Skip to main content

Interpreting marketing science results

Had a linear regression analysis (LRA) or causal impact analysis (CIA) done? Here's what to do next!

S
Written by Strategy Organization
Updated over 2 years ago

Overview

  • The Data Intelligence team can assist proving the value of a channel or tactic

  • Marketing Science methods can be deployed to achieve this:

    • Linear Regression looks at historical data and is fast and easy to do

    • Causal Impact is essentially a fancy pre/post test but is more accurate

  • The purpose of this is to prove the value of a tactic when performance is “poor”

    • If attribution says Facebook has a $0.50 ROAS and client wants $2.50

    • If in-platform says that Trade Desk has delivered 0 conversions

    • If the client “just doesn’t feel the impact on the business”

The Analysis Comes Back Good

  • The linear regression analysis (LRA) or causal impact analysis (CIA) come back

  • The results look good or are better than attribution (e.g. $2.50 instead of $0.50)

  • Work with the data intel team to explain and present these findings to client

  • If the analyses convinces the client that the channel is actually effective:

    • Apply coefficient to in-plat and last click numbers

      • The total volume of conversions or revenue must remain the same

      • The proportion "attributed" to each channel will change

      • At first, reduce revenue from organic and/or direct

    • Potentially scale up spend if the channel is significantly outperforming goal

      • Example: business goal CAC of $50. Last click says Meta CAC is $150, so we reduce budget. But an LRA shows that the “real” CAC is $25. We can 2-3x our budgets without Meta’s CAC of $50 (however total blended CAC of all marketing might increase with more spend, be careful here)

    • Scaling spend helps the client grow their business as the ROI is positive

  • If the analysis does not convince the client that it is effective:

    • Recommend that we validate the findings through experimentation

    • The easiest and most common experiment is through a Matched Market Test

    • Some channels can’t MMT (e.g. Influencers, Affiliate, PR, Email)

      • Significant budget changes or other holdout tests can work instead

      • Run a second measurement analysis (generally a CIA) to validate

    • If experiments are not allowed and the client wants to reduce or remove the channel altogether, then we can still treat its removal as an experiment:

      • Remove the channel per their request

      • Work with the data intel team to measure its absence

      • Report on latent impact of removing channel weeks or months later

The Analysis Comes Back Bad

  • Explain the rationale as to why it might not be effective, adhering to what we think is the true likely cause given our expertise. Examples:

    • Investment level is too low

    • Creative isn't aligned with message or audience

    • Account hasn’t had time to optimize

  • Recommend that we validate these findings through experimentation (see above)

  • Come prepared with pivot solutions and other services that might assist the client’s needs in the absence of losing this channel (swap CTV for YouTube, etc)

Did this answer your question?