Multi-Channel Attribution When MMM Falls Shortback to all NEWS

Insights | Dec 14, 2017

The world of marketing analytics has been hungry for change. Marketing mix modeling (MMM) has been a commercial solution for marketing measurement since the 1990s. The methodology – correlating weekly level data by geography/DMA – was a function of the available data at the time, and a media market whose currency was the Gross Rating Point (GRP). Marketers have long complained of the slow time to insights and a lack of tactical detail, especially in a rapidly emerging digital media environment.

The explosion of online media channels powered by “big data” produced some hopeful alternatives to MMM. The most common alternative is digital attribution, which leverages digital clickstream data to provide user-level insights, personalization capabilities, and includes applications such as embedded analytics for the media buying process. You would think that this would spell the end of our old friend, MMM. But it has not. How has this traditional approach to both media and analytics hung in there? Here are a few big reasons:

1. Offline media

Because digital attribution uses user-level data derived from a DMP, data on offline media exposure doesn’t exist. Attribution, therefore, is blind to users’ exposure to traditional channels like TV, print, and radio. Even so-called “multi-channel” attribution is only named “multi” in the sense that it measures multiple digital channels. Although shrinking, offline media is still a considerable part of the marketing mix and marketers need some tool to measure it. Until the industry can solve for mapping online and offline media channels to a household or user, we still need MMM.

2. Offline sales

Digital attribution analyzes user-level sales and activity, but a significant amount of sales are still offline and anonymous for many big box retailers, drug stores, and other parts of the economy. Ignoring them is a mistake. Because MMM correlates weekly store-level changes in sales with weekly market-level changes in media weight, it can estimate what marketing is driven offline, online, and anonymously. The digital attribution approach requires user-level sales to be linked to user-level media exposure, which means it ignores sales where that data does not exist.

3. Correcting for bias

Because major offline sales influences are excluded from attribution, it risks overstating the influence of digital media. For example, if someone sees both a TV and display ad and makes a purchase, digital attribution will credit the display ad with 100% of the sale. But because MMM doesn’t require customer-level data, it can incorporate both and measure digital while controlling for offline media’s impact.

4. Data infrastructure

If you can’t already tell, setting up the proper data infrastructure is critical for attribution to be useful. You need to map user-level data across platforms, devices, offline, and online to get an accurate picture of media performance. Many companies have spent months if not years ironing this out. MMM is comparatively more straightforward as you can use first-order data without needing to know all the second-order, user-level data required by attribution.

So, while MMM is here to stay for a while, it has clear gaps that multi-channel attribution can immediately fill. With attribution, we have the benefit of:

  • Tactical, day-to-day management of digital channels
  • Shorter time to insights
  • Higher level of granularity of results
  • User-level insights
  • Integration with the digital buying process

For all the reasons above, multi-channel attribution and MMM continue to be used together. Increasingly, marketing analytics providers are seamlessly deploying them in a unified platform, commonly called UMIA, or unified marketing impact analytics. But, as marketers push to have more and more customer-level data, and as analytics and optimization engines adapt to these new data sources, the trend is clear: MMM hasn’t left the marketplace yet, but it’s been given its hat.

Matt Sato is a Senior Vice President in the New York office.

3 Tips for Effective Measurement

Using Avatars to Measure Biases, Credibility + Trust

Avoiding the Fake Research Trap