybe you just updated your website and want to know if your conversion rate has increased. With a minimal amount of coding, you can use this simple package in the R programming language to determine if your actions are having a statistically significant effect on your business.

Causal impact analysis did an action that we took, have a statistically significant effect on the outcome,

Wesley Duckett -   / wesley-duckett-077a03aa  

You don't need to know a lot about R in order to run this analysis, but you will need some basic functions.

Our Scenario
Step One, we gathered our data. We grabbed purchases for the cohort that we are targeting, we also want to get the purchases of other groups, those that are not being targeted by our ad.

Step Two: Using the Causal Impact R package, we then read in the purchase data.

Step Three: Clean up the data. While we don't technically need the data column we like to keep it. Then you designate the pre.periond and the post.period.
pre.period - c(1,30)
post.period - c(31,50)

Step Four: We run the Causal Impact package on the data from step three.
impace - CausalImpact(Purchases, pre.period, post.period)

Step Five: We plot our data, which will give us three graphs, with a dotted line that shows where our event took place.
Original - the solid black line shows us our actual observations that we saw. The dotted blue line shows us a prediction of what our model would look like if we did not take that action.

The Causal Impact Analysis will return analysis of the average purchases per day after the action taken (56) versus a prediction of the average daily purchases (45) had we not taken this action. This also returns the confidence of the prediction, in this case, it is a 95% confidence interval. This means that 95% of all of the predictions lands within the predicted.