How marketers can implement marketing mix modeling with Mark Stouse

In this episode of the Marketing Analytics Show, Mark Stouse, CEO of Proof Analytics, discusses everything you need to know about marketing mix modeling.

You'll learn

  • What marketing mix modeling is and when it’s a more useful method compared to MTA
  • What groundwork teams should lay before implementing this model
  • How to sell the idea of marketing mix modeling to the leadership team

Subscribe to the Marketing Intelligence Show

Learn from Supermetrics' experts how to use data to fuel growth and maximize the ROI of your marketing spend.
Subscribe to the Marketing Intelligence Show


Anna Shutko:

Hello, Mark, and welcome to the show.

Mark Stouse:

It’s a pleasure to be back on your show. Thank you so much. Looking forward to the conversation.

Anna Shutko:

Yes. I’m also looking forward to this conversation. And Mark and another colleague of mine, Evan, have recorded webinars. So, if you’re interested to see Mark in action, go to the Supermetrics website and feel free to check out the webinar. But today, we’re also going to talk about a very interesting topic, which is MTA and MMM, which stand for marketing mix modeling and multi-touch attribution, as two approaches. So, Mark, my first question to you would be what are the MTA and MMM? And how does MMM work?

Mark Stouse:

Sure. So, MTA is all based upon… It’s rooted in data. We call it touch data. This is the whole idea that in a digital world, every single time a customer touches or accesses a digital artifact, there is a record that is made of that. And that you can track that data and begin to understand a lot of things about that customer from his or her intentions to buy or not buy to how they kind of move through their decision-making process. And there’s a lot about MTA properly executed, completely executed that is really, really valuable, particularly if you’re trying to understand repeating patterns in a customer buying decision process. When you see a lot of people kind of moving in the same way, accessing your materials, your content in the same way, more or less in the same motion, and across the same sort of timeframe, that’s very valuable. There’s no question about it.

I think that the challenges to MTA and to touch-based marketing attribution in general… So, there’s also in addition to multi-touch, there’s first-touch and there’s last-touch, and there’s even a couple other minor ones thrown in, variations on that same theme. The problem with all of that, right, is that Apple iOS 14 and 15 and Google and more and more other private companies, platform companies, are basically saying, “No, you know what? We’re not comfortable with all this third-party tracking that’s going on,” which is essentially the essence of MTA or any kind of touch attribution. And so, they have made it, in the last year or so, significantly harder to collect that data. So much so that probably companies’ marketing teams that were using MTA have seen the amount of third-party data that they have access to drop by two-thirds, sometimes even more. It’s just pretty much eviscerated that effort.

The additional problems have been with EU legislation, California legislation, which further constricts the use of third-party data. From the very outset, the whole goal here is to understand cause and effect and to optimize your spending in light of what your customers are doing or valuing in the overall buy decision. And for all of its really good points, MTA, mathematically speaking, is not a basis for optimization. It is not a basis for allocating spend, time lag effects alone. Pretty much eliminate that opportunity. In other words, that means that you’re spending money, and it’s not having an effect right away. It’s having an effect at some much later time.

And also, it’s a multi-variable effect. We live in a multi-variable world. This is the whole reason why people say, “Well, correlation does not imply causality.” It’s because, at its heart, correlation is between two things. You’re looking for a relationship between this one thing and this one thing. And that’s not going to indicate causality most of the time because we live in a multi-variable world. We’re not living in a situation where one thing causes one thing. We’re living in a world where many things come together in a unique or almost unique set of relationships and effects to drive a particular outcome over a particular period of time. And that’s not MTA. MTA can’t do that under any circumstances.

So, this is what brings us to MMM. So, about 35 years ago, a number of very, very large CPG companies, Procter & Gamble most notably, began to use multi-variable regression analytics to explore all these relationships and to get to a sense of ROI and to get to a sense of optimization. And they really did it to a fairly well. And MMM, marketing mix modeling, has been a really significant part of CPG and retail marketing ever since. The problem has been not the math or even the data, it’s been how hard it has been to operationalize the analytics, so that you get a helpful insight and that you’re able to make a better decision than you otherwise would. That’s been the problem. So, to be more specific, MMM historically has been really expensive. So, you get kind of like one or two models for several million a year. So, that right there kind of tells you a lot about who’s able to afford it and who isn’t.

It’s also very slow. The modeling is very slow. The latency on the recalculation of the models is also very slow. It’s not unusual for models to only be recalculated once, twice a year. And so, even though in that model, there is a prediction about the future, it’s delivered so late that it’s impossible for most organizations to get to use that prediction to change or otherwise control their future, right, to get a jump on things, so to speak. It’s also very hard to understand. The outputs are very difficult to understand if you are not a data scientist. So, all of those things taken together, and by the way, this goes way beyond just marketing and go-to-market, it’s a problem that has been there for analytics across the board.

That’s really where we are today with bringing automation to MMM or to multi-variable regression analytics if you want to be technical about it. So, we’ve automated those, and we’ve strapped a really intuitive UX on the front that allows business teams, in other words, people who are not trained analysts, not trained data scientists, to immediately understand what’s going on and what they need to do. And the really cool thing about it actually and really what made it all come together, in the end, was that we discovered that when you automate these algorithms, things start to behave a lot like a GPS on your phone, in fact, actually exactly like the GPS on your phone.

And so, if your goal is to say, “Well, here I am. This is my place in the marketplace right now. And this is where I need to go. And I’m going to plot a course to get there.” But just like in my car, things happened around me. Headwinds and tailwinds and events change, and maybe there’s an auto accident. And so, the GPS is not only tracking what I do, but it’s tracking all the stuff that’s around me that relates to my journey. And so, sooner or later it will say, “Hey, Anna. There’s been a lot of reasons why this route is no longer the best one, and we want to reroute you over here. And you’ll be eight minutes late to your dinner instead of 40 minutes late to your dinner. And we’ll show you that you have arrived on time.” So, that is actually very much what Proof Analytics does and why this whole category of analytics is exploding right now.

Anna Shutko:

Awesome, Mark. Thank you so much for such an elaborate and really, really in-depth introduction to both MTA and MMM. I really liked how you outline the benefits of both. So, my next question here would be, so when the company has already decided that “Okay, we want to go with MMM,” what groundwork should the team lay before starting to implement it? I know you’ve briefly touched upon

this, but if you could elaborate on this a bit more.

Mark Stouse:

Sure. I mean, look, this is actually very much like the scientific method of inquiry. The scientific method of inquiry starts with a question. It’s a question that you wanted to get the answer to. And then, you develop a hypothesis. Right? This is what I think the answer to that question is. For marketers, the hypothesis is typically, “What are you doing right now?” Because whatever you’re doing, whatever categories of marketing you’re involved in, whatever channels you’re involved in, you’re doing that because you believe that it’s having a positive effect in some way. So, that is your hypothesis to the question that says, “What’s the value of what I’m doing?”

So now, you have a hypothesis to test, and the hypothesis generates the model. I’m simplifying it but not by much. It will tell you, in other words, how to test it. You know what? This is the model. These are the data sets that need to be attached to of the model in order to compute it. It’s fairly straightforward, particularly in such a digital world where we have access to so much data. The biggest issue is having a clear idea of which datasets need to be in that model. And so, that would be number one. But kind of even before number one is this whole conversation that needs to be had between marketing and the business leaders to find out what questions they really want answered. So, you’re going to need to really prioritize the questions. And so, that’s, I think, a really key thing. The last piece of course, right, is having the right data.

And fortunately, we live today in a golden age of data. We have a lot of data. It’s very easy to get more data, even with a swipe of a credit card, initiating a subscription. You can immediately get your data that applies to you, your company in a particular area, maybe all the way back 10 years, if it’s necessary. Right? So, it’s not hard to get the right data. The other thing that’s very cool about that is that the data that’s for sale is typically already been cleansed and all that kind of good stuff, so that simplifies matters. Supermetrics is actually a great part of this. Right? Really super important part of it because the ability to bring all your data streams together into a platform like Supermetrics, which then auto cleanses and auto harmonizes all that data.

So, regardless of what you do with it next, whether it’s putting it into Proof for automated analytics, or putting it into Tableau for visualization, or whatever it is, that data is all ready to go. It is in good, solid shape. It’s ready for you to use. To put that in perspective for just a moment, cleansing and organizing, and harmonizing data usually takes about 80% of a data scientist’s time. So, when you’re able to automate it as Supermetrics has, the value to that operational process is just gigantic. You’d be hard-pressed to overstate it.

Anna Shutko:

Awesome. Thank you so much for outlying a very clear algorithm on how marketers should proceed forward when they decided to implement the MMM. And I’m super happy to hear that Supermetrics is a useful product for these kinds of purposes. Another thing I was really curious to learn more about is the communication between marketers and the leadership team. So, you’ve touched upon this topic from the angle of the right questions people should be asking, but I’m very interested to learn more about how can marketers sell the idea of moving from MTA to MMM to the leadership team. Are there any specific arguments they can bring up given everything we’ve just discussed?

Mark Stouse:

Yeah. I mean absolutely. So, I would say that the biggest argument of all is that the business leaders want to know what’s working and what’s not working on a very dynamic basis. So, they realized that the world outside the company is constantly changing. And so, those changes have a lot of impact on the investments and the actions that the company makes. And so, if the headwind gets to be too strong, it could overwhelm something that the company believes is effective, in the past has been effective, and all of a sudden, it stops being effective. And if you don’t have MMM, you will never know that. You’ll never understand that that has happened quickly enough to make a change.

And so, again, this is where the GPS comparison really is spot on. If you didn’t have the GPS, if you just had a map, you could easily go way too far down a road before you realized that you had made a mistake, or before you realized that suddenly the road was closed, and then you had to circle back. Right? You lost a lot of time. With a GPS that doesn’t happen anymore. That totally saves you that wasted trip.

And so, that’s really what business leaders want. They want to get the most that they can, that’s optimization, and they want it re-optimized as frequently as necessary in order to minimize waste and maximize the upside. The other thing that MMM will give you as far as optimization is concerned, is it will identify a point of diminishing returns. And again, this is all contextual within what’s happening in the marketplace at any given moment. But a point of diminishing returns means that if you continue to spend money in this area past this certain point, you’re sort of wasting your money. You’re not going to get the same kind of benefits incrementally that you were getting for that money. And so, that’s an area that is really important to business leaders, not just in terms of marketing and go-to-market and things like that, but in every area of the business. They care about that.

So, in the end, summing this up, business leaders look at investment in a particular part of the company as an investment deal. Right? They want to see a business case made, and a business case then says, “Okay, if you give me this kind of money by, I don’t know, this number of months or even a year maybe or something, you’ll get this amount of money back in terms of whatever matters to them.” Right? It could be more revenue, more profitability, better cash flow, could be more people, more the right people hired. It could be all kinds of things, but there will be an ROI. There will be a value that has been created that would not have otherwise been created if they hadn’t made that investment. And that’s the key to it right there.

Anna Shutko:

Thank you. These are really, really solid arguments. And if we talk about starting to use MMM a little bit more, so about the process part, can you please share what are some of the common pitfalls when it comes to the MMM implementation? So, what should marketers be aware of when the company starts to implement MMM. What could go wrong?

Mark Stouse:

So, the number one thing is… I mean, so this is actually not new. This is something that software people talk about all the time and software customers talk about all the time. So, there’s the old saying about people, process, and technology. Right? And process and technology exist to serve people, not the other way around. And people are always the long pole in the tent, meaning they are the most critical component. So, the change management piece, the process enablement piece, the human being enablement piece. And so, I think that the most important thing is to realize that if your people are not successful with the process and with the technology, you’re not going to be successful, period. It’s just the way it is. Right? So, the most important thing, when you’re moving into an analytics-lead decision-making mindset is to help people wrap their heads around that.

Now, the irony is that we all do this all the time, but we do this very intuitively. We make analytical decisions, but we do it very intuitively. And we typically don’t use a process. Not really. We don’t recognize it as a process anyway. So, the key is to say, “Okay, you know what? Here are your goals. This is what we’re all trying to achieve.” And everyone nods and says, “Yes, I get that. I agree with that.” Okay. Here are some processes that are designed to guide you and not only in terms of the way you use the software but in terms of how you think about all this. And this is how we’re ultimately all going to be successful together, as opposed to here’s some really cool software, here’s your license, have fun. Right? That doesn’t work. So, there is a tremendous amount in the first, say, I don’t know, two months or so. There’s a tremendous amount of enablement that happens. And that’s part of the program. It’s indivisible from the program.

And then sooner or later, people either outsource it, the analytics, to maybe a small data science team using Proof or something like that. We have a lot of partners that do this kind of work. Or they take it inside, and they have an internal team that essentially does that, all that work, and collaborates with not only marketing and sales, but other parts of the business to say, “Okay. What are your questions now? What do you want to continue to explore?” So, the other thing that’s really cool, I think, and really super important to remember is that you cannot boil the ocean. So, you start with five or 10 questions that you want answers to. That’s five or 10 models that you’ve created, and you work on those, and you learn some cool stuff. And then, you say, “Okay. Based on what I’ve learned, I have these additional questions. And so, let’s go modeling those and figuring out what’s going on there.” And then, you just kind of gradually build it out.

And there’s always a point at which customers plateau for a little bit. They kind of hit a spot where they’re consuming a lot of analytics and kind of wrapping their head around it and getting more and more leverage on it. And then, all of a sudden they’ll decide that they want to do some new stuff. And so, there’ll be some new models created. There’s a slope and then a plateau, and then there’s another slope and then another plateau. And then, there’s a point at which most organizations kind of say, “Okay. You know what? We’ve gone as far as we want to go right now. That might change sometime in the future, but right now the body of analytics that we’re working with are totally awesome, and we’re very happy with it. We’re just going to kind of sit here for a little bit and really get expert about it.”

So, that’s the way I would answer that question. At the end of the day, we just have to remember, this is all about making people successful, which is a hard thing a lot of times for software people to remember. Because a lot of times, software people think of software as replacing people. And it really doesn’t most of the time. What it really does is it makes people able to make better decisions to do a job faster, to do a job better. But it’s still that person or that group of people doing that job. It’s just software is a major accelerant on what it is that they’re doing.

Anna Shutko:

Thank you so much for sharing all the useful tips. And now, if the audience would love to learn more, where can they find you?

Mark Stouse:

Sure. Absolutely. They can find me on LinkedIn. That’s pretty straightforward. I’m very active on LinkedIn. You can also find our company on our website, which is Proof Analytics, just like it sounds, .ai.

Anna Shutko:

Great. Thank you so much for coming to the show.

Mark Stouse:

Hey. Thank you, Anna. Great talking to you.

Turn your marketing data into opportunity

We streamline your marketing data so you can focus on the insights.

Book Demo