Marketing mix modeling: Going beyond traditional MMM

Deren Tavgac
February 6, 2023
3 min read

Marketing mix modeling (MMM) is an advanced analytics approach that companies use to better understand how their marketing efforts are driving key business outcomes, and more efficiently optimize their marketing spend.

Although data-driven companies understand that a mathematical, data science approach will consistently outperform more simplistic analytics (or worse, gut decisionmaking), traditional Marketing Mix Modeling approaches possess many limitations that have significantly impeded their adoption by brands. A successful Marketing Mix Model that will generate 15-20% improved ROI requires multi-data source collection, unification, processing, wrangling, modeling and reporting, which can take companies quarters or years to complete. As a result, Marketing Mix Modeling solutions have historically been surface level, manual and onerous (both in terms of long setup times, and cost), which has slowed adoption.

Thankfully, analytics operating systems like Redbird are ushering in a new era where MMM workflows can be no-code, automated, always-on, robust and maintained at a lower cost. This means companies are able to build and run workflows rapidly, with extensibility into unlimited new data sources and lightning speed.

What is Marketing Mix Modeling?

Marketing Mix Modeling (MMM) is a statistical analysis technique that helps businesses use an analytical approach to measuring and quantifying the impact of various marketing inputs on sales and overall business performance, and make better decisions around marketing budget allocation. By analyzing historical data, MMM enables businesses to understand the correlations between different marketing variables such as advertising, promotions, pricing, and distribution, and their influence on consumer behavior and sales outcomes. This comprehensive analysis empowers businesses to optimize their marketing strategies and allocate resources effectively to achieve optimal ROI.

In an age where cookies are being scrutinized, privacy laws are being enacted to make multi-touch attribution more challenging, and customer journey mapping is more resource intensive than ever, there still remain some reliable standards to analyze the performance of ad spend across key channels (digital marketing, offline, etc). Companies need to understand details of how far their ad dollars are going, their return on investment, and where to place their next dollar to maximize revenue or conversion. Furthermore, as omnichannel environments become more common, creating multiple avenues for customers to engage with organizations, it's more often the case that multiple avenues of advertising drive conversion simultaneously, rather than a single instance of an advertisement.

Media or Marketing Mix Modeling addresses these concerns by taking a statistically-driven approach rather than a user-action driven model. Essentially, all ad channels are considered together, and specific attributes of each channel are derived through best fits using a machine learning or AI model. These attributes typically consist of things like the delay between ad-spend and conversion, the maximum length of time over which a day’s ad spend may drive conversion, and how quickly conversion decays over time.  Crucially, since all channels are modeled simultaneously, interaction terms between channels are also considered.

How is Marketing Mix Modeling Done?

The modeling procedure is typically a four-step comprehensive measurement process. Let’s go through the steps assuming we’re using our MMM model to extract the channel-level impact of ad-spend on the sales of a particular item:

Extract the baseline sales profile 

Since sales are obviously affected by more than just ad-spend, we need to develop a model or framework to take into account non ad-spend related activity that may drive sales of our item. A number of approaches are viable here, from basic flat-line modeling all the way through advanced AI forecasting models. What we’re going for is an informed approximation of what the sales may have been in the absence of any ad spend. At Redbird, we’ve used both high-end AI forecasting models and very simple approaches like fixed-value baselines in the case of events like a new product launch. At the end of the day, a reasonable approximation of the sales expectation and how sales ought to change both intraday and intra-year is enough to get started.


Coined by Simon Broadbent in the 1970s, adstocking refers to the time-delayed and prolonged effect that advertising has on consumers. In order to properly model these effects, it isn’t enough to try to simply correlate ad spend on day X with sales on day X. Rather, ad spend on day X often impacts sales from day [X+N] to day [X+Z], where N is typically anywhere from 0-14 days, and Z can be as long as a month. To account for these time-delay effects, we incorporate an iterative process whereby various time-delays, decay parameters, and peak day effects are all modeled independently such that various channels may have different optimal values, within some predefined constraints. Once extracted, these values can be sanity-checked and further constrained as necessary to improve the runtime of the model and ensure viability for a given use case.

Return on Ad Spend (ROAS) Fit Curve Extraction

Once the adstocking parameters are fit for each channel, we can build scatterplots of the incremental revenue vs the ad-spend for that channel by day. These scatterplots can be fit by any diminishing returns curve, typically a logarithmic function or a Hill function. These functions can then be used to parameterize how efficiently a channel uses its investment, and when it becomes fully saturated. 

Each channel is independently fit and can be fit to different functional forms where necessary in order to maximize adherence to historical data.

Optimization of Ad Spend

Finally, once all the ROAS fit curves are derived, ad-spend can be optimized across different channels. For example, if we find that our typical daily ad spend on paid search is low on the saturation curve, another dollar of investment in that channel will lead to significant and steep returns on conversion downstream. If, however, the typical daily ad spend on paid search is nearly saturated, then there’s really not an incentive to continue to invest in that category, and it's much more effective to put our next ad spending dollar elsewhere.

The Redbird approach for optimization is highly dynamic and can achieve a number of separate objectives, including maximizing revenue or conversion given a fixed ad-spend, or finding an optimal ad-spend to guarantee a certain return on investment or advertising effectiveness. The Redbird team has also worked on cross-slicing within channels, including demographic analysis, channel whitespace analysis, and geographic location-based analyses. Constraints and insights can be generated in or across any of these slices, which can significantly boost targeted return, ROAS, and Marginal ROAS (mROAS).

Traditional Marketing Mix Modeling Challenges

There are multiple challenges with Traditional MMM modeling. To name a few:

  • Baseline Modeling – It’s crucial that any good MMM model measures incremental ad-revenue from a hypothetical environment where there was no ad spend. Effects like Covid, severe weather, holidays, intra-week effects, etc. all can drive incremental sales or conversion and, in an ideal state, should all be built into the baseline modeling approach.
  • Collinear Entanglement - It’s often difficult to untangle collinear effects, where multiple channels may have similar ad spend and create similar incremental revenue
  • Channel Significance – Often times, certain channels are experimental or novel, leading to minimal investment that may or may not drive significant action. Handling non-significant channels appropriately (especially if there are many) can also play a critical role in attributing incremental revenue to ad channels.
  • Causality – It’s been said many times that correlation does not imply causation, and rigorous statistical tests are required before we can have any degree of confidence in calling specific channels causal to conversion, rather than simply correlated. MMM models ought to restrict allowed values for derived metrics to ensure effects are downstream of any ad spend and fall within expected ranges.
  • Data availability – The approach is data hungry and typically requires 2-3 years of history to ensure enough statistical variation in each ad-spend channel. For a rough guideline, analyzing more than 100 channels requires 3+ years of history, while analyzing less than 20 channels may have some statistical viability with only a year’s worth of history.

Key Components of a Successful Marketing Mix Modeling Approach

  • Data Collection and Analysis: The first step in Marketing Mix Modeling involves the collection and analysis of relevant data, including sales data, marketing expenditure, market trends, and other external factors that may influence consumer behavior. Robust data collection and comprehensive analysis lay the foundation for generating meaningful insights that can drive strategic decision-making.
  • Independent Variable Identification: Identifying the key marketing variables that have an impact on sales performance is crucial in Marketing Mix Modeling. These variables can include advertising expenditure, promotional activities, pricing strategies, and distribution channels (e.g. direct mail, offline channels, digital channels etc). Understanding the relative impact of marketing activities on sales allows businesses to pursue optimal budget allocation and focus on the most impactful marketing initiatives.
  • Modeling Techniques: Various statistical modeling techniques, such as regression analysis, time-series analysis, and econometric modeling, are employed in Marketing Mix Modeling to quantify the relationships between marketing inputs and sales outcomes. These modeling techniques help businesses predict the potential impact of different marketing strategies and assess the effectiveness of their marketing campaigns.
  • Marketing ROI Analysis: Marketing Mix Modeling enables businesses to evaluate the return on investment (ROI) of their marketing activities by assessing the incremental impact of marketing inputs on sales revenue. By understanding the ROI of different marketing initiatives, businesses can prioritize and allocate resources to the most profitable marketing channels, thereby maximizing their overall marketing effectiveness.

Redbird’s MMM approach

Redbird’s approach abstracts away the traditional challenges into a customizable user interface that can be leveraged via a code or no-code approach, depending on the user’s expertise. For example, a baseline model can be provided as an input to the tool or one can be auto-generated for you, using AI or machine learning forecasting solutions. Collinear and causal effects are intrinsically handled in the underlying modeling constraints, ensuring that statistical uncertainties properly model the realistic properties of each marketing channel. Finally, data volume guidelines can also be provided to users, ensuring the model remains grounded and isn’t affected by one-off or long-tail events.

Redbird’s next-gen Marketing Mix Modeling platform uses a grounded yet flexible statistical model to derive the best possible fits to spend and conversion data for each independent channel. We’re able to provide an MMM offering both to power users, who want to code and leverage infinite flexibility in how the model is deriving channel-level parameters, while also providing a no-code offering for an easier out-of-the-box solution. 

By leveraging Redbird’s data operating system, marketing managers are able to get started months faster than existing or legacy tools and iterate on modeling multiple times per day if necessary, rather than the quarterly turnaround that is typical of the industry. Furthermore, non-experts are able to start generating insights even without deep MMM expertise. The no-code building blocks available in the Redbird MMM toolkit allow users to rapidly prototype and develop robust, accurate models for downstream analysis.

Through more automated, always-on, no-code approaches, an analytics operating system like Redbird can help customers generate >20% improvements in media spend efficiency and marketing performance across all touchpoints. Once baseline MMM models are in place, modern marketers can expand their understanding of business drivers using Experience Mix Modeling, which incorporates new data sources and identifies additional opportunities for business optimization across all customer journey touchpoints.

MMM Options

Companies currently have 3 options when it comes to implementing their MMM strategy:

1. Legacy MMM providers. Companies that specialize in MMM solutions and services

  • Pros: OOTB least common denominator solutions
  • Cons: Manual, 6-12 months setup, not easily customizable, company resources required for data collection, and very costly (seven-figure annual contracts on average)

2. DIY Custom Build.  Utilizing internal engineering / data science resources

  • Pros: No software licensing cost, assuming open-source and proprietary code is being leveraged
  • Cons: Significant time and resource cost investments. In-house engineers need to maintain data pipelines and models. (1 year+ initial set up, seven-figure annual resource cost including maintenance)

3. Analytics Operating Systems.  No-code, automated, always-on tools

  • Pros: Always-on (model can be run on demand), lower software licensing cost, capabilities democratized to non-technical users, fully customizable, rapid set up time and extensible into additional data sources
  • Cons: Reliance on a third-party vendor

With the emergence of always-on, automated, lower cost analytics operating systems like Redbird, MMM has become a no-brainer from an ROI perspective. Once baseline MMM needs are being met, given their flexibility and power, these tools also afford companies the ability to easily enrich their analyses with more granular data that can extend beyond basic marketing sources into broader experiential moments customers are experiencing across different touchpoints. We call this Experience Mix Modeling.

Introducing the Experience Mix Model - a full business CX optimization engine

Given the reduction in time and custom-coding effort required to ingest and process granular data from a variety of sources, an analytics operating system like Redbird is capable of helping marketing leaders expand MMM into a more robust, insightful view which we call the Experience Mix Model (EMM). In addition to traditional MMM data sources, EMM includes a robust set of customer insights sources such as surveys, ratings, call center conversations, customer journey touchpoints, reviews, ecommerce funnel metrics, A/B testing, supply chain, CSAT/NPS, post-purchase interactions, and many more. Through advanced data science modeling, always-on reporting and custom interactive web applications, teams can seamlessly understand which parts of the customer experience are driving business results, and optimize the customer journey accordingly.


MMM Modeling is a powerful and flexible tool for marketing teams that remains viable even in the face of privacy concerns and the removal of multi-touch attribution data from many data collection platforms. While its statistical nature obscures the journey maps from an individual user standpoint, there’s tremendous potential to identify which advertising campaigns and channels are impactful from a holistic standpoint. There are then many opportunities to further investigate why channels are so impactful, and how companies may better leverage their ad campaign spend in the future.