Media mix modeling (MMM): The app marketer’s handbook
“Forecasting via media mix modeling will return” was one of Our top 5 mobile app marketing predictions for 2023. Having made its initial splash in the 1950s, Media mix modeling (MMM) is not new, but its relevance has recently skyrocketed in light of the new era of user privacy in digital marketing. In fact, in late 2022, Meta announced it had seen an 80% increase in MMM adoption compared to the previous year.
With the introduction of Apple’s App Tracking Transparency (ATT) framework in 2021, marketers can no longer access user-level data on iOS unless a user opts into tracking (see iOS 14.5+ Back to basics guide). Similarly, Google is seeking to limit the sharing of user data on Android to third parties, reducing reliance on cross-app identifier data in an effort to strengthen user privacy in its Google Privacy Sandbox on Android. This is one key place where the privacy-friendly media mix modeling comes in.
At its core, MMM is looking at how your marketing budget was spent and the result of your spend and using that to inform your future marketing endeavors. Thanks to advancements using data analytics and machine learning, MMM is having a renaissance, which we hint at in our article: Media mix modeling: The comeback kid of advertising analysis.
In short, because MMM is not powered by user-level data but instead utilizes aggregated data from a variety of sources and channels, it provides a solid marketing measurement analysis fit for 2023 and beyond. Learn everything you need to know about mix media modeling in our guide below.
But really, what is media mix modeling?
Media mix modeling, also known as marketing mix modeling (MMM), is a statistical analysis used to ascertain and predict the impact of marketing activities on a business’ return on investment (ROI). This technique employs a top-down approach using data science techniques like multi-linear regression to examine the relationship between dependent variables, such as engagements and conversions, and independent variables like ad spend across channels.
The MMM framework allows marketers to include external factors, such as the effect of COVID-19, the rollout of iOS 14.5, inflation, etc, as well as offline and digital marketing efforts. The model will assign a value to the impact of campaigns across channels, which marketers can use to see the ROI of their marketing efforts, determine future steps—such as ad spend—and make marketing predictions for upcoming campaigns.
The media mix modeling ratio is:
1. The marketing channels in use.
2. The amount of money spent on each channel.
3. The previous campaign results and insights.
The ratio above is the base of an MMM framework, flexible in its ability to include other components. Today’s digital media mix modeling can handle plenty of data input, but it’s up to the marketing team to determine which variables to use, which will in turn ultimately influence the model’s usefulness.
10 questions app marketers can answer with MMM
With MMM, marketers can answer questions like:
- How many conversions did each media channel drive?
- What’s the ROI of each of my marketing channels?
- What’s the optimal spend level for each channel to maximize KPIs?
- Does the way a campaign is executed on a channel impact its performance?
(E.g. frequency of ad, creative, or targeting)
- Where should my ad spend go in the near future?
- What is the optimal mix of channels to reach a certain user segment?
- How do channels and geography impact my marketing effectiveness?
- What’s the role of earned, owned, and paid media?
- How to best optimize marketing touchpoints by campaign, audience, geography, timing, and publisher to maximize ROI?
- What is the impact of external factors on the total revenue?
Not bad, right? What’s even more impressive, is that an MMM can answer more business-specific questions, such as, “If I change X factor, how will my revenue be affected?”
Below, we’ve included an example graph created by an MMM, displaying the revenue contribution based on a certain mix of individual marketing channels. You can see it combines all marketing efforts to show the generated ROI compared to the weekly cost.
What data does an MMM need?
Traditionally, a marketing mix model was built around the 4Ps: Promotion, Price, Place, and Product. For mobile apps, however, not all of these aspects are as relevant (e.g., price of an app). While a major selling point of this model is that it’s flexible in the data points you can plug into it, we recommend including the following in your mobile app marketing’s MMM.
Monitoring daily ad spend is a must for mobile marketers. You should be able to identify the highest-performing channels, seeing which deliver the best user acquisition rates. For an MMM to be able to inform your overall marketing strategy, it will need your media spend data. Consider utilizing a tool like Adjust’s ROI Measurement solution to ensure your monetary data is accurate and unified across your digital channels before you feed it, along with your offline media spend data, into your model.
No, seasonality isn’t just for brick-and-mortar stores. This factor can certainly affect the success of your app’s campaigns and is worth studying. For evidence of the effect of seasonality, check out these articles: Unlock Q5 to level up your gaming app, Mobile app trends over the holidays 2022, and Touchdown! Mobile app usage ramps up on Super Bowl Sunday 2023.
For example, let’s say you know that mobile app usage hops up on Lunar New Year. If your app has the resources, you can localize in the markets celebrating and then take advantage of this holiday to promote your app by updating its metadata, icons, and in-app events to reflect the Lunar New Year celebration. By feeding your MMM with seasonality data, you’ll be better prepared to acquire and retain more app users during calendar events.
App store ranking
Undoubtedly, your app store ranking is a crucial factor in your app’s visibility in an app store. This aspect falls under the first of the 4Ps: “Promotion”, and the use of App Store Optimization (ASO), a specific tactic app developers and marketers use to promote their app directly in the app stores by increasing their app’s overall app store ranking.
ASO takes into account the keywords in your app’s description, content localization, your app’s assets and screenshots, and much more. As your app store ranking is something you will continually monitor and seek to improve through routine ASO efforts, it is an important component to include in your MMM.
For a comprehensive study of ASO, check out The mobile app marketer’s definitive guide to app store optimization.
Another factor to consider is your company’s press coverage. While the influence of PR varies by app vertical, if you think your brand’s press releases, bylines, and guest posts are making an impact on your app installs, then track it. It can help to look at your app’s daily active users (DAUs) for an idea of press coverage affecting the interactions with your app.
Naturally, you’ll want to include mobile attribution data in your MMM. This can be data from the self-attributing networks (SANs) you partner with like TikTok, Facebook, or Google Ads, data from the SDKs of your mobile measurement partner (MMP), first-party data (Yes, we still recommend Getting the opt-in), and more.
Explanation: Media mix modeling vs. attribution modeling
MMM and attribution modeling are different from a technical perspective. They do, however, address similar business needs around meeting objectives. Today’s app marketers are fixated on mobile attribution, through which they uncover their best users and channels by monitoring individual user interactions. In contrast, marketers are less familiar with MMM, which offers a more macro-level view, looking at all relevant outside factors (like seasonality, market conditions, competitors, etc.) and other aggregated data that impact the company’s ROI.
We’ve broken down the main differences between attribution modeling and marketing mix modeling in the table below.
In addition to using attribution as your marketing source of truth, we recommend topping up your marketing strategy with an MMM for additional insights. By monitoring attribution, you can course-correct campaigns in real-time for optimal results, and using an MMM, you’ll be able to properly plan your campaigns and forecast business outcomes. Combining both modeling types can be particularly useful for brick-and-mortar businesses that also have an app. We’ll dive into this more later.
Why use MMM
Benefits, limitations, and use cases of MMM
To gain a better understanding of what an MMM can bring to the table for your marketing strategy, let’s explore the challenges and opportunities in media mix modeling.
Benefits of an MMM strategy
As is necessary, a greater number of privacy laws and policies are being put into place worldwide. None of these impedes a marketer’s ability to use MMM. Marketing mix modeling is and will continue to be privacy-friendly—from GDPR to Apple’s restrictions on accessing the IDFA on iOS devices to Google’s upcoming Privacy Sandbox initiative reducing reliance on cross-app identifiers for Android devices. Many claim that this advantage alone makes MMM a future-proof marketing tool.
Gain an overview of your marketing ROI
When done correctly, an MMM can help you correlate different results to the marketing efforts that made them successful. In doing this, you’ll be able to understand historical data trends better, justify the ROI of your marketing initiatives, and more accurately forecast the outcomes of your future campaigns.
Make accurate predictions
In the same vein, one study revealed that brands that utilize data-driven simulations to inform their planning saw at least five times the growth of those who did not. By effectively plugging in previous data, you should be able to predict with precision future revenue and KPI results. This removes the guesswork from planning campaigns, as you can forecast campaign performance based on how you scale your ad spend.
Optimize budgets and campaigns
With the insights gleaned from your model, you can uncover which of your marketing channels are best-suited for ad spend and adjust your budgets to deliver maximum ROAS. Modern automated marketing mix modeling can provide insights in real-time to help marketers examine the success of a running campaign. Note: The details you can optimize, i.e., creative-level, depend on the level of granularity you provide your model with.
Limitations of marketing mix modeling
May seem complex
Let’s not forget MMM’s statistical background. Whether you have a data scientist on your team designing the right MMM for you or you decide to partner with one of the several media mix model companies to use their solutions, a copious amount of data is involved. However, once you have the right setup, your model should easily provide your marketing team with actionable insights.
Not built for causality
This model provides correlation, which is not equal to causality. While often, a MMM answers questions marketers have, as mentioned above, it may not always be able to answer the cause or reasoning behind a question. Nonetheless, well-built models can serve to sufficiently address this limitation enough to offer predictions and channel lift.
Needs time and money
Like most good things in life, setting up an MMM requires much of your budget and time. However, if built correctly, your media mix model should deliver results that will make it worth the resources you have invested.
Requires lots of data
App marketers and developers with a new app are unlikely to benefit from an MMM when launching their apps unless they use data from a similar app as guidance. The standard for these models is at least two years' worth of data because they require multiple years of historical data to produce their marketing averages.
Three MMM use cases
Let’s say you have a recipe app. You can employ an MMM to analyze how seasonality affects installs in different regions. In North America, the holidays of Thanksgiving and Christmas may prove mighty as more people are at home, and install your recipe app as they desire to cook for their families. This may also prove to be true in the Middle East, during the celebration of Ramadan, as well as in China during the Lunar New Year.
To investigate, have your model analyze data from the last three years by region using seasonality as a variable. You’ll then be able to see how much weight it plays on your marketing efforts' success and create a reliable forecast for your upcoming seasonal-based campaigns.
In this scenario, let’s imagine you have a finance app. While historically successful, finance apps, particularly crypto apps, recently experienced a downturn, also known as the “crypto winter”. Believe it or not, this downturn is something you can plug into your MMM. It is a macroeconomic factor that will influence your app and is, therefore, something to examine.
Lastly, envision you have a retail store with its own e-commerce app, selling clothes in-app and in-store. To gain a better overview of how to optimize your online and offline ad spend, you use an MMM.
You can learn if installs will increase by lowering your budget for direct mail and increasing it for connected TV (CTV) ads, specifically shoppable content. With the right data, you can predict the outcome of a campaign primarily run on CTV and whether or not this is the best next step.
For more vertical-related content, check out our Mobile app trends 2023 ebook, covering insights collected from 5,000+ apps.
How to set up your marketing mix model
Review your resources
First things first: You need to determine if you have the resources to build an MMM, likely utilizing either in-house or outsourcing.
In-house option: Data scientist
If you have a data scientist or general analytics team, see if they have the time and knowledge to build your statistical model based on the questions you’d like it to answer. As designing an MMM is quite technical, they should be familiar with media mix modeling tools and libraries. After building your MMM, your team member(s) should be able to interpret the results and run optimizations and forecasts for you.
Libraries to build your own model
While not directly associated with Google, the LightweightMMM, also known as the Bayesian marketing mix modeling library, was built by Google scientists as an open-source media mix modeling python library. Your team can use it to train your model to acquire channel attribution information and to predict the optimal budget allocation across your media channels. Access the Bayesian MMM library here.
Robyn MMM package
Created by Meta, this model is a semi-automated MMM open-source package that is powered by machine learning. Robyn offers intuitive model comparison so you can determine which model is best for you. It also has a budget allocator for marketers to maximize their ROI by optimizing their budgets. You can install Robyn here.
In 2022, Adjust was proud to be one of the select partners for the Meta MMM Incubator Program centered on leveraging Robyn to build a new solution. During this period, Adjust built our prototype called Budget Planner.
For guidance on the functions of the two frameworks above as well as insight on how they differ, check out Deloitte’s How-To Guide for Advanced Marketing Mix Models.
Outsourcing option: MMM Vendor
Alternatively, you can partner with experts from one of the marketing mix modeling companies to outfit your team with an MMM that fits your business needs. In addition to consultation and building the model, most offer automated marketing mix modeling software with forecasting and optimization capabilities. While outsourcing saves your team a great deal of effort, it will cost you.
Questions to ask your MMM vendor
Should you decide to outsource the building of your media mix modeling, it’s essential to thoroughly analyze a vendor before working with them. Ask them the following questions.
- How do you collect your data?
Your marketing mix will come from various sources and should be evaluated using an econometric model. If the vendor collects and standardizes the data manually rather than via an automated process, it will likely be predisposed to error.
- What inputs will be included?
This question will clarify whether or not a vendor knows the industry and what is best for your app marketing. You want to ensure the vendor isn’t limiting essential data and also isn’t adding meaningless data points just for the sake of show.
- How will you ensure the data inputs are accurate?
See if the vendor will run tests to validate the accuracy of the data. Without clean data, the statistical analysis of the model will not be correct, and therefore, the insights will be unhelpful or even misguiding.
- What level of granularity will be provided?
The goal of MMM is to take specific actions based on insights gleaned from your model. Should a vendor only provide channel-level analysis, you won’t be able to optimize any further than this. A media mix modeling company should be able to drill down and provide insights based on specific campaigns, geographies, and more.
Build your media mix model
Whether you decide to partner with an MMM vendor or build your MMM yourself, these steps will guide your model-building process.
- Determine your questions
- Collect data
- Build your model
- Analyze output
- Make predictions and plan
1. Determine your questions
As illustrated in our 10 media mix modeling example questions above, decide which informational gaps you’d like your MMM to fill. Consider questions around budget optimization, conversions by channel, campaign execution, offline activities, and any external factors you wish to observe.
2. Collect data
Round up the data—both the dependent and independent variables—from the various marketing sources that you’d like to include in your model. Bear in mind that it’s recommended to include data from at least, if not more, than two years to increase the data points that can be derived from the model.
Data app marketers should consider feeding to their MMM:
- Installs, sessions, retention
- Paid and organic activity (focus on impressions)
- Macroeconomic factors like seasonality, when relevant
- Ad spend
- Owned media (E.g., blog/website)
- Influencer marketing
- Email marketing
- Competitor intelligence
3. Design your marketing mix model
When you create your MMM, correctly specifying it is key. You want to ensure it not only answers all of your business-specific questions but that it also answers them correctly. This requires fitting the model to the data and setting the model’s parameters. It’s recommended to test it for at least four to six weeks before implementing it.
4. Analyze the output
After running your model for some time, you need to evaluate its ability to impact your overall business goals. You need to examine its output in terms of efficiency, effectiveness, and ROI.
To review your model’s efficacy, ask these questions:
- Is the model actionable?
- Can we use its outputs to inform our next marketing steps, such as frequency of ads, channels used, and ad budget allocation?
- Are the predictions trustworthy and driving incremental business performance?
Like most marketing functions, MMM is an iterative process. You’ll need to routinely validate whether it serves your team’s needs and perform marketing mix optimization accordingly.
5. Make predictions and plan
This final step is the reason you’ve gone through steps one through four! Now that you have built your model to fit your needs, it’s time to extract the insights that will inform your upcoming campaigns. This involves not only looking backward at the historical data visualized but running simulations to discover the best mix of your marketing tactics to meet your KPIs.
App marketers can utilize the output from media mix models to understand the big picture of their marketing efforts. For example, you can determine how often you should show ads to a segment of your target audience. You can also see which channels you should allocate more ad spend toward, determine what’s not working, understand organic uplift, and gain an overall view of your ROI.
MMM best practices for app marketers
1. Select the right data sources
Gather data from your various marketing sources, including your own app analytics, third-party ad platforms, and offline data sources. Make sure the platforms you partner with offer transparent and accurate reporting, providing data in the right format to be included in your Marketing Mix Modeling.
Two tips for app marketers:
- Go for the opt-in: While MMM does work well with aggregated data, working with compliant, user-opted-in first-party data will only serve to improve the accuracy of your model. Including data from your MMP like in-app purchases, conversions, engagement, and other in-app activity is recommended to best monitor the customer journey. The good news is that Adjust data shows opt-in rates are consistently on the rise.
- Take advantage of Datascape: Note that you can use Adjust’s Datascape solution to collect all of your digital marketing data in one place. Datascape guarantees your data stays clean and accurate, which is of utmost importance to ensure accurate results and forecasts from your model.
2. Differentiate video formats and platforms
Videos vary in format, purpose, and where they are hosted. Therefore, they should not be grouped together, but treated individually. For example, an ad on Connected TV will differ from a TikTok in-feed video ad or a product demo on YouTube.
We suggest breaking out your platforms and programmatic partners (if using) as individual line items and then, within those, separate the media channel for each one. Factor in viewability, watch time, and audibility, as each has a different influence on your marketing.
3. Test changes to identify patterns
The MMM framework allows marketers to be bold and test out hypotheses using statistical analysis. Don’t waste this opportunity to see, for example, how increasing spend on a certain channel while decreasing it on another could pay off big for your brand.
Forecasting is one of the best advantages an MMM has to offer. Consider developing monthly, weekly, and daily forecasts divided by your regional markets using historical data in what is also known as utilizing the power of predictive analytics. By doing so, you can create the most fine-tuned models for your forecasts. You can discover, for instance, how much ad spend is needed per channel to achieve a year-over-year revenue goal.
4. Make your measurement approach hybrid: MMM and attribution
Media mix modeling is not a replacement for mobile attribution. Rather than having the current mindset of “marketing mix modeling vs. attribution”, app marketers should see the latest technological advances in both as a sign that these approaches should be united to maximize your bottom line.
MMM proves its value by providing a top-down view of all your marketing efforts and allowing you to make accurate predictions. However, it does not let you drill down into the granular details of user behavior. This is where mobile attribution, with its bottom-up perspective, fills the gap as it lets marketers understand the individual touchpoints in the user journey with completely objective accuracy and reliability. Combining MMM with attribution is particularly valuable for companies growing at scale and those with both a physical and digital business model.
Example: CTV attribution + MMM
In our article, How to create a market-leading mobile app OTT campaign on CTV, we discuss why measuring impressions and engagements is critical to assessing your CTV campaign’s performance.
But, did you know you can input this data into an MMM to inform your overall marketing strategy?
You can take into account factors like seasonality: In the last two years, are more people streaming on CTV devices in the last week of the year? If so, what does your historical data show regarding your CTV campaigns? If you were to increase ad spend for CTV and decrease it for paid search during this time period, what would be the result? Your MMM could advise you if this action would pay off.
Despite being around for over half a century, MMM has come full circle and it deserves app marketers’ attention for its powerful potential. We believe the technology backing MMM will only continue to develop, and when paired with mobile attribution, can provide an insightful overview that tangibly links your marketing efforts to revenue.
As mentioned above, an MMM is not a replacement for attribution but both should be used together. Let Adjust’s mobile measurement and analytics suite feed your MMM with all of your digital marketing data and fill any granular gaps left by your model.
By digital data, we mean Adjust provides:
Mobile attribution: Measure channel performance, identifying best campaigns and creatives.
ROI measurement: Track ad spend, ad revenue, conversions, and LTV trends.
ROI for CTV: See the impact of your ads on mobile and CTV with CTV AdVision.
An overview with Datascape: View all your digital marketing data in one place.
Smart alerts with Pulse: Set customizable alerts for KPIs and anomalies.
We’ll help you gather clean, accurate data for your model, letting automation remove the manual and time-consuming labor. Plus, with real-time measurement data, you can optimize campaigns as they’re running for maximal success—down to the creative level.
We’re always innovating, staying ahead of the latest tech advancements. For example, our foot’s already in the door with Google, as a beta tester of its Privacy Sandbox on Android, and we’re currently developing a groundbreaking MMM solution. So, if you’re looking for a trustworthy mobile measurement solution, Adjust is here for you.
And that’s about it! We hope you found this guide to media mix marketing helpful. Should you wish to see first-hand how Adjust can transform your mobile app marketing with its data-driven insights, request your demo now.
Want to get the latest from Adjust?