Building Training Data for MMM

Updated by Eddie Chou

Rockerbox MMM helps brands explore the relationship between marketing investments and revenue based on their historical data. In order to deliver to Rockerbox that historical data, Rockerbox supports two methods:

  • Automated ingestion for major ad platforms listed here as well as revenue for customers selling through Shopify
  • UI for customers to upload:
    • revenue from non-Shopify platforms
    • ad spend from platforms not already automatically ingested
    • impression-based marketing, such as email and SMS sends
    • other non-spend factors, including economic indicators, promotional dates, seasonality, and other relevant variables beyond direct marketing spend and impressions.

Process Overview

  1. Rockerbox automatically ingests spend from major ad platforms
  2. Customer prepares revenue, additional spend, and other marketing data (collectively these are known as features) to include in MMM
  3. Customer uploads the data via the UI Uploader
  4. Customer selects which of the features between 1) and 3) to include
  5. Customer kicks off a model build
  6. After a few business days, Rockerbox will alert the Customer that the new model is available in platform
To refresh the MMM model, customers will repeat steps 2-4 with updated data.

Preparing Data

An important and often most time-consuming part of building MMM training data is compiling the required data from all the disparate sources. Any data you already have consolidated into readily available reports will assist in this process.

As you collect your data, we recommend creating a single spreadsheet or dataset that includes:

  • Date (ISO 8601 format)
  • 1 column for each "feature" you plan to include, where each row includes that feature's value

    Feature Type

    Values to Include

    Example Features

    Revenue

    Revenue in USD

    Amazon eComm, Retail

    Marketing Spend

    Cost in USD

    DSP spend, Reddit spend

    Impressions

    Impression counts or total sends

    Email or SMS sends, PR impressions

    Extra Features

    Binary 1 or 0 for whether a promo was active that day;

    Binary 1 or 0 for key product launch dates;

    Discount rate

    Promotions, product launches, site-wide discounts

Uploading Data


How did we do?