Skip to content

Schedule 2024

Opening the black box: how to validate your media mix model

Michael Kaminsky  (Founder, Recast)

Location: Room 32A

Date: Tuesday, October 22

Time: 9:05 am - 9:45 am

Pass Type: All Access Pass, Digital Pass, Main Conference Pass

Theme: Marketing Analytics

Session Type: Session

Track: Measurement & Modeling

Vault Recording: TBD

Audience Level: Intermediate

Media mix models have the potential to be very powerful tools for constructing optimal marketing budgets. However, these types of models are often opaque and notoriously difficult to validate and know that they’ve picked up true causal signal. In this talk we will walk through how to think about validating media mix models in order to avoid huge amounts of wasted marketing budget. This talk will cover an introduction to MMM for those new to the field, as well as actionable advice for people who are further in their MMM journey.

Takeaway

  • Marketing mix modeling has grown in popularity over the last few years as a response to changes in the privacy landscape, but many stakeholders do not know how to properly assess the accuracy and efficacy of their models
  • Marketing mix modeling is a complex regression analysis technique
  • Because of this, we need ways to validate the results, outside of the modeling framework itself, to know that the model has picked up the true causal relationship between marketing spend and a business KPI
  • We need measures of both internal- and external- validity of the model
  • We can use techniques like parameter recovery, backtesting, and experimentation to validate the models