Episode 48: Why Privacy-First Marketers Are Turning to Media Mix Modeling

In this episode of The Digital Clinic, Michael Wiegand, Director of Marketing Sciences at Wheelhouse DMG, walks through what media mix modeling now makes possible for med tech and healthcare marketers. 

Michael spent years skeptical of MMM. The price of entry was too high, the timelines were too slow, and the results arrived long after the decisions they were meant to inform. A new generation of open source models has made MMM accessible to mid-market brands, and what Wheelhouse built on top of those models has made it genuinely actionable: a tool for simulating budget decisions before you make them, not just accounting for what happened after. 

Listen & Subscribe:

Media Mix Modeling for Healthcare and Med Tech: What’s Now Possible 

Aaron Burnett: Welcome to “The Digital Clinic,” the show where we dig into what actually works in digital marketing for healthcare and med tech. The strategies, tools, and thinking that move the needle when the rules are tighter, the stakes are higher, and “that’s the way we’ve always done it” is no longer a viable answer. 

The measurement tools most marketing teams rely on are losing fidelity. Cookie deprecation is in full swing, and for healthcare and med tech, increased data privacy regulations have rendered most of the conventional targeting and tracking tools used by other industries too risky and off-limits. 

Attribution models may offer only a vague sense of what already happened and very little insight into what to do next. For years, media mix modeling promised a tantalizingly powerful solution to this challenge. But with model build costs typically starting at $500,000, time to build at six months, and model refresh times requiring about the same time as the initial build, it was decidedly out of reach for all but the largest and most well-funded companies. 

But some very capable open source models recently upended that paradigm, making powerful media mix modeling and its benefits much more accessible. Michael Wiegand is Director of Marketing Sciences at Wheelhouse DMG. He spent most of his career as a skeptic. But after building, running, and seeing the results of our own media mix modeling, he’s not a skeptic anymore. 

One reason? The first time we used it for one of our clients, the effort delivered a 72% increase in net profit while spending 15% less on media. Here’s our conversation.  

This podcast is sponsored by Wheelhouse Digital Marketing Group. Wheelhouse provides exceptional performance marketing for healthcare and medical device manufacturers. 

Every Wheelhouse client saw record performance in 2025, even after implementing HIPAA-compliant data solutions. Find out more at wheelhousedmg.com. 

Why CMOs Are Struggling to Measure Marketing Impact 

Aaron Burnett: So Michael, let’s talk about media mix modeling. When we think of the challenges facing CMOs and VPs of marketing, what is it that they have struggled to know that has caused us to focus on media mix modeling? And what does media mix modeling promise or deliver for them? 

Michael Wiegand: Yeah, I think that’s a great question. 

A couple of the things that I think have driven this latest wave of interest have really been tied to the fact that third-party cookie lifespans have been collapsing over the last several years. Things like the rise of Apple Safari’s ITP, Intelligent Tracking Prevention. A lot of our clients and a lot of folks in the industry are collecting more data than ever into their CRMs, bringing in spend data, bringing in impression data from all of these ad platforms where they’re running media. 

And a lot of that data is starting to eclipse the two-year period where it is viable for them. They’re able to detect all of their channels’ impact on the broader business and the business KPIs that they really care about. So I think that’s been, in my experience, the thing that’s driving this latest wave. 

Aaron Burnett: So let’s talk about the fundamental difference between what you can learn through attribution modeling, even very sophisticated multi-touch attribution modeling, even assuming that you have full fidelity in tracking with cookies. What can you know from media mix modeling that you could never know from multi-touch attribution? 

What Media Mix Modeling Reveals That Attribution Cannot 

Michael Wiegand: Yeah, I think the biggest thing that it reveals in our clients’ businesses is the direct, indirect, and joint impacts of all of their marketing channels and how they work together to create results. Even if you had a really dialed-in multi-touch attribution model, the maximum lifespan for something like that is around a 90-day window. 

And I think with media mix modeling, there are a couple things you can do that are much different. You can analyze the lag time that it takes for upper funnel media to impact downstream conversions, and that is a really key part of it. It also does a really great job of telling you where the next best dollar is spent in your marketing mix, and I think that’s the thing that so many media buyers are trying to solve for. 

Where’s my next big bet? 

Aaron Burnett: Yeah. If you spend X, what will you get? What Y will you get? 

Understanding Joint, Indirect, and Upper Funnel Effects 

Aaron Burnett: So let’s pause and dig in a little bit on joint effects and indirect effects because to me, this is part of the magic of media mix modeling. In attribution, you can see direct attribution or sometimes implied attribution from a bottom funnel channel, maybe a mid-funnel channel. 

It’s always been not just harder, but almost impossible to attribute anything with any degree of certainty to upper funnel channels, and it’s been absolutely impossible to attribute any credit to offline or traditional marketing like television or print or catalog distribution. So how does media mix modeling solve for that? 

Michael Wiegand: So there are a couple things that media mix modeling does when you’re ingesting data that makes it particularly adept at detecting all of those things. The first is it looks at impressions over time and the impact of those impressions. So you don’t even necessarily have to get a click-through to your website to be able to understand the channel’s impact on your results. 

We can look at just how these impression trends and spend trends in all of these channels relate to and affect the lag time of my conversions. And that opens the floodgates to understanding a lot of those upper-funnel impacts, when you’re doing things like connected TV buys, when you’re doing things like display, where the click-through rates are so low that you’ll never get a visit to your site from those channels. 

But it is seeding your message into the user’s and potential customer’s mind, and then when they need that thing, they understand that they’ve seen it several times in the wild, and they come back through other more lower-funnel channels to get there. So I think media mix modeling does a great job of framing that entire journey. 

Aaron Burnett: The other thing that is very powerful here, and we’ll get to this in a moment, is that the model and the platform on which we’ve built our solution is an inference model. It is exactly as it sounds. You’re using a model that can infer the effect of something where that direct effect is not at all explicit in the data, so that you can see that when there’s a traditional television buy in a particular market, you see the downstream impact, even though it’s not at all explicit in the data. 

Michael Wiegand: Yes, absolutely. And in a landscape where so many CMOs and CFOs are making decisions based on outdated attribution approaches, I think this allows us to really educate them: we can actually detect the causality and the causation of a lot of these upper-funnel channels that we were blind to before. And it’s really opening those doors for us. 

Aaron Burnett: Yeah. All right, media mix modeling has been around for a long time. It has, though, for the longest time, been the domain only of the largest brands and the largest agencies. It certainly was out of our reach. Historically, it was a half-million-dollar price of entry to even get started. 

So, tell me a little bit about the historical context or dynamics around media mix modeling and what’s changed now. 

Why Media Mix Modeling Was Out of Reach for Most Businesses 

Michael Wiegand: Yeah. I would actually consider myself, for a long time in my career, a skeptic. And I think partially it was due to just the unapproachability of media mix modeling for mid-market and SMBs traditionally. 

It was for enterprises and enterprises only. You had to be not only able to spend a half million dollars to build the model and to train it, but also your media spend in all of your channels had to be astronomical in order to detect any effects in the model. And that’s not even counting the amount of time investment that you had to put into these old models to get them to work. 

On average, it would take anywhere from three to six months to build a model. By the time you had any good data from that model and you could refresh it, the business and the industry had already moved on. There are new platforms to buy on. There are new problems to solve in the marketplace, and you were none the wiser for having done all of that. 

So mostly what it told you was… 

Aaron Burnett: …what you should have done six months ago. 

Michael Wiegand: Exactly. Yeah, if I had my crystal ball and could see out in front of those things, then maybe I would have done something differently. But yeah, the time investment and the monetary investment were astronomical, and it didn’t move at the speed of our clients’ businesses. 

How Open Source Models Changed the Equation 

Michael Wiegand: So I think now what’s changed, and we can dig into this a little bit more, is there are a lot more very powerful open source models available, and we can select one of those and tailor it to our business needs and really accelerate how fast we can get to insights. And we can do it for a fraction of the price that it used to cost us with outside vendors or other agencies. 

Aaron Burnett: Yeah. I know our interest was piqued initially when a client shared with us that they had gotten a proposal from an outside consulting firm to develop a media mix model for them, and the price was $800,000 and it was going to take just short of a year. And so we started looking at this about three years ago at this point. 

Tell me a little bit about what we found when we initially evaluated the market and the options, and where we’ve ended up. 

Michael Wiegand: I think in 2023, predating my time at Wheelhouse, there was a lot of investigative work that had happened on the team that I inherited around third-party vendors to do this. 

And in the marketplace, there were a few more approachable options. In 2023, we investigated a third-party vendor to build a model, and it took six months to build. It cost us around $75,000 to our business to produce that model, and it did produce some good signals. But I think the economics of building with an outside vendor and the fact that we didn’t really have very granular control over the model’s parameters is why we decided not to continue with them. 

After that, in 2024, I think we did some due diligence and looked at some other third-party platforms to maybe do this, and it was more around $150,000 a year. And again, the controls of the model were still very much a black box. You didn’t quite know the mathematics behind it. You didn’t have a lot of control variables that you could set to really impact the model’s accuracy. And we realized at that point that proprietary platforms were not the route that we needed to go. 

So more recently, beginning of 2025, there have been some very powerful open source models that have become available, and it’s made it so that anyone that has some technical know-how can grab those models off the shelf, tweak them to their liking, and really build some accuracy and tailor it to the clients and the businesses that they work with. And I think that’s what’s really been our evolutionary path in media mix modeling. 

Aaron Burnett: So, tell me a little about where we’ve landed and what we’ve built. 

Why Wheelhouse Chose Google Meridian for MMM 

Michael Wiegand: We landed on the open source solution, Google Meridian. And again, this is after waves of researching outside platforms and looking at the pros and cons of each of these things. But in early 2025, I believe it was January of 2025, Meridian came into public view. 

And I think the things that attracted us to it were the fact that you could control for things like organic search volume and query volume and understand that as a baseline. We know that our media doesn’t work in a vacuum and that our prospective clients and prospects don’t just engage with paid channels. 

So we need to be able to control for those organic pathways and answer the question: what would happen if our clients spent nothing on media at all? And then use that as a starting point and layer our channels on top of that. So Compass, our proprietary marketing data warehouse offering, is privacy-first and security by design, and it was very important to us that any open-source model that we adopted could play nicely with that infrastructure. 

We are able to sit our Meridian models on top of that Compass data that we have pieced together. It’s data that we’re aggregating from so many different marketing channels that our clients are involved in. We’re bringing in Salesforce information, and we’re doing that in a privacy-first way. 

And it was very critical to us that we could bake our media mix modeling into that whole ecosystem and have it play nice. 

Aaron Burnett: Meridian, the base case for Meridian, is a model that tells you quite a lot about what has been happening. And it will show you, again, those indirect and joint effects looking in the rearview mirror. 

We got that as we went through the heavy lifting of implementing. What have we built on top of Meridian, though, that gives us the capacity to, as you put so well, help our clients understand where to spend that next dollar and what they’re going to get? 

What Wheelhouse Built on Top of Meridian: Scenario Planning and Forecasting 

Michael Wiegand: Yeah. There are a few things that we’ve adapted the model for and been able to bolt on, and one of the big ones is scenario planning. 

Our clients come to us with different budgets in mind, from a quarterly outlay to an annual outlay, and they really want to understand with a high level of fidelity: if I make this investment, what am I going to get in my KPIs? What am I going to see downstream? What’s my cost per KPI going to be on all of this? 

And so we needed a way to not only make educated guesses about what we would have done in the last 30 days if things were ideal, but also to look forward, 90 days, six months, a year, and be able to draw pretty accurate conclusions about that with a high level of certainty. And we were able to add some libraries to Meridian’s open source stack and really be able to take the model’s recommendations and apply them going forward, based on what we anticipated spending over the next eight weeks, six months, or a year, and really be able to understand with a high degree of confidence what we would get in reality based on that. 

We do have some rigor around back-testing that. Once we are creating the forecasts out of Meridian, we’re able to look at the actuals as they come in, compare them to our model recommendations, and make some fine-tuning adjustments and make sure that the next subsequent runs are smarter and smarter every time as a result of it. 

Aaron Burnett: That’s actually a really important distinction. Again, a distinction between conventional analytics or even business intelligence and attribution modeling versus media mix modeling. Those former are static. They tell you what happened, and that is what happened. When looking in the future, when looking at media mix modeling, again, it’s an inference model. 

It’s also a model that learns iteratively. And so as we implement it and more and more data is added and that data is consistent with recommendations, there are natural algorithmic adjustments that take place within the model that fine-tune and make subsequent recommendations all the more accurate. 

Correct? 

How Bayesian Inference Powers Model Confidence 

Michael Wiegand: Yes, absolutely. It uses Bayesian inference. That is the core math behind Google Meridian. And what that’s doing in the background as you feed it historical data is it’s running hundreds, sometimes thousands of simulations based on the media spend that you had, and trying to understand the range of outcomes that would have happened in the past and that will happen in the future. 

And it’s giving that information to you at a very high degree of confidence. So just to frame this, we have not made a recommendation based on data yet that has been less than a 90 percent confidence interval. So that’s how accurate the model has to be in our mind before we’re going to start making decisions on tens of millions of dollars in ad spend based on that. 

We’re not going to blindly shove things into a model that we can’t hang our hat on. So I think that’s one key piece in all of this, is that Bayesian statistical analysis gives us a very high degree of confidence, and it helps us lean into the model gradually and understand whether or not our results are following various saturation curves. 

Aaron Burnett: So, let’s talk about one of the first applications or implementations of media mix modeling for a client. Tell me about the client. Tell me about what we had to do to build their model, and then we can talk about results. 

Building the First Client Model: Data Inputs and Requirements 

Michael Wiegand: Yeah, absolutely. So, the first thing is the inputs. When we work with clients, we have to have two-plus years of weekly historical data in all of the channels that they want to be modeled. 

You do have to have a substantial amount of historical data for this to work appropriately. And you have to have things like impressions, clicks, and cost, all that stuff that you would expect. But then you also have to be able to segment down to particular campaign types within each channel. So when we give clients a recommendation, it’s not just, give this much more money to Google, give this much more money to Meta. 

It’s, “Shift this much budget into PMax within Google, shift this much more money into dynamic product ads with this particular channel.” So, it’s very granular down to campaign types, and so we needed all of that back data from our clients at a base level. The other thing that we needed was to look at non-digital channels and try to understand and wrap our brains around the external factors that are impacting this client’s business. Things like market demand, competitor activity and movements, changes in messaging, seasonality, and geographic data as well, being able to break down our results historically by things like state or metro area. 

And really the last thing that went into it was direct and organic traffic. Again, that baseline of what would I have gotten in the marketplace if I spent nothing on media at all. Once we had all of that picture together and our client was fantastic in working with us and getting all of those offline data sources injected into the model. 

Once we had all of that together, we started doing our first few runs with this. And just to give you a perspective, our first model run on this was a 94 percent confidence interval on our model, and our margin of error was under ten percent. And again, that’s something that we just set as a rule for ourselves in any model that we run. 

But particularly with this client at the beginning, we wanted to make sure our margins of error were well under ten percent before we would lean into any of the recommendations. 

Aaron Burnett: The inputs and the sophistication of these models, and this model in particular, you alluded to some of this. In this case, we were dealing with a dual brand, actually a retail client, with a highly seasonally dependent product catalog. And so we integrated weather data at a national level down to geo. 

We integrated catalog distribution data and inside sales data in this model as well, correct? 

Michael Wiegand: Yes, absolutely. So part of that inside sales data was also things like their call center. They have folks on phones that are making outbound calls to vendors, and we know that some of those vendors are looking at the website. 

So how does that impact the call volume that they’re receiving? And that was a big piece of all of this, roping together all of those things that could impact those online decisions and the online decisions that could impact the offline piece of the business as well. 

Aaron Burnett: All those things that historically a VP of marketing or CMO would wonder about. 

“We had a lift in conversions in a particular area. I wonder if it was the catalog, or I wonder if it was something else.” 

Michael Wiegand: You can know now. You don’t have to just lean on anecdotes. This is tying a clear connection to all of these things. 

Aaron Burnett: Let’s skip to the results. 

We implemented these recommendations, and then let’s back off, and we’ll talk about the recommendations and what maybe was surprising and maybe even a little bit disconcerting about the initial recommendations. 

Michael Wiegand: God bless our digital advertising team. I think they were the first recipients of these recommendations before we trotted them out in front of the client, and they were a little skeptical. 

They were a little bit worried, let’s be real. I think they gave us a lot of caveats about what they were seeing in all of these channels on the ground, and that actually informed the way that we leaned into the model at first. So we set up some recommendations. It was earlier on this year. 

We talked to our internal teams about it. We came up with a plan of attack for how we were going to implement in each of these channels. But I think a few of the things that jumped out to us and gave us pause was the impact of upper funnel awareness in Meta. That was something that traditionally our multi-touch attribution data came back telling us was not producing a really great ROAS. 

It was concerning for us to potentially lean into this more. We were actually in the process of cutting budget out of those campaigns before the model results cropped up. But what the model showed us with those indirect and joint effects of that Meta spend really gave us confidence that there was runway. 

And as we leaned into the awareness plays with Meta, that actually had some virtuous cycle downstream effects on what we were doing with Google brand. The fact that people would see an ad on Meta and then come through on a different channel and convert, and being able to demonstrate that connection and lean into the model with more spend there and see the results actually get better as an outcome of that, was incredible. 

Aaron Burnett: We were all nervous because the model suggested what would be intuitively the opposite of what most digital advertising managers would want to do. So the model said, basically, “You’re at saturation at bottom funnel in some of your bottom funnel campaigns. You should actually be investing in upper funnel campaigns and awareness-building campaigns,” where, as you said, the results are not at all evident in the data that we would conventionally use. 

So we made that shift and, as you said, we worked into it, did a little bit, and oh, that worked, and then a little bit more. And ultimately, what did we drive for that client? And by the way, I should stress these were budget neutral recommendations. So for those who might have had experience where the ethos is the agency is trotting out recommendations because they want to tell you that you should double your spend or that sort of thing, that’s not what we did. 

This was budget neutral. And what did we drive? 

The Results: 72% Net Profit Increase on 15% Less Spend 

Michael Wiegand: Yeah, we ended up driving a 6.3% increase in revenue growth for this client, and not only was the revenue growth important, but we leaned in really heavily and tilted our strategy into high margin campaigns for this retailer. And so that 6.3% revenue growth actually ended up driving a 57% increase in gross profit during that time that we were leaning into the model and a 72% net profit increase. 

Not only did our recommendations have a hand in telling them to lean into awareness, but for their bottom of funnel channels, it really helped us key in on the types of campaigns that were going to impact not just top line revenue, but bottom line profit for our clients. And that’s the thing that I think has been a game changer with all of this. 

And as we leaned into the model more and more, we didn’t just stay budget neutral. We actually ended up spending 15% less in the window of time that we made the recommendation. So not only did we generate more of our client’s KPI, but we were able to save them money on their marketing spend and get better results. 

Yeah, that’s been amazing for us. That’s the full holy grail for digital advertising. 

Aaron Burnett: We’ve touched on this a little bit. Is media mix modeling a replacement for multi-touch attribution or incrementality testing? Is it a complement to them? Where does it fit in the pantheon of the BI and data analysis tools that we have available to us? 

How Media Mix Modeling Fits Alongside Multi-Touch Attribution 

Michael Wiegand: Yeah, I think it’s very complementary. 

It’s not at all a replacement. And I think the fact that you have to have that prerequisite of two-plus years of historical data means there are some clients and startups that are not going to have that amount of back data on hand. So we need to use approaches, more traditional multi-touch attribution approaches and incrementality testing, to help them drive results in the short term while they prepare for media mix modeling in the future. 

So I think these approaches can be very complementary. Media mix modeling is very much more strategic in nature, and it tells you how to shift your overall mix around. And multi-touch attribution can be a lot more tactical and can be a lot more involved in your day-to-day. 

So we’re not telling our clients to ignore these signals if they have them, but to temper them when they have both available. Be able to look at both sides, make an educated guess, and move forward with all of that. 

Aaron Burnett: Yeah, that makes sense. We have alluded to the fact that traditional media mix modeling took six months to build usually and another six months to refresh each time. 

It was very expensive. Let’s contrast that with what we’re able to do now. There still is an initial model build time. There is a period of time when we are gathering data. What sort of period of time is required there? And then once we build the model, what utility and frequency is available to us? 

Build Time, Run Frequency, and What a Mature Program Looks Like 

Michael Wiegand: Yeah, that’s a great question. I think our initial model builds in every case that we’ve applied to our clients so far has been somewhere between six to eight weeks. It’s not easy to get our hands on a lot of this inside sales data and work with other media partners like TV and radio buys and all of the offline distributions. 

Those typically come in on a quarterly basis. It’s not immediate in that sense. But it’s becoming weeks and not months to be able to implement something like this, which is really key. And then once we have the models stood up, we can run them as frequently as we need.  

The first client that we mentioned is actually in the middle of their busy season right now, their peak season as a retailer, very seasonal business. And we can run the model now every two weeks to be able to fine-tune as we’re spending. We can run it every two weeks, whereas for some other clients, we may only run it monthly. 

So it gives us that latitude to be able to run it with a lot more frequency, run it with a lot more fidelity, and to be able to make more educated guesses about shorter windows of time in the future. 

Aaron Burnett: I know we’re running it for other clients as well. Share what an ideal end state is for a mature media mix modeling program for a client. 

What’s the cadence? What are the outputs? How do those outputs change or improve over time? 

Michael Wiegand: I would say the cadence for most clients that are on a sophisticated program is going to be a monthly run. We’re going to take stock of what happened in the previous month, add that to our two-plus years of baseline data, and be able to rerun on a monthly basis and really come up with ways that we can adjust the model. 

And what’s lovely in having both the rearview mirror reporting of the model and the scenario forecasting is as we lean into a particular scenario, we can back-test that against our most recent 30 days of performance and see how close we were to the ideal spend according to the model. And the closer that we get to having to make zero shifts in our channel mixture, the better obviously. 

We want to fine-tune and be able to predict that with a high sense of confidence. The other thing I would say is you are naturally folding more data into this model over time, and I don’t just mean the most recent data from your existing channels. New channels too. So as you are testing into new display platforms, programmatic, as you’re doing connected TV buys and leaning into those things, you’re adding new channels to the model too, and you’re seeing how those play with all of your existing modeled channels. 

So I think that’s a key portion there. And then the other thing I would say too is you’re making budget decisions with a lot more clarity, and you’re sticking to those as you lean into the model more. It’s giving you more and more confidence to be able to say, “We can spend X and get Y.” 

Aaron Burnett: As we’re talking with current clients and prospective clients about media mix modeling, what reception do you get? Is there enthusiasm? Is there skepticism as a hangover from media mix modeling as it existed for years? And what sort of objections, if any, are you hearing? 

Common Objections Around MMM 

Michael Wiegand: I think the main objection that we’re seeing from clients that we’re talking to about this, and prospective clients that we’re talking to about this, is that they feel like their data house isn’t clean enough. 

Their baby is ugly. They don’t have enough to feed a lot of this into a sophisticated model and have it produce good results. And I think our Compass technology really helps them solve for that issue. We are able to aggregate data, clean it, and transform it for them, get it into a HIPAA-compliant environment, and then we’re able to run our media mix modeling on top of that. 

So it really dovetails nicely with our other offerings in the marketplace and helps us get to these kinds of results in an approachable way for those clients. I think, too, there’s a little bit of skepticism about Meridian being a Google product. It’s open source, and I think folks are going to say, “Is it going to tell me to shift all my money into PMax?” 

And that’s actually not been the case. In the client we mentioned earlier, it told us to take money away from PMax. So I think those objections are natural. It’s a Google product, they’re going to tell you to spend more on Google platforms. But that has not been our experience at all in running with this. 

It’s really been channel agnostic. 

Aaron Burnett: And as you alluded to earlier, that’s not been our experience, and one of the things that we vetted so carefully when we looked at all of these models is the actual math, the calculations, and the algorithms behind them, so that we could see that there was not bias. 

And for each of these models, we tweak and tune. And even if there were bias, we could control for that. 

Michael Wiegand: Yeah, absolutely. That’s another thing I didn’t mention earlier, but a mature client is going to be adding more and more control variables over time into their model. Controlling for organic search query volume is only step one. 

As you’re getting more sophisticated with your understanding of all those potential outside impacts, the weather, your catalog drops, your direct mail campaigns, all of these things start to become a fixture in the way that you run and rerun. 

Aaron Burnett: So, I guess one more question. For current clients using Compass, what does it take for them to start to take advantage of media mix modeling? 

Michael Wiegand: Very little, honestly. We have a lot of these digital data sources tied up and wrapped up in a bow in Compass and we’re really able to deliver reporting that is holistic on all things digital. As they are able to turn the keys over to us on things like CRM data and the offline data that they’re doing, that allows us to get to results a lot faster with something like media mix modeling, as we’re not having to piece together everything. 

We’re only having to add a couple slices of the pie to get it back to where we need for model confidence. Yeah, again, it’s a matter of weeks, not months, that we can get this up and running for clients. 

Aaron Burnett: This has been a really helpful conversation and very interesting. Thanks for taking the time, Michael. 

Michael Wiegand: Yeah, of course. Thanks, Aaron. 

Closing Takeaway: Why the Team Trusted the Model Over Attribution Data 

Aaron Burnett: When Michael presented media mix modeling recommendations to our digital advertising team, they were understandably nervous. The model was telling them to do the opposite of what they planned. They were already moving to cut Meta awareness spend based on attribution modeling. 

But with a more complete view of not only digital channels, but weather data, catalog distribution, inside sales, and call center data, media mix modeling came to a very different conclusion. The team had to decide which one to trust: attribution modeling, which was steeped in platform data, or the more sophisticated and nuanced analysis that told a very different story. 

Thankfully, they trusted media mix modeling. Total revenue grew by 6.3%, but the real joy was in margin. Gross profit increased 57%, net profit went up 72%, and media spend decreased by 15%. This is the holy grail for digital advertising, and it’s what happens when you shift from conventional attribution modeling that can only measure direct channel contributions to a more nuanced approach that can infer and model direct, indirect, and joint effects for digital and offline marketing. 

For marketers in healthcare, med tech, and other privacy-first industries where cookie deprecation, shrinking attribution windows, and HIPAA constraints are already limiting what you can measure, media mix modeling is worth taking very seriously. It doesn’t rely on granular user-level data that creates compliance risk. 

It works on aggregated business data you already have: spend, revenue, sales cycle, offline inputs, geography. And it tells you something your attribution models can’t: where to put the next dollar and what it will return. If you’re looking for him, you’ll find Michael Wiegand on LinkedIn and at wheelhousedmg.com. 

I’m Aaron Burnett. We’ll see you next time on The Digital Clinic. 

Sponsored by Wheelhouse DMG