“Impressions” are misleading in marketing mix modeling

At Recast we always use spend — never impressions — to represent channel activity in our marketing mix model. We do not care how many gross ratings points (GRPs) your TV vendor claims to have delivered or how many impressions Facebook says you got on your ad. We don’t care about clicks or even the amount of website visitors that any vendor wants to claim. That puts us at odds with how traditional MMM vendors do things, but we have strong conviction in our choice to model spend not impressions.

We believe:

  • The goal of MMM is to drive bottom-line improvement in the business i.e. make you more money
  • Bottom-line improvement requires that we measure how marketing spend (dollars in) maps to the core KPIs of the business (dollars out)
  • Intermediate measures like “impressions” only muddy the water since every business would rather have fewer impressions that lead to more revenue than more impressions that lead to less revenue
  • Models using impressions are less interpretable, because you have to convert spend to impressions first, then impressions to sales. If you’re not careful you can get this seriously wrong by failing to account for saturation (cost per impression rises at higher spend levels!)

This way of working can be really shocking to marketers coming to Recast who are used to working with MMM vendors that put a lot of focus on things like GRPs. Unfortunately, many vendors like to focus on measuring impressions because:

  1. They’re relatively “easy” to measure and tend to be easier to work with when building a model by hand
  2. Impressions scale linearly with dollar investment into a marketing channel so vendors are extremely incentivized to sell impressions even if it doesn’t drive any real impact
  3. Big ad buys sometimes come with ‘free’ impressions or GRPs, thrown in to sweeten the deal or make up for past under-delivery

Beyond that…

We think impressions are bad metrics to include in an MMM for a few reasons:

  1. They aren’t measured apples-to-apples across platforms. An impression from Facebook is not the same as an impression from Youtube is not the same as a TV impression and definitely not the same as an estimated impression from a billboard campaign.
  2. Platforms often change how they measure impressions over time. So even within Facebook an impression today might not be the same thing as an impression 3 weeks ago
  3. Impressions are not closely tied to value: in many channels it’s easy to scale up and get a lot of low-value impressions but a raw impressions number will mask that
  4. There’s no such thing as ‘free’ impressions: this is just an accounting loophole to get away with charging a higher CPM on their rate card, ignoring the true CPM

All of these reasons make impressions a bad metric to use for running a business and for marketing mix modeling. At Recast, we want to focus on identifying which marketing channels are actually driving bottom-line improvements in the business, not just driving vanity metrics like “impressions”.

About The Author