After years of collaboration, Amie and her team at E29 Marketing tossed a curve ball my way. They requested I write a blog on an enticing and enigmatic topic as mystical as a unicorn: marketing mix modeling. In today’s data-driven world, brand marketers need to grasp the how, what, and why of this concept so they can make the most out of their marketing plans and budgets

"No problem," I said, posing one crucial question: how versed will the audience be in terms like objective functions, constraints, and gradient descent algorithms? The answer was less than I'd hoped. The instructions were to write for marketers, intelligent yet perhaps without an advanced mathematics degree.

So... let's liken the topic to cooking. My grandmother was a culinary alchemist, always meddling with her secret blend of ingredients, constantly noting cryptic annotations on recipe cards.

She was optimizing her recipes on instinct, remembering what seemed to work and didn’t, and constantly trying to outdo herself.  That time I used the nutmeg, she might have thought, worked out really well.  But I also used extra honey that time.  Was it the nutmeg or the honey that mattered?  Ah, but there was another time when I used extra honey without the nutmeg and that wasn’t great.  Oh, but wait, that was the time I tried baking at a lower temperature, so I can’t necessarily blame it on the honey.

Now imagine if we could magically analyze grandma's cookbook, decipher her notes, and create the ultimate guide to cook her signature dishes?  It's not far from reality today, with “Recipe Mix Modeling” (you didn't see that coming, did you?)

Every event could be fed into the computer with some measure of a taste-test result and all the inputs:  how much of which ingredients, temperature, cooking time, the particular cookware.  The analyst could build a model that not only goes back and quantitatively estimates the effect of each recipe decision, but is capable of simulating what would have happened in scenarios that were never tried.  The algorithm could identify the optimal recipe mix.

This is exactly what Marketing Mix Modeling (MMM) does. It uses your marketing ingredients to whip up a successful soufflé.  MMM tells you how much of each ingredient (TV, print, social media, SEM, etc.) leads to the optimal recipe. Whether your goal is maximizing returns on a fixed budget or minimizing spend to hit sales targets, MMM provides the ideal mix.

It's not as simple as just adjusting one area up or down. The ingredients interact. If you increase social media spend 10%, the model may reveal you should then slightly decrease TV and print budgets, and increase SEM. The model runs simulations to find the right amounts of each ingredient to accomplish your goals.

So if the model identifies an opportunity to boost SEM effectiveness 10% by increasing spend, it would also suggest complementary adjustments - like a 5% TV budget increase funded by reducing print and social media. MMM looks holistically across marketing channels for the perfect recipe.

MMM not only helps allocate the budget but shapes messaging, positioning, pricing, and promotions.  It can assist with planning campaign launch dates and durations and tracking progress against goals.  Is it magical?  Absolutely (thanks to proven mathematical techniques). Isn't it aptly likened to a unicorn?

Well, I have to admit that it is just about as hard to build one of these models as to catch your average unicorn!  But the work isn’t in the magical incantations data scientists use to control their algorithms; the work is in the preparation.  Thinking in terms of our cooking analogy, step 1 is to find all the ingredients, and some of them aren’t at the supermarket so you need to visit the specialty spice store and the farmers’ market and stop by Aunt Hillary’s place to pick some rosemary from her garden (and you know how Hillary likes to talk and talk).

The data? Start by identifying your target outcome:  Customers?  Sales?  Revenue?  Choose the model’s "KPI" and build from there.  You’ll need all marketing data from the past 2-3 years, from Instagram campaigns to PR efforts, TV ads, and even that DM postcard campaign you’d hoped everyone had forgotten.  Some data files are easily downloaded, others pulled from creative decks, emails and inconsistently formatted Excel files maintained by media agency interns. It might start as a mess when you do it for the first time. Seriously.

The data needs to be time-based.  Ideally weekly, but sometimes it is daily or monthly.  And of course not all your data will line up.  One source uses Sunday-Saturday weeks and another sends Monday-Sunday weeks. Another source only sends monthly data, and only then do you realize how unfortunate it is that months are not actually exactly four weeks long.

And you don’t just need your marketing data.  Other factors affect your business results besides just your marketing (much as any CMO wouldn’t want to admit it).  In our cooking example, it is about more than the ingredients.  Cooking time and temperature matter.  Even the humidity that day can matter!

And so you need information about relevant holidays and time periods.  Does back-to-school time matter for your business?  Was there a Covid lockdown right in the middle of a big PR campaign?  Did a competitor run a big promo or launch a new product?  Did you open new stores or have a major supply chain issue that caused out-of-stocks one week?

All of that data has to be collected and then stitched together into a single giant repository.  That is, all the pages of my grandmother’s cooking notebook need to be copied neatly into a spreadsheet and organized in a way that the machine learning algorithms can make sense of it.  But don’t think you’re ready for the modeling… oh no!  Not yet.

Data errors creep in.  The wrong files were sent by that junior media planner.  Something was missed.  Marketing mix models are big endeavors and in turn inform big decisions, so you don’t just plow ahead without taking some serious pains to double-check everything.  Think of it like grandma taste-testing everything in the kitchen to make sure she doesn’t poison Aunt Hillary or any of the grandkids.  Well, at least not the grandkids.

Are you exhausted from reading all this?  Imagine how the data science team must feel having to do all this!  But the reward is that they can then make the algorithms, guided by experience and domain knowledge, work out that optimized recipe.

At the end, you’ve done a lot of work. A lot of work. But at last you have your model. Now, I don’t mean to burst your bubble, but keep realistic expectations.

Your model is insightful, but imperfect. It's a guide, not a crystal ball. The future echoes the past but isn't an exact match. Even the best models have limitations, as marketing doesn't happen in a vacuum. So what is the model good for? An informed seat at the decision-makers table.

Without a model, marketing decisions come from experience, intuition and debate. Think of the model as a data-driven marketing-expert robot, kind of a weird member of the marketing team (like Aunt Hillary), who gets to sit at the table with you and offer its thoughts and recommendations, backed by the data. But it doesn't wholly replace human judgment.

Consider the model's guidance closely, but weigh it against other inputs and expertise. Use it to inform decisions, not make them outright. The model brings immense value in optimizing spend, allocating budget, and guiding strategy. But it serves more as a co-pilot than an autopilot.

AI is all over the news, but we aren’t at the point where we can hand our decisions over to it.  The AI can help with quality checks of data, but it doesn’t know that you’d never have spent that much in June so that budget entry must be a typo.  The AI can help build the model, but it won’t realize halfway through the modeling that the September results are off because you forgot to include the one-off promotion you did with that new social app that offered you a “deal” on their pricing.  AI systems right now can make this process faster and more efficient, but both the marketing mix modeling and the decisions driven by it still require human thought.

That is to say, the robots can help bake the cake, but they can’t taste it so maybe don’t take a bite until Aunt Hillary tries it first.

So if you aren’t going to hand the reins over to Hal 9000, who will build your model?

Well, why do you think E29 Marketing called me in the first place?  You know where this is going.

To be honest, you do have options.  Let me bucket your choices into three basic categories.

First, internal handling might seem budget-friendly, but most companies lack the resources.  Do you have an analytics team with spare bandwidth and specialized modeling expertise?  Unlikely.

Then there are consulting firms, both big-name and niche specialists.  It's a viable option but can be costly, and they might need time to understand your business.  And their accountability post-engagement?

Lastly, your agency - but choose wisely! Many agencies boast world-class data science teams. But boast is often the operative word. True marketing mix modeling masters are rarer than unicorns who've won the lottery.

Vet carefully to ensure your agency has legitimate MMM superpowers. They should understand your business and data inside-out. Look for extensive experience specifically in marketing mix modeling across many clients and industries. Don't just take their word for it - ask for case studies and client references.

If you find a qualified agency, they can be the perfect fit. Understanding your business, accountability for results, and a vested interest in an accurate model. It's worth hunting patiently to find this unicorn agency, rather than settling for the first donkey in a party hat that trots along.

E29 Marketing, just coincidentally, fits the bill. Our world-class data science team has decades of MMM expertise across a variety of industries (feel free to ask about it). But enough shameless plugging - the key is finding the right agency partner to trust with igniting the full power of marketing mix modeling magic for your brand. The payoff is well worth the search.

But hold on, I hear you say, this is letting the fox guard the henhouse!  The cat guard the pigeons!  The sorcerer guard the unicorns!  The agency may be biased in that they want to show that the historical marketing (that they did) was great work and that the client should spend more money with them.  I hear this a lot.

It’s a misconception.  This caveat comes from a misunderstanding of marketing mix modeling. You want an agency that seeks insight into what is not working, just as much as what is working.

If evaluating your historical marketing performance or calculating return on marketing investment (ROMI) is the goal, there are simpler methods than full-on marketing mix modeling.  ROMI efforts could be efficiently executed by your agency’s analytics team, although trust is necessary to alleviate concerns of bias.  Interestingly, CEOs and CFOs are usually fine with the marketing department overseeing the ROMI effort (regardless of the vendor they choose), despite the marketing department having a similar bias!

Marketing mix modeling can give you some idea of how well your marketing performed versus how well it might have done with optimal conditions.  But that is not its primary purpose.  The purpose is to guide future marketing decisions.  How to allocate budgets, when to run promotions, and where to place media, among others.  This process involves building a robust mathematical model of your business, so the concern about the fox guarding the henhouse is a red herring.  It's in your agency's best interest to build a model that will optimize your future marketing performance.  After all, your success is their success.

While an agency-led marketing mix model might not always be the best fit for every brand, it certainly merits consideration.  A reputable agency will always operate in your best interest, employing their skills and knowledge to build a model that optimizes your marketing strategy.

MMM takes work, but summons powerful magic - if not quite a unicorn.  As for Grandma’s cookbook?  That remains a family treasure – but with some help from Recipe Mix Modeling, who knows?  We might finally decode her secret ingredient.

Subtlety isn't my strength (that's data science), so let me be clear: Need marketing mix modeling? Reach out and E29 will deliver a (metaphorical) marketing robot ASAP!