Marketing mix modeling (MMM) is an important analytical technique in the field of marketing, that employs statistical methods, primarily multivariate regression analysis, to measures the impact of marketing activities on business outcomes. With further information of marketing spend per marketing channel, it can evaluate the ROI (Return On Investment) of marketing campaigns and subsequently guide budget planning decisions.

Input Data

Business outcomes refer to sales, revenue, or other KPIs (key performance indicators). They are the dependent variable in the modeling. All other data are independent variables used to explain the variation of the dependent variable.

Marketing activities, as the key independent variable to measure its impact, refers to how many “eyeballs” have been exposed to the media, for example:

  • digital/online media (e.g. Facebook, Google Search, Youtube, Display, TikTok) corresponds to impressions
  • TV and radio corresponds to GRPs (Gross Rating Points), i.e. Reach x Frequency x 100
  • print (e.g. newspapers, magazines) corresponds readership

Besides marketing activities on these paid media examples, organic media may also be of interest to measure. Here, organic media refers to any marketing activities without a clear marketing spend (e.g. subscribed newsletters, app push notifications, official social media posts), so it won’t be applicable in ROI evaluation and budget planning.

Besides marketing activities on media, non-media may also be of interest to measure its impact, or be used as control variable in the modeling, such as promotions (including price discounts and bundled offers).

Other control variables (or called context variables [1]) that can help explain the dependent variable include competitor activities, macroeconomic factors including seasonality / holidays, economic growth, government policy changes, COVID-19 impact, etc.

Marketing Domain Knowledge

Adstock and saturation are two key concepts in the modeling process that pertains to the marketing field.

Adstock: the effects of advertising can lag and decay following an initial exposure - lag in the sense that not all effects of advertising are felt immediately, instead, memory builds and people sometimes delay action until following weeks; decay in the sense that this awareness diminishes over time. Popular modeling choices for adstock include:

  • geometric adstock: a percentage of marketing impact in period 1 are carried over to period 2.

    \[A_t = T_t + \alpha A_{t-1}\]

    in which $A_t$ is the adstock at time t, $T_t$ is the advertising value at time t, $\alpha$ is the retention rate, or geometric decay rate, whose typical values are 0.3~0.8 for TV, 0.1~0.4 for OOH/Print/Radio, 0~0.3 for Digital [1].

  • Weibull CDF adstock

  • Weibull PDF adstock

Saturation: each additional unit of advertising exposure increases the response, but at a declining rate. Popular modeling choices for saturation include:

  • Hill function:

    \[\text{Hill}(q, ec, \text{slope}) = \frac{1}{1 + (\frac{ec}{q})^\text{slope}}\]

    in which $\text{slope}$ controls the shape of the curve between exponential and s-shape, and $ec$ is the inflexion point, whose typical values are 0.3~1 [1].

References

[1] An Analyst’s Guide to MMM, https://facebookexperimental.github.io/Robyn/docs/analysts-guide-to-MMM.

[2] Meridian Model Specification, https://developers.google.com/meridian/docs/basics/model-spec.