Putting the “Media” In Media Mix Models

If you talk attribution to nearly any marketer today – most of whom “grew up” professionally on digital media  – they will talk about audience identifiers and digital footprints and geo-location data and cohorts and deterministic models.

That’s not media mix modeling (mmm). Media mix modeling is algorithmic, powerful probabilistic modeling. It’s not dependent on cookies.

There’s a simple, but important tenant to media mix modeling, which is garbage in = garbage out. And therein lies the rub.

Typical off-the-shelf mmm solutions rely on a black-box algorithm that a really smart data scientist developed. But that solution, and the team supporting it, never worked as a media buyer or planner. They might not know that an online display impression is not a TV impression is not a billboard impression.

This is a problem.

Take out-of-home media, for example. Many models struggle to obtain an attribution “signal” for billboards, and there’s a wonky reason for this. Billboard impressions are flat – they are the same number every day of the week; every week of the month; every month of the year. That’s not reality and only someone who knows the methodology of Geopath (the out of home industry measurement company) could fix this.

FutureSight.online was built by media teams for media teams. Sophisticated data science with models like Bayesian and neural networks are at the center, but it’s media experts who wrangle the inputs.

And that’s why it works.