Part 2 of The Incrementality Battle: Regression Analysis
As I discussed in Part 1 of our Incrementality Playbook, creating a model to help drive investment decisions is a great first step. There is a need for marketers to understand where the most efficient next dollar is spent, but more importantly, whether that conversion was truly incremental or would have happened regardless.
Once you have the model in place, the next piece is to take a more mathematically robust step: understanding diminishing returns to allow for even better budget allocation decisions. It is important to note that Part 1 and Part 2 are not mutually exclusive—we can leverage an incrementality model in conjunction with a regression model to understand how our last dollars invested are performing.
Part 2: Regression Analysis
“Regression analysis is the hydrogen bomb of the statistics arsenal.” - Charles Wheelan
The mere mention of regression analysis usually sends a large portion of the population running for the hills, but it is a powerful tool in our toolkit. To make this analysis more digestible, below are some useful tips:
- Run a regression for each tactic separately: Branded search, Non-branded search, Competitive campaigns in search, Facebook/Instagram Prospecting, Facebook/Instagram Remarketing, Programmatic Prospecting, Programmatic Remarketing, Affiliate, etc.
- We have found that, in most cases, forcing a best fit line through 0 at the Y intercept is optimal (see more in the charts below).
- Build and test multiple models for each channel and tactic. Select the models with the best fit to the data.
- Don’t forget to take into account seasonality, trend, and differing spend levels across all channels, or other factors that impacted performance. Remove seasonality, normalize data, or adjust the time period analyzed if needed.
- Identify and remove outliers or noise (e.g., periods of testing).
- Last click attribution, while not ideal, is ok as a starting point! You won’t have a deduplicated total cost to acquire an incremental conversion, but the insights will still allow you to make better cross-channel investment decisions in short order.
For those of you a little further along on your journey, you can work to incorporate a data-driven attribution model as your source of truth for a more unified view of the value of each channel. There are still limitations, especially when dealing with the lack of view-through data from Facebook/Instagram, but a data-driven attribution model will give you a baseline to measure success against as you test and scale.
By bringing Part 1 and Part 2 of the playbook together, marketers are able to better allocate their marketing spend to the tactics which are truly having an impact on their bottom line and bringing in revenue that is incremental to the business.
If you’re interested in learning how to do this for your company, reach out to Rise.