top of page

Embracing Marketing Mix Modeling’s Even Smarter Future

A Conversation with LiftLab’s newest Advisor, Professor Koen Pauwels


One of the world’s leading experts in Marketing Mix Modeling (MMM), Professor Koen Pauwels is a testament to the versatility of the field. His work spans three continents and has brought him into affiliation with brands as varied as Amazon, Heinz, Kayak, Kraft, Marks & Spencer, Microsoft, Nissan, Sony, Tetrapak, and Unilever. 

Pauwels is a Distinguished Professor of Marketing at the D'Amore-McKim School of Business at Northeastern University. He has also taken a position as Advisor to LiftLab. 

We sat down with Professor Pauwels for a wide-ranging conversation about using MMM to solve some of the knottiest issues in marketing – and how new data plus better experimentation is making MMM’s answers more relevant than ever. 


What follows is an edited version of our discussion.


LL: When people think of you in the field, what are you best known for? What’s your research primarily about? 

KP: You could say I’m something of a “big picture guy.” I have a background in econometrics and strategy, and I advise companies on a macro-level: How should you think about how big your budget should be? How should it be allocated? 

As part of that approach, I’m interested in two interrelated questions. One is long-term impacts, something I’ve been fascinated with throughout my career – my dissertation was literally titled Long Term Effects of Marketing. I looked at the classic “four P’s”: if you change something about your Products, your Price, your Place (including going online), or your Promotion action, how do each of these have a long-term impact on the bottom line? 

At the same time, I’m known for my work with vector autoregressive models, or VAR. VAR is a kind of modeling that is really good at looking at long term effects and how variables can be both a driver and driven by other things. A typical example here is brand equity. Brand equity drives your sales, but it's also driven by your marketing actions and a whole host of other external factors. In more technical terms, the question is: How do you study something that is both the dependent variable in one equation and the independent variable in another equation? 

Brand equity is just one example of a web of reactions that marketing can trigger – others include consumer reaction, competitor reaction, government response, and so on. Throughout my research and advisory work, I ask how these impacts interplay, what the net effects are of any given action, and how senior managers should think about allocating budgets and driving initiatives accordingly. 


LL: What excites you most about where the MMM practice is headed? What’s the MMM opportunity for marketers on the horizon? 

KP: I see three major trends driving the MMM opportunity right now. The first is the ability to use MMM to help fill the measurement vacuum now created by data deprecation. The second is more precise models made possible by more granular inputs. The third – and perhaps closest to LiftLab’s core thesis – is the new opportunity to validate and strengthen models through experimentation.

Let’s start with MMM’s place through data deprecation. On top of regulation and a general push for greater data privacy, it looks as if Google is poised to sunset third-party cookies on Chrome. Individual-specific attribution will become a lot harder, creating a host of challenges for marketing practitioners.  Growth marketers, especially, will face significant changes.

Of course, marketing mix models have always excelled in bringing in a wide array of inputs, individually tracked or not—whether you’re talking about out-of-home, radio, TV, digital, or any other channel. It’s an ideal approach to help marketers thrive in the face of all that data deprecation. 

Add to that the long-term benefits of MMM – instead of telling you what happens in the next seven or fourteen days, MMM can help guide decisions for many months going forward—and you can see why MMM stands to emerge as an incredibly powerful tool in the new data landscape. 


Second, as I mentioned, you have that new data precision. Even as many data sources go away, we also know that there’s a wealth of new, unprecedented granular inputs–granular both in terms of time and in terms of geography. For example, in the old days, I estimated everything weekly because that's how decisions were made. Now, you can estimate things at the daily or even hourly level. 


Third, I'm very excited about the ability to combine marketing mix modeling with other methods – and particularly experiments – to validate findings. With experiments, we can determine if the models built from historical data are just true in the past or if they’re working for the marketer right now as well. 

In my work, I typically model based on historical data, and then we move on to field experiments – randomized control trials – to understand how accurate the models are. And then what’s great is that the data from the experiments offer variation in the data, which we can model against again and thus improve the work. So, I see models and experiments as a virtuous spiral or circle, getting you higher and better in accuracy and effectiveness with every round.


LL: What drew you to LiftLab as an advisor? What did you see as unique about the company? 

It’s the experiments that excite me about LiftLab. LiftLab’s experimentation approach is intriguing to me. Often, marketing experimentation takes a kind of binary, “zero-one” approach where you cut spend entirely in some areas and keep it in others – it’s an approach that’s helpful. Still, it misses the crucial nuance that makes the model truly actionable. LiftLab, by contrast, has these experimental conditions where they cut spend by 50% or by 15%, where they’re not just looking into what levers to pull but also how much to pull. Because of those experimentation conditions, after the experiment, we can see how much we should spend more or less, instead of simply saying that a given channel “added significance” without really telling you much about what you should do afterward. 

In addition to the experimentation approach, I was drawn in by some very interesting validation. I sat in conversation with a very large company that was benchmarking its marketing mix models against the primary competitors, and they named exactly three competitors they saw as significant. LiftLab was one of the three. Out of all the marketing science providers out there –and there are many – this major company had chosen LiftLab as a company to watch. And I said to myself: “Hey, that’s some real external recognition of the potential.”


Yorumlar


bottom of page