Back to Gartner for Marketers Blog

Should You Do Attribution or Mix Modeling, or Both?

By Martin Kihn | October 06, 2016 | 2 Comments

Attribution and marketing mix modeling solutions measure the impact of marketing and media efforts, and suggest ways to improve that impact. Increasingly, marketing leaders use both methods — either in parallel or unified within the same platform — to expand the scope and precision of their measurement.

It takes skilled talent, experienced with idiosyncratic, complex marketing data, to navigate the methods and models available and select the right approach (or ensemble of approaches). The line between software and services is particularly blurred in this market. Most marketing leaders can only execute simple projects in-house.

Nonetheless, there are some powerful software tools that continue to improve. We recently published a couple of Market Guides to attribution and marketing mix modeling and advanced analytics service providers for marketing, as well as a basic guide to understanding the space. (Clients can find them here.)

Marketing Mix Modeling

Marketing mix modeling (MMM) is a time-tested way to measure the high-level impact of a range of marketing and media tactics on your business. It is usually used to guide investment decisions by showing which channels and tactics work better than others. Specifically, MMM applies econometric regression techniques to aggregate data to estimate the impact of marketing activities on a desired outcome, such as sales.

This data typically includes many months of historical information about media spending across both digital and offline channels such as TV and print. They also include nonmedia factors that are likely to influence the desired outcome. Nonmedia factors include promotions, coupons, competitive activity, seasonality and consumer sentiment.

Note that models including only media-related data are known as media mix models. These are a subcategory of marketing mix models.

MMM is a top-down approach because it starts with aggregate data at the level of campaigns and markets, not individuals. Virtually all major brands and many midsize marketers use MMM. The customization required, particularly in identifying and collecting the required data, means that most MMM projects include professional services or consulting support.

At a minimum, MMM providers must:

  • Incorporate aggregate data from a variety of sources in different formats (e.g., store transaction logs, media spend receipts, third-party vendor data).
  • Provide methods to update, cleanse, deduplicate and normalize data, particularly aligning time series.
  • Give analysts a range of standard visualization, exploration and statistical modeling tools.

Multitouch Attribution

Multitouch Attribution (MTA) providers support the analysis of the large amounts of data generated by digital media, particularly paid search, display advertisements and email. MTA methods require collecting information about individual consumers over time. This information ideally includes every exposure the consumer has had to a marketer’s messages and his/her response (including no response).

Using techniques from simple rules to complex algorithms, the provider develops a model of the impact of the messages on a desired outcome. Because it uses individual-level rather than aggregate data, attribution is often referred to as a bottom-up approach.

Typically, only marketers with significant digital media spend and/or a significant share of online conversions use attribution providers. Useful attribution requires high volumes of data and advanced analytics support. Cost and complexity puts full multichannel MTA out of reach for most smaller marketers (although they may use rule-based or single-channel methods). Most attribution providers include significant services support to implement their software.

At a minimum, MTA providers must:

  • Ingest and/or collect a very large volume of user-level data rapidly with high fidelity and throughput.
  • Create, harmonize and maintain a discrete user-level record of as many known and probable unique users as possible, over time.
  • Apply a variety of regression, predictive, categorization, ensemble and other machine learning methods to the data.

Unified Measurement

Mature marketing leaders understand the benefits of combined top-down and bottom-up approaches. In practice, however, the different data sources and levels of data capture make manual reconciliation of the two approaches difficult. As a result, providers have begun to offer paired solutions.

Unified providers should also have a documented approach to combining MMM and MTA and a willingness to explain how the models are reconciled. For more detail on these topics see “Understand Attribution and Marketing Mix Modeling” (clients enjoy it here).

Not sure which methodology is right? Navigate through the high-level considerations to determine the options best suited to your data.

Decision Tree: How to Identify Which Measurement Solution to Use

foto

Source: Gartner (September 2016)

Comments are closed

2 Comments

  • Great read as always! Very interesting as we are moving from digital to physical sales, so this kind of information is always helpful.

  • Interesting article. I agree that there is no standard approach to analytical attribution and would like to take this one step further, even. Our clients have shared their frustrations with attribution modelling because this simply doesn’t deliver on their goal – smart media budget allocation. And here’s why I think so: attribution modelling is currently based on a siloed channel point of view. We believe that this is outdated and that attribution should be based on customer journeys, or customers more specifically, regardless of the channels they use. The value of customers or estimated value of prospects should sit in the lead when determining your media spend. We’ve started to elaborate on this idea here if you’d like to know more: https://www.marketingweek.com/2016/05/13/customer-strategy-not-attribution-is-a-marketers-route-to-the-boardroom/