From MMM to AMM to EMM to Hmmmm

The consensus is clear: Not All Reach Is Equal™.

No one in their right mind still believes that a native ad hugging the bottom of an article creates the same kind of impression as a 15-second spot running in an NHL game. Yet most media investment decisions are still being driven by a technique that ignores that reality.

Media mix modeling (MMM), the gold standard for diagnosing marketing effectiveness, by default uses gross impressions as the base input to the model. In a model that hasn’t been heavily customized, tweaked, and prodded, display ads and TV spots mathematically hold the same potential to influence perception and behavior. Since display ads are “cheaper” and nearly always on, they’re more likely to get credit for influencing incremental sales.

That’s, quite frankly, insane – and currently my favorite theory to explain why ROAS as a whole was in decline for most of the past decade.

We don’t have to surrender to that status quo. We can correct it by using new metrics and data sources.

Why is this critical now? MMM has been the gold standard for measurement for four decades, but as more media has shifted to digital, most organizations chose a shortcut for measurement: using some form of first- or third-party attribution. In the best-case scenario, this would be supplemented with other econometric modeling incorporating brand effects, but by and large, most were typically using the simplest form of attribution possible. That method is gone now. Cookie deprecation and loss of data flow from major “walled gardens” mean attribution methodologies are no longer reliable.

Thankfully, MMM is getting more accessible. Motives and trustworthiness aside, Meta released an open-source MMM capability called Robyn in 2022, and just this week, Google released Meridian, their open-sourced capability MMM (now in limited release).

In honor of this new platform, let’s design a better one. Three things will help un-break MMM:

1. Remember That an Impression Isn’t the Same as Attention

Again, not all reach is equal. But a statistical model needs something more quantitative than intuitive to express that fact.

Proposed Solution: Attention Mix Modeling (AMM)

Experiment with each of the leading attention-verification platforms to add the attention factor alongside gross impression volumes fed into the model (get granular – apply this not only at the channel level but also format mix, and as granularly as possible geographically to match the sales variability in the model).

2. Stop Ignoring the Force Multiplier

Breakthrough creative has proved to be the single clearest way to make campaigns more effective. MMM needs to account for the incremental impact that creative can deliver (or sacrifice, when under-delivering).

Proposed Solution: Emotional-Resonance Mix Modeling

Creative quality may be subjective, but emotional resonance testing provides a strongly correlated predictor of short- and long-term impact. Put it to work in our effectiveness models. Use simple star ratings from System1, resonance benchmarks from Nexxen’s EQ Max, Ipsos Breakthrough scores, or even the legacy Ace Metrix quality scores on an asset-by-asset basis to account for creative quality. Incorporating this into a model will require coordination (and lots of testing to backdate for all assets) but should offer far more insightful texture for the model to play with when developing response curves.

3. Reward Creatively Earned Contagion

MMM models should properly be called paid media mix models. While some may include a factor for public relations, even fewer effectively incorporate social conversation as anything other than base.

The main challenge is that earned media still is most commonly expressed through an unrealistic number of gross impressions. That overinflated quantity of impressions may compensate for the fact that in nearly every instance, earned coverage should come with an attention advantage. Contextually, it’s naturally endorsed or given greater credibility by the media entity or person amplifying the story, and in most cases, the form of that amplification captures a meaningful amount of time spent (TV story, article on a website, social share, etc.). This should all rank much higher than display and in-feed placements in terms of active attention per impression.

Proposed Solution: Quality Time Mix Model

Build on the AMM coefficients to account for the quantity of time spent with the brand through unpaid channels. The trick will be in isolating the earned time as specifically as attention derived from creativity – the volume of attention/time spent with the brand as a result of intentional creative provocations.

Would this be three new models, or one? I don’t know. I’m not a data scientist, I just stayed at a Holiday Inn Express last night.

Making an effort to better reflect the reality of how attention influences memory and action will make it more likely for us to invest in better work. Maybe even less wasteful (or annoying!) ads. One can only hope.