As a media mix modeling novice with a lot of questions, I sat down with global marketing agency Assembly’s Joel Coppersmith, Global Director of Measurement and Effectiveness, and Johnny Francis, Head of Data Science, to learn as much as I could about media mix modeling in a quick interview.
Coppersmith and Francis are building a media mix modeling tool, SCENE, that lowers the barrier to entry for businesses interested in gaining more insights about their media spend. (SCENE is built on top of a proprietary data platform called STAGE.)
Media mix modeling (often referred to as MMM or econometric modelling) has historically only been for large enterprises with huge media budgets. It can be extremely expensive, time consuming, and contingent on data quality, but AI, ML, and automation are helping alleviate the traditional barriers to entry for MMM.
Read on for expert insights about the MMM of the past, the MMM of the future, and more.
Media mix modeling is a technique used to determine the optimal allocation of advertising and promotional budget across various media channels to maximize the effectiveness of marketing campaigns.
Media mix modeling helps businesses draw connections between sales, revenue, and brand health and “explanatory variables” such as advertising, pricing, and competitor landscape.
A media mix model can also help businesses understand what their baseline sales would be without any promotion or marketing and how factors like the weather, seasonality, or the economy are impacting sales.
Joel Coppersmith (JC): To put it simply, media mix modeling is the process of drawing out the relationship between a number of variables.
So in the case of marketing, you're typically looking at something like sales, revenue, or sometimes brand health metrics like awareness and consideration, and so on. And then you want to understand all the factors—known as ‘explanatory variables’—that contribute to movement in those metrics. That would typically include things like the advertising you’re doing but can also include a range of factors like your pricing, distribution, your competitor’s pricing, macroeconomic factors, even weather for some businesses can be quite significant.
You’ve got a whole range of different possible factors for marketers. Obviously, there's a bunch of stuff you can't control (like the economy, the weather), but there's also stuff you do control (your marketing spend), and that would typically be a small contributor to the overall sales and revenue but a major discretionary expenditure for the business.
And so you're just looking to understand what the relationship is between these explanatory variables and your sales using the modeling. MMM also gives you a measure of baseline sales. So those are the sales you would have got anyway, even if you hadn't done any recent advertising, maybe because of the strength of your brand, perhaps from historical advertising, having a lot of stores that people are going to see, maybe changes to your category, you know, someone dropping out of the market.
One of the biggest advantages of MMM is that when you do get a measure of how many sales are being driven by marketing activities and the different types of media channels like TV, radio, online display, etc., that is then considered incremental, so that’s sales and revenue you would not have got without spending that money on media.
JC: I've only ever seen them being used interchangeably. At one point I thought, well, maybe “media” is supposed to be specifically about media as opposed to including other pieces of marketing, but not really. They’re entirely interchangeable.
Media mix modeling provides data-driven insights and guidance on how businesses should allocate their marketing budgets effectively, which increases ROI, improves strategy, and optimizes spend.
1. Measures the impact of all of your marketing activities, external variables, and the baseline of your bottom line
2. Provides a holistic understanding of channel performance, not just a breakdown of digital channels
3. Gives businesses data they can use for budget planning
Johnny Francis (JF): There is a wide range of factors that MMMs are used for, and I'd say probably the most valuable part is the measurement aspect. At its core, MMM is a really useful tool for measuring the impact of all your different marketing activities, external variables, and the baseline of your bottom line.
The main value you gain is a holistic understanding of your channel performance, and I think secondary to that is the fact that the measurement that you’re getting is incremental. Without this kind of model, you're left with some parts of measurement, but none of them really capture the big picture stuff in the same way that an MMM would.
So a counterpoint would be something like, multi-touch attribution in the digital space—what that would do is give you an understanding of how your channel mix breaks down across all your digital marketing efforts. But what it’s doing is attributing the entirety of the value of any given sale back to the different channels that are in your digital media mix.
Now the problem with that is that because it's not accounting for your offline advertising or the weather or the base sales you would have without any advertising, for example, it doesn't give you a truly incremental view of what those marketing efforts are doing. It can be very useful to do that, but MMM is much more rooted in providing the true value that all that activity actually brought to your business.
And there are a lot of other things MMM is useful for. With that data, you can then make better decisions around budget planning, particularly between your channel mix. Understanding which channels are performing or what works well at different time of year can be essential to how you want to plan your budget throughout the year.
Data collection for creating a model is one of the hardest parts of implementing media mix modeling and requires a decent amount (ideally years) of historical data to provide useful insights.
Models take into account historical performance data from a businesses as well as public data around weather, stock prices, the consumer price index, and more.
While the work required upfront to create a model can be challenging, it also provides businesses with a chance to streamline their data storage practices so the model can continuously improve and take in new data.
JF: In my experience, data collection, as in getting all the data that you need for the model (not just from media, but from the external sources and so on), is probably one of the more time-consuming parts of the entire process.
In terms of getting data, there are publicly available data sources that are very useful. So weather is a key one for some businesses, then there are also economic factors, stock prices, consumer price index—all that stuff is freely available. It's really gathering internal data that can take time.
Often businesses’ media spend is held in spreadsheets and is tucked away somewhere—and the other thing about MMM is you really do need like a decent amount of historical data to be able to understand the seasonal factors that are at play as well as multi-year trends.
If you were starting a media mix model and the provider asks you to give them channel spend for something like direct mail for the last five years, that data might just be sitting somewhere on a server in an old, poorly formatted Excel spreadsheet. Figuring out where that data is and if it is high quality enough to run a model on is one of the more time-consuming parts of the process.
Part of our vision for SCENE is the ability to help automate a lot of that data ingestion. So the first time you go round, gathering all that can be quite heavy lift. But once you've got it and you know where you've located it, and you understand where it comes from, the opportunity is there to moving forwards, say, “Let's organize this in such a way that this doesn't need to be so difficult in the future.”
And what we have with SCENE’s supporting platform STAGE is a massive data lake with an awful lot of media data in it that we get via API. The goal with these tools should be to over time help the client’s business to increase the level of automation they've got set up for their data to eliminate the more manual work in the future.
Media mix modeling can be time consuming, especially to do it right. Businesses all have different data, goals, and marketing strategies, so a one-size-fits-all model doesn’t really work.
SCENE leverages machine learning to help ease some of those challenges by using the data inputs to create hundreds of models quickly. From there, businesses can evaluate the models and fine tune the ones that are most in line with their goals and reality.
This drastically shortens the time-to-launch for a model by getting data specialists a pretty good model much more quickly than was possible in the past.
SCENE also makes MMM more accessible to the average marketer—its easy-to-digest dashboard replaces the traditional bi-yearly data dumps associated with MMM. Marketers can more easily pull data and react to marketing performance using data that’s updated monthly and packaged in a digestible way.
JF: MMM can be very time consuming, and it can definitely be time consuming to do it right. You can do it quickly, but it may not get you a model that makes sense or is accurate, so a large amount of the work that goes into it takes time when you’re creating a model the traditional way.
Part of the difficulty here is that every business has unique marketing needs: They have different data, different goals and sell strategies, so every model has to be fairly tailored. You can't just build one model and then plug and play into that for a different client—it won't work that way.
So the way that SCENE works alleviates a lot of those difficulties by using machine learning to essentially do the modeling process hundreds of times very quickly. Instead of building one model, you build hundreds. From there you evaluate and score those models. The model will continue to regenerate with inputs and feedback you’ve provided, and by the end of the process you have something that is pretty close to making sense.
This gets us closer to the part at the end where we have manual specialists go in and actually tweak the model and make changes that make sense with what they're seeing, what the clients strategy is, how they've been spending, etc.
That gets you to a really good final model without having to do like months of manual work to get everything up and running. That positively affects the cost of MMM because if you're automating a large part of that process, you’re not having to pay for like a specialist to set everything up.
In terms of lowering the barriers to entry for new people wanting to do MMM, it reduces the time, it reduces the cost, and it puts something that's good in your hands.
And lastly, a tool like SCENE helps make the data from the model more accessible for the average marketer. A traditional MMM delivery might be like a 200-slide presentation that goes through very intricately, like all the different effects that were noticed in the model and how all the different action interactions happened. But for a typical marketer, who just wants to make a few tweaks or needs to dive into one specific area of the data, it’s a pain to sort through these data dumps every few months.
SCENE has an interactive dashboard where you can go in very quickly and find the exact metrics that you need to do your job well. The outputs are much, much more streamlined and actionable.
JC: SCENE has a very media-focused approach, so it’s not about understanding every possible driver of the overall business: It’s about understanding how your media investment is working. The core visualizations Johnny mentioned allow you to make quick decisions and optimize your budgets in the moment.
As far as the data requirements go, they're not massively dissimilar to a kind of traditional MMM, but a traditional MMM realistically delivers data every 6 to 12 months, whereas this is happening every month with SCENE. Once you've built that model and you've got the data flowing into it regularly, you'll be able to refresh it on a monthly basis. You can make decisions a lot more quickly, and your model will only get better over time.
It's not just learning on data from a campaign you ran six months ago or a year ago, it is learning on data that you were running weeks ago. And so that was the reason behind the vision for SCENE and is why it runs the way it does.
You’re a good candidate for MMM if…
1. It makes economic sense for you to spend $100–300K on measuring the impact of your media spend
2. You care about marketing effectiveness
JC: The way that the market is going, there are a lot more good candidates for this than there were previously. The barriers to entry have come down significantly.
So you kind of need to be of a certain size, but not that big. It’s certainly not just for the people who have traditionally done MMM like McDonald’s or Unilever, it’s so much more accessible now.
Basically, it needs to make economic sense to you to spend $100–300K on measuring your media spend—if that money is best spent on understanding how to improve the value of the other money you are spending rather than just putting it directly on media, you could be a good candidate.
I think the other thing as well is you have to actually care about marketing effectiveness. You can get any number of metrics from any number of platforms—like every media buying platform will give you a slew of metrics—and they are all intended to help you spend money better within that platform.
But none of them tell you what the best way is to spend your money or to give you the best possible return across all your media. Even things like digital attribution don’t do that, but MMM does. So if you care about understanding both the immediate sales impact and the slightly longer term impact of the media budget, then MMM could be really useful for you.
1. It isn’t a marketing silver bullet, but with the barriers to entry for MMM lowering, it’s a powerful tool for understanding how all your media works together.
2. As long as you’re willing to get started and be open to learning, you can get a lot of value from MMM.
3. Just do it.
JF: Yeah, I was gonna make a joke about like, yeah, just e-mail me if you’re looking for an MMM provider.
SD: Might add that…I am the content marketer, after all, you know.
JF: You gotta have some marketing in there!
It's just really important to have an understanding of how all your all your media works together, and MMM is not the only thing in my opinion you should be looking at in isolation, but it does cover some really fundamental bases that everyone should know. With the barriers to entry lowering, it’s no longer the absolutely colossal global companies that can afford this anymore or make use of it.
It isn’t a marketing silver bullet, but it does give you a lot more understanding, and I would say that probably upwards of like 80% of the questions that we get asked about marketing measurement and understanding marketing activity would be solved by an MMM.
And that’s missing from a lot of people’s view on marketing measurement, I think.
JC: I think it’s important to say that if one of the things that holds people back is questions over how accurate MMM is that, yes you need to have an amount of logic there that makes sense for your business. I’m absolutely 100% not denying that you can have terrible models that are clearly nonsensical.
But the most important thing is that you’re willing to get started and understand that whatever technique you’re using, like literally any possible measurement technique we’re using—the most complex or simplest—they’re all wrong. But some of them are extremely useful.
And what you need to do is understand how to create a collection of useful techniques for your business. Just get started, figure it out. It’s doable.