Get in touch with us. We'd love to hear from you.
The Rebirth of Marketing Mix Models
Marketing mix models (MMMs) refer to econometric time-series equations that help quantify the impact on a conversion metric (sales volume, footfall, leads) from a series of drivers and environmental variables. The drivers usually are marketing metrics – organic and paid – performance and brand, while environmental variables tend to consider macroeconomic indicators, like weather data or event-based data (ex. weekend vs. weekday, Black Friday, etc.).
For example, a bank would be able to identify the main channels (like campaign videos, social activity or outdoor campaign) and events (a corporate social responsibility announcement, bank-sponsored sporting event, launch of a rewards program, increase in interest rates) that bring new savings account openings, so marketers can make well-informed, tactical decisions.
In the last decade, the explosion of ultra-targeting and the massive number of data points that can be harvested and amassed for each user turned a method that was selling insights at an aggregate level into a second-hand analytics toolkit. The ability to mine tens to hundreds, if not thousands, of interaction points for each user seemed to make MMMs obsolete. However, due to privacy concerns and increased regulations, MMMs are starting to make a comeback.
The Growth of MadTech
First, it’s important to take a step back and look at how we got here. Marketing technology (MarTech) has focused over the last 10 years on software and tools that can help marketers and marketing analysts gain a holistic, 360° view of their users and customers. Solutions like customer data platforms (CDPs), which are used today, allow a wealth of data to be made available for extracting actionable insights and producing ultra-targeting algorithms. For a CDP to work effectively, your advertising technology (AdTech) needs to be well architected with the correct tag management system, and all platforms need to be enabled to allow for the harvesting of the user journey footprints. While the wider MadTech (Mar/AdTech) ecosystem of solutions and players continues to grow, a good understanding of the tech stack and the “ideation-to-eyeballs” supply chain (agencies, media planning providers, retargeting services, etc.) is critical to being able to leverage the data flowing into your CDP — data that should convert into positive return on marketing investment (ROMI).
Since CDPs entered the market, data flows have been feeding the data platforms with hundreds of attributes on individual customers. Heuristics and matching algorithms have bridged the gap whenever the user journey was lacking or the sparse attributes couldn’t show a clear or certain linkage. This was especially important not only in order to coordinate the information trail along the user funnel but also to the omnichannel experience (as we know multi-channel marketing campaigns are more likely to have a higher ROMI compared to single campaigns). Best of breed, well-architected, robust MadTech stacks have focused on managing and optimizing bigger, better, more targeted omnichannel campaigns with the use of tags, cookies, pixels and mobiles IDs…until now.
Enhanced Focus on Privacy
Younger consumers, a new wave of consumer privacy requirements and increased regulatory pressure have ushered in an enhanced focus on privacy and data protection. With these market changes, advertisers are becoming more concerned with platform-lock-in risks, the uncertainty surrounding matching algorithms’ accuracy without cookies and IDs, and the difficulty brands may have in collecting and building first-party data pools. Given this new environment, brands and marketers need to rethink their measurement strategies and audience-building methods. In fact, "over half (52%) of those surveyed by WARC for the Marketer’s Toolkit said they are looking to find ‘new measures of effectiveness’, while 42% acknowledge the need to invest in new technologies to measure audiences."
Going back to our earlier example, the bank would have been able to link different data points and channels, which would make the audience building richer and enable the connection between potential customers and current customers’ omnichannel journeys (call center interaction, Facebook retargeting, Google search, mobile ads). With privacy concerns and increased regulations, this exercise of ultra-targeting is much more challenging.
Rebound of Marketing Mix Model
And this is why we are seeing the rebound of MMM, a data-driven statistical methodology that is privacy-friendly and can help analyze the diverse factors that impact the selected KPI. Traditionally, some of the limitations of using MMMs were the long lead times and difficulty in scaling; however, once the architecture, data models and processes are in place, the models can be run every few weeks.
A combination of MMM and brand drivers can help to understand channel contribution and set up a target ROMI. An adequate learning agenda allows you to assess campaign/channel lift and inform model calibration. Within the marketing planning cycle, an optimization process can use the current funding, achieving a predicted lift in marketing effectiveness by reallocating budgets among the advertising channels. Essentially, MMM helps break down channel contribution. Attribution and control/exposed experiments will help measure the lift, while the experiments also can aid in the recalibration of reach curves and the compilation of new priors for Bayesian inference. In the in-flight phase, programmatic advertising helps at an operational level.
The return in marketing science to MMMs is becoming widespread and even recommended by big tech publishers. MMMs enable you to not only overcome privacy and governance concerns since it only makes use of aggregate data, but with the right infrastructure and MLOps in place, it can function as an always-on model or ecosystem. Leads, sales, cross-sells or whatever key metric you are measuring can be refreshed constantly – even at an audience level targeting your market segments – and recalibrated by lift experiments and attribution model results. In practice, it’s a rebirth of the model that wasn’t gone, just underused.