Over the next five years, marketers will experience the demise of user-level tracking. We’ve already seen this happen when Apple’s Safari rolled out progressively harsher restrictions on cookies via its Intelligent Tracking Prevention (ITP) initiative. The decline of user-level web tracking accelerated at the start of 2020, when Chrome announced that it also would phase out all third-party cookies.

A similar story has been playing out in the realm of app tracking, ever since Apple first allowed users to opt-out of device ID (or IDFA) sharing back in 2016. The announcement last year that IDFAs would become opt-in only, and therefore largely unusable by advertisers, signaled the end of device-level tracking on iOS. With many expecting Android to make a similar move in the near future, the user-level tracking that advertisers have become so accustomed to appears to be on its way out.

Since user-level tracking has been so pervasive within the digital advertising ecosystem, where do we go from here? Is there a way to measure in a unified manner? Maybe. Media Mix Modeling (MMM) helps measure different channels with a single, unified approach.

What Is Media Mix Modeling?

MMM first came on the scene in the ‘60s and ‘70s, when marketers didn’t have access to user journey data the way we do today. It doesn’t look at individual users, nor does it try to understand what influenced them in terms of the specific ads they’d seen prior to conversion. MMM takes a much higher-level approach to attribution. It’s essentially a mathematical technique that tries to understand the relationships between a set of input variables (e.g. how much you spend on your various marketing channels) and an output variable (how many conversions your brand generates).

MMM tells us how effective each of our different channels are based on how strongly they appear to influence our KPIs. 

Let’s break it down with a simple example.

Putting Basic Media Mix Modeling to Work

Digital marketers are always trying to quantify the impact of upper funnel channels on a brand’s KPIs. Let’s imagine we find ourselves in this situation, and we’re trying to understand YouTube’s role in our brand’s overall conversions.

A typical way we might approach this is to look for correlations between the amount we spend on YouTube each day, and our brand’s organic volume. We could do this by going into our various platforms, pulling date-segmented data for the YouTube ad spend and our overall conversions, and end up with a graph that might look something like this:

A quick glance at this graph shows that there appears to be some correlation between YouTube spend and the brand’s conversion volume.

One way that you might seek to quantify this is to perform what’s known as a regression analysis on the data. This mathematical technique gives you an equation that estimates your brand’s overall conversion volume on a particular day, based on the YouTube spend on that day. It might look something like:

This equation implicitly gives us a cost per conversion (CPC) for YouTube and allows us to quantify how effective it is at driving conversions.

From Basic to Advanced Media Mix Modeling

Advanced MMMs are scaled up versions of the YouTube example. The only differences are that:

• In practice, MMMs will typically look at the impact of multiple channels on a KPI.
• More sophisticated MMMs are able to take into account the delayed effect of advertising. Meaning, we understand that we don’t expect the impact of a channel like YouTube to happen immediately after a user sees our ad.

What’s So Great About Media Mix Models?

A cynical view might treat MMM as a nice-to-have, but not something to center a marketing strategy around. There are a couple of reasons why this misses the point, and why MMM is a serious approach to attribution:

• No user-level data – Remember, MMM doesn’t need any user-level data. This makes it incredibly privacy-friendly, and also protects you from any changes that platforms make to their tracking capabilities.
• Platform agnostic – Pretty much any user-based approach to tracking can be biased toward certain channels. MMM doesn’t have such biases.
• Choice of inputs – The inputs to your model don’t have to just be about the spend. You could include variables like competitor advertising volumes, or even factors like weather (if your brand is traditionally affected by it).

With the imminent demise of all but first-party cookies and mobile device IDs, marketers are going to be pushed to find new ways of measuring their channels. It’s not often that marketers talk about adopting older approaches, but now is a great time to activate Media Mix Modeling.

If you have any more questions, let’s talk.