Skip to main content
What data is needed for Marketing Mix Modeling (MMM)?
Lauri Potka avatar
Written by Lauri Potka
Updated over a year ago

Data specifications for Marketing Mix Models depend on the modeling scope. For example, Sellforte's Digital plan targeted for ecom companies includes only online media and online conversions using standard Sellforte data schema, whereas Growth and Enterprise plans can incorporate offline media and offline sales as well. However, all Marketing Mix Models leverage three broad categories of data (shared below) and have following characteristics:

  • Time-series format;

  • As granular as possible (daily data is typically sufficient);

  • Structured, machine-readable format.

1. Sales data

The main objective of Marketing Mix Modeling is to understand how marketing activities impact sales, which is why granular and high quality sales data is the foundation for all models. Following considerations are important when deciding on the data to be used:

  1. Dimensions of the sales data. Typically dimensions include:

    • Geography (for example, 'country')

    • Sales channel ('brick & mortar', 'online', ..)

    • Product group

    • For FMCG/brands with multiple brands: Brand

  2. Unit of the sales data: Volume or value

  3. Source of the sales data. There are a three main options to consider

    • Conversion data from online platforms (e.g., Google Analytics): This data can typically be easily automated using data connector services, such as Supermetrics

    • Sales data from an ERP or similar company-owned system: This data is often very granulary but requires a custom data connector to connect to the MMM model

    • Sales data collected by a 3rd party: This data has the lowest granularity (weekly, in some cases daily) and is often collected for brands without direct consumer channel

2. Marketing activity data

Marketing Mix Models use data on marketing activities to estimate the impact of different marketing activities on sales. Thus, the accuracy of data for a specific media has big impact for the model's ability to estimate the impact for that media. Following are typically dimensions that should be considered:

  1. Media hierarchy

    • Media Type: Online media, Offline media, Own media

    • Media Groups within Media Types: TV, OOH, Radio, Display, Search, etc.

    • Individual media channels with Media Groups

  2. Campaign hierarchy

    • Campaign types: brand, tactical, seanoal, etc.

    • Campaign names of individual campaigns (different for each company)

The dataset should include for all media:

  • Investments

  • Exposure metrics (such as TRPs for TV, Impressions for Display)

  • Others metrics relevant to the medium (e.g., clicks)

3. Non-marketing variables

There can be variables which have large impact on company sales, but are not related to marketing. It is important to control for the effect of these variables, so that the Marketing Mix Model doesn't attribute their positive or negative effects to media. These variables can include for example the following:

  • Pricing & discounts: Decrease in prices typically increase the volume of products sold

  • Distribution network: Having products more accessible to consumers has a positive impact on their sales

  • Macroeconomics: Downturn can have negative impact on sales of some products and positive on others

  • Weather: There can be products whose sales is positively or negatively impacted for example by rain or tempertature

  • Competitor activities: In certain industries, competitive activity can impact sales

  • COVID-19: In recent years COVID-19 related restrictions imposed by governments has had impact on many companies' sales

  • Holidays: Certain holidays could have positive or negative impact to some products

  • Seasonality: Different product have different seasonal patterns (this is often observed from the data)

This is not a comprehensive list, and including non-marketing variables is a model is always a case-by-case consideration.

Did this answer your question?