Whether you call it Marketing Mix Modeling, Media Mix Modeling or simply MMM, this type of marketing investment analysis has come back to the forefront of conversations and interest within the media industry, among advertisers and agencies alike.
In this blog post, the first of its series around marketing measurement, we provide a refresher on this long-standing measurement technique and explore how it has evolved in the past couple of years given the imminent demise of Multi-Touch Attribution (MTA) as we know it, and advances in machine learning and automation.
What does MMM solve for?
Given the proliferation of channels and messages that a consumer is exposed to, marketers are under pressure now more than ever to prove the value of marketing to the Board. In fact, “The CMO Survey” from August 2021 revealed that 59% of CMOs felt increased pressure from the CEO and 45% from the CFO to demonstrate the value of marketing in driving business results for their organization.
Marketing Mix Modeling does just that. This statistical analysis aims to determine the incremental impact of a brand’s marketing activities on its business Key Performance Indicators (KPIs) e.g. sales, web engagement, etc. It disentangles the business impact among:
- internal factors, which the organization has control over, and related to: product (e.g. product innovations, packaging changes), online & offline media channels across paid, owned and earned media, price & promotions, distribution (e.g. inventory, geo footprint), and brand equity
- external factors, which the organization has no control over, such as: seasonality, weather, macroeconomic factors, competitive activity, etc.
Along with allowing marketers to acquire a full view of the elements impacting their business KPIs, this analysis also provides them with forward-looking budget recommendations to inform their marketing strategy and maximize business outcomes such as revenue or profits.
Marketing Mix Modeling answers questions such as:
- How many incremental sales (or other KPI) were driven by each of my marketing tactics?
- How much was due to external factors I have no control over?
- How much was due to my long-term brand equity? i.e. consumers’ inherent propensity to buy
- How did the impact vary by product and/or geography?
- What is the Return on Investment (ROI) of my marketing efforts by channel/sub-channel, partner/platform, and creative execution?
- What was the impact of social media on my business, across paid but also earned?
- For video, how did different copy lengths or dayparts perform?
- For which channels have I hit diminishing returns (i.e. where spending an extra dollar will return less than a dollar) vs. channels where there is further opportunity to spend for extra profit?
- What is my optimal budget and how should I allocate it in the future for maximum returns?
This list is non-exhaustive but, given the broad spectrum of the analysis, MMM generally sheds light on any hypothesis that a marketer might have around a campaign (given the right data are available) and provides detailed and customized answers to all their burning marketing effectiveness questions.
How does MMM work?
MMM needs a comprehensive amount of data. Generally, at least 2-3 years of weekly data is needed around sales, marketing activities and external factors in order to set up a robust marketing mix model.
This can be a time-consuming process. However, advances in technology and computational power can simplify and speed up the ETL data collection process, enabling to collect and manipulate multiple data formats from varying sources. In turn, open-source database management systems like MySQL or SQLite can be linked to cloud platforms such as Azure or AWS to help manage, store and analyze large amounts of data.
With the right data collection procedures and infrastructure in place, future updates become even faster and less resource heavy, and easily performed at the advertiser’s desired cadence and given measurement goals and priorities.
The approach used in MMM is generally a multiple regression analysis measuring the impact of internal and external factors (the independent variables) onto the business KPI of interest e.g. sales (the dependent variable).
Although regular regression models might be used in some cases, multi-level modeling techniques such as the Hierarchical Bayesian model have surfaced as a better way to model very large data sets that cover many dimensions such as multiple products across multiple markets and sales channels. Not only does this allow us to identify potentially different responses to marketing across those dimensions but it also facilitates the appropriate alignment of marketing data to those dimensions within the same model. For example, promotions should be aligned to the store level whilst broad reach media is best aligned to market or national levels.
Whichever modeling method is used, for MMM to be a viable modern attribution solution, it needs to resolve three key issues:
- The ‘last touch attribution bias’ problem present in marketing channels that are used to support customer pull through and fulfillment, such as branded paid search.
- Last touch bias, known as ‘selection bias’ in statistics, arises when marketing channels are more of a means to purchase for consumers where their decision to purchase was already made. For many consumers it is only once they have decided to purchase that they then visit the website or search for the product. Consequently, a sizeable proportion of site visits or search are simply an artefact of the sales process. This creates an endogeneity or identification problem, leading to biased estimates of the truly incremental impact of those marketing tactics.
- There are various more advanced statistical processes to deal with these types of biases in the data but there is also a role here to use a test & learn in-market experimentation program to help better identify the true incrementality of these tactics.
- Accurately measure the long-term, brand-building impact of marketing.
- Short- and long-term marketing effects are different in nature; a short-term effect is temporary and lasts only as long as the marketing program itself, however long-term effects may build more slowly and persist for much longer. To accurately measure these impacts, we need to employ a dynamic modeling technique and explain long-term base sales as a function of changes in consumer perception and brand metrics.
- Provide the granular and near ‘real-time’ results that marketers have come to expect.
- It is often argued that MMM is too slow and lacks the necessary granularity to handle the ‘real-time’ attribution problems that solutions such as MTA purport to solve.
- However, MMM can provide similar learnings if it is based on higher frequency daily and hourly time series data – updated weekly to deliver rapid in-campaign attribution.
- Advances in technology and computational power can further speed up the modeling process, where supervised Machine Learning (ML) techniques allow the analyst to rapidly build a set of initial models for subsequent causal analysis.
To learn more about these issues, read our blog post.
MMM provides a holistic and data-driven view of the impact of marketing activities on a company’s bottom line with data tables and charts of outputs that can be easily repurposed by the client for any internal company meeting or report.
Outputs can come in the form of PowerPoint presentations but also dynamic dashboards that clients can securely access at their own convenience and that can be automatically updated as new data/models get generated.
Typical insights that MMM provides are:
Leveraging the data from the media mix models and related insights, one of the most valuable aspects of MMM for advertisers is that it can be used as a powerful marketing budget optimization, scenario planning and sales forecasting engine.
ML and AI-powered software with an intuitive user-interface can help clients set and allocate their budget across their portfolio of media tactics, products and geographies and assess the sales impact of different media mix scenarios given certain assumptions around external factors evolution.
Working hand in hand with your brand’s media agencies can help with strategic and tactical campaign decisions and supplement the media planning constraints already available in such tools.
Implementing MMM at distinct levels of expertise
MMM projects vary in size and complexity. As does the experience of marketers with such analysis.
If you are starting with MMM
For marketers embarking on this journey for the first time, starting small would be a good way of getting your feet on the ground. That could mean for example only starting by modeling a couple of key products at the national level.
Obtaining key stakeholder buy-in around this initiative is also a crucial part of the process.
- On the one hand that means securing the right (and potentially gradual) funds with Finance with agreed measurement objectives and metrics for success.
- On the other hand, it is important to secure buy-in from internal marketing and brand stakeholders and external agency partners as MMM insights might lead to changes to areas such as product, marketing resource allocation, etc. that might not be favorably viewed by some.
As initial learnings start getting disseminated across key stakeholder groups and data collection processes are being put in place, expand scope to cover additional products/SKUs, introduce geographical granularity such as DMA level, and even expand to international markets. Whichever makes sense in terms of your business objectives and what you are looking to learn.
If you have some experience with MMM but are looking to change vendors/scope
Ask yourself, what are the major pain points in the way MMM is currently being done in your organization that is preventing it from being the holistic marketing measurement and optimization solution that it is meant to be and how is it failing to answer your key business questions?
Be clear about that upfront with your new vendor and make sure to provide them with a comprehensive brief. A good vendor will be able to inform you of what is possible or not given, for example, data limitations and they will be able to back up their claimed ability to align with your needs.
Look for a solution that combines technical rigor and scalability, as well as a partner with in-depth marketing analytics experience and knowledge of your industry.
For more in-depth information about how to successfully implement MMM, read our blog.
To speak to one of our experts about your MMM needs, please contact us.