What is a Media Mix Model?

Let's start at the beginning, what is an MMM?

A media mix model is, at heart, a big regression model that is designed to tell you the impact of your media spend.

So, let's say you spent $10 million on a TV campaign and then you are faced with the rather sticky question of what that campaign actually did for the company. Perhaps your sales doubled overnight and it's clear to see just by looking at the top-line.

In practice, that's unlikely. Any company ready to spend on TV is likely beyond a size where such an immediate doubling of sales is possible. That's where MMM comes in. TV could have an impact on sales but then again, plenty of other things do too: Seasonality, longer-term growth trends, maybe the weather. And if you take those other factors into account, and then layer on your daily or weekly TV weight you can tease out how much of an impact your spend is having.

It's an imperfect science of course. For starters, there's no guarantee that your spend has any measurable impact on sales. And the noise from others factors may well drown out any media signal. But MMMs are great for stripping out any causal assumptions and starting your measurement from scratch. There's no hand-wringing over the last-click or view-thru windows. What's more, the fact that you have to control for big external factors (like seasonality) is both a bug and a feature. Because while it adds to the complexity of the project it means that you don't just get media impact as an output. You also get an understanding of seasonality, trending, and the impact of any other relevant external factor. Therefore, MMMs are a boon to both marketing and finance teams.

Another big feature/bug of media mix models is the aggregation of data. Meaning that, instead of looking at user-level paths to conversion, you are rolling the data up to time periods and geo units. On the one-hand there's definitely some loss of signal at that less granular level, but this also means that data is anonymised. That's good news in a world where data privacy concerns and, let's be honest, the self-interest of certain tech giants, means less access to user-level data.

So with all these positives why is the world of media measurement not all MMM all the time? The reasons are pragmatic. MMMs are imperfect, and acknowledging that imperfection means that reporting will change with the model. And there's nothing a conversation between marketing and finance needs less than changing reporting. The other blocker is cost. MMMs have traditionally been time consuming and expensive.

But these factors can be mitigated. Attribution reporting is nice and stable and works well for reporting. MMM can be there as a calibration tool to ensure that the numbers being reported represent what they claim. And the incredible progress we've seen in marketing science over the past 5 years has really opened these methodologies up and enabled teams to build in-house. This reduces the cost and turns a historically black-box model into something with deep internal knowledge.

To wrap up it's an imperfect science but for most marketing teams out there the benefits well outweigh the costs. To quote George Box "All models are wrong, but some are useful".

So if you are convinced and would like to learn more about building a useful model our next article "How do you build a Media Mix Model" is for you. Or if you like to talk to someone about it, reach out to us at info@codarossa.co