What Is A Media Mix Model?

What Is A Media Mix Model?

Written in collaboration with Roumen Vesselinov

?      Companies employ different media, different campaigns, and different ads to market their products and services.

?      Media Mix Modeling (MMM) provides a holistic view of how all the marketing activities work together and what the unique contribution of each one of them is to sales.

?      To accomplish this, MMM is built using complex econometric techniques.

?      A MMM may look something like this:


What econometrics methods are used to create MMM?

?      Pre-whitening of the time series – autocorrelation (ACF) and cross-correlation (PACF) functions, outliers, seasonality, holiday effect, structural change (one-time effect, e.g. new legislation, etc.)

?      Testing – lag evaluation, unit roots, integration and co-integration, periodicity

?      Model estimation –ARIMA, VAR, VEC, GARCH models (type determined after testing)

?      Post Estimation evaluation of the model – AIC, BIC and other Information criteria, residual analysis (QQ plot, ACF/PACF, GARCH, etc.)

?      Model Interpretation: Evaluates the clean individual effect of factor X while controlling for lagged values of Y (outcome) and controlling for all other factors included in the model.

An example:

?      Outcome: Web Revenue

?      Multivariate ARIMA(1,1,0) model

             Yt, Yt-1 – Web revenue for current day t, and with lag 1 day

             Xt, Xt-1 – Google branded search impressions, current and lag 1 day

Control Block – all other relevant variables: holiday effect, emails, other search engines impressions, promotions, discounts, all advertising channels, etc.

Results:

?      Multivariate ARIMA(1,1,0) model for Google branded search

Effect: Increase of 1000 Google branded impressions corresponds to increase in web revenue by $180, controlling for all other factors.

Efficacy: Investment of $60 (cost per 1000 impressions) in Google branded impressions corresponds to $180 increase in web revenue.

Impact Factor: Effect/Cost = 180/60= 3.0.

             One dollar investment brings about 3 dollars in web revenue.  

I hope this explains it all. And if not, you can always ping me and ask for more.

Why do you need MMM? click here

What skills are needed to build MMM? click here

要查看或添加评论,请登录

Nino Ninov的更多文章

  • The Very High Cost of Employee Turnover

    The Very High Cost of Employee Turnover

    We all suspect that the cost of employee turnover is high but how high? I recently conducted a survey sponsored by…

  • Branding Needs TV Advertising

    Branding Needs TV Advertising

    It is difficult to build a brand without TV advertising. Not that it is impossible; Google and Facebook were not built…

  • Price as Signal for Quality

    Price as Signal for Quality

    It is well known that price is, among other things, a signal for quality. The famous Van Westendorp Price Meter asks…

    2 条评论
  • Skills To Build Media Mix Models

    Skills To Build Media Mix Models

    Recently I was contacted by a recruiter who was scouring the web for people specializing in media mix modeling (MMM)…

  • Why Do You Need A Media Mix Model?

    Why Do You Need A Media Mix Model?

    Companies are swimming in a sea of marketing ROI analytics: Google Analytics, Facebook Analytics, Just-Trust-Me…

    1 条评论
  • Rumors of TV Advertising's Death are Premature

    Rumors of TV Advertising's Death are Premature

    The rumors about TV advertising's death are very premature. Prime time CPMs are through the roof compared to prior…

    1 条评论
  • Attribution Accelerator Conference: Summary of Presentations

    Attribution Accelerator Conference: Summary of Presentations

    On October 12 I attended a one-day conference in New York City dedicated to multi-touch attribution (MTA). The…

  • Tom Brady, Benoit Mandelbrot, and Statistical Probabilities

    Tom Brady, Benoit Mandelbrot, and Statistical Probabilities

    At some point during the LI Super Bowl somebody estimated that there was more than 90% probability of the Falcons…

    2 条评论
  • In Defense of Pollsters

    In Defense of Pollsters

    Pollsters have been blamed for predicting wrongly that Clinton will win the presidential elections. I find these…

  • Can Machines Have Feelings?

    Can Machines Have Feelings?

    Of course not is the intuitive and immediate answer that this question would most likely get. But let’s not be so fast.

    1 条评论

社区洞察

其他会员也浏览了