Marketing mix modeling is having a moment. In a digital marketing world dominated by “black box” algorithms, GDPR, and the deprecation of third-party data, MMM provides marketers with a tangible blueprint for optimizing their media investment.
Originally conceived as a statistical model for measuring the importance of above-the-line marketing tactics like broadcast, MMM looks at current sales and marketing data to anticipate how future investment shifts will directly impact sales.
As you might imagine, this can give marketers a powerful edge in the current market. But, what limitations does it have? How much data do you need? Also, how do you get started?
To unlock the secrets of MMM, we sat down with the experts at Adtriba, a Funnel Technology Partner. Janos Moldvay, CEO, and Tim Kreienkamp, Chief Data Strategist, were happy to share some of their top tips and expertise to help you take full advantage of MMM.
Left to right: Janos Moldvay and Tim Kreienkamp
The marketing mix modeling resurgence
As Janos and Tim tell it, MMM isn’t a new approach at all. In fact, it first gained popularity in the 1970s and 80s. However, since the emergence of GDPR in 2018, marketers have rediscovered MMM’s value .
“We saw a lot of interest in MMM starting in 2019 due to cookie restrictions,” said Tim. “While GDPR meant Europeans had to care about this issue first, the American market soon caught on thanks to other privacy restrictions implemented by Google as well as Apple’s anti-tracking software.”
Marketing mix modeling isn’t solely for those above-the-line tactics anymore. But instead, it’s a valuable tool that can help you begin to understand contribution and incrementality.
How does MMM work?
At its core, marketing mix modeling is a statistical tool that looks at different variables (i.e., your media tactics) and determines how important they are to your organization’s goal (most often, sales).
For those readers who have taken more advanced math courses or earned business degrees, this may sound very similar to linear regression. While MMM is similar, Tim quickly pointed out its key differences.
“Yes, MMM is a type of regression, since it can model continuous outcomes,” said Tim. “However, it’s not linear due to the nature of advertising. Over time, advertisers see diminishing returns due to utility declines, or they hit peak saturation. Plus, ads that you run today can impact tomorrow and beyond.”
It makes sense if you think about it. Tactics like brand advertising build long-term customer affinity over the years. At the same time, your company may be running high-converting ads based on limited-time promotions. Plus, all of these messaging strategies can be executed across various media. MMM tries to make sense of all of this.
So, if you were so inclined (and had a fairly simple media plan), could you create a marketing mix model yourself?
The classic dilemma: build vs. buy
If you really (and we mean really) love statistical modeling, you could build your own models from scratch.
“In this instance, I would recommend starting with regular linear regression models, then build in further complexity,” said Tim.
However, Tim pointed out that there are plenty of open source MMM models to get you started. In fact, Adtriba uses a stable of multiple models at once, while also adding their own customizations. Depending on the client and issue they are trying to solve, they use different models.
The best decision, build or buy, will likely come down to your own expertise, ability, and available time. In any case, it helps to have a partner like Adtriba to implement your model and interpret the results.
When we asked Tim about the most underestimated challenge in MMM, he quickly pointed to the data requirements.
“You don’t necessarily need huge volumes of data,” said Tim. “Just megabytes of data can be fine, but you need to be able to collect the right data.”
Often, Tim and his team need to access daily or weekly performance data. To achieve this, they almost always use Funnel. But without the marketing data hub, they require a client’s BI resources to pull that performance data daily. The issue, of course, is that the BI team may accidentally pull incorrect data.
“I’ve seen cases where the wrong query is used, and then the acquisition costs somehow triple in the model,” said Tim.
What are the most common client challenges?
Nearly all MMM projects are, on some level, aiming to optimize budget allocations. When approaching Adtriba, most clients are attempting to measure parts of their media spend (like offline media) that lack the hard data of digital advertising.
As more marketers become more data-driven, this need to quantify and accurately measure the vaguer parts of their media budget will become more critical. Yet, according to Tim, the best insights that MMM can yield sometimes don’t come from where marketers may think.
“Clients are often surprised to see that Instagram is often undervalued in the upper Funnel,” said Tim. “Running a comprehensive MMM project can help uncover those insights and enlighten clients about how they can better leverage existing tactics.”
In the case of Instagram serving your upper-funnel efforts, some businesses may want to allocate more brand awareness or product launch messages to those media placements.
The biggest misconception of MMM
While marketing mix modeling can uncover surprising truths, it’s not a panacea for every problem that marketers face. Additionally, it doesn’t work for every business model.
“Unfortunately, MMM doesn’t really work well with small tests,” said Tim. “It needs substantial investment to show a signal due to the nature of the data.”
In other words, the model can’t recognize any statistical significance in a one-off micro campaign that runs for weeks or days. Any so-called “insights” may be illusions of the statistical noise — just like in A/B testing or other experiments.
The model depends on your business model
Marketing mix modeling works particularly well for consumer brands with substantial, sustained ad spending. After all, those are the businesses it was originally designed to serve. Consequently, B2B business models will struggle to use marketing mix modeling effectively.
If you look at B2B SaaS companies, you could roughly tie a marketing tactic to a KPI (like a sales-qualified lead), but from there, the picture becomes much more blurry. The lead comes into contact with one or more salespeople, tries a demo, may have to sell the value of the SaaS product to internal stakeholders, and more. That’s often too many variables and too long of a sales cycle for MMM to account for.
According to Tim, there are even some B2C brands that can struggle to employ MMM.
“Imagine you’re a popular beer brand,” said Tim. “You advertise to customers, but sell to distributors and wholesalers like grocery stores. Grocers may always place their orders on a specific day, but that day could be different for each grocer. This can skew the MMM, so you might need to use weekly-level data instead of daily.”
Other business models can also face unique challenges when employing MMM, though it’s not impossible.
“Luxury can use marketing mix modeling quite effectively, but it depends,” said Tim. “If you’re selling something like yachts and only do a few sales per week or month, that can be challenging to account for in the model. However, hitting 30 or 50 unit sales a day can be quite valuable for MMM.”
The long arm of AI’s influence
It’s no secret that AI's impacts are hitting every marketing specialty (as evidenced by our annual Marketing Data State of Play), and MMM is no different. However, Adtriba has been riding the AI wave for quite some time. As Janos, Adtriba’s CEO, pointed out, they are an AI company.
“MMM, and Adtiba as a result, require machine learning to function,” said Janos. “We’ve actually been using AI since our inception.”
Fun fact: Adtriba uses the same machine learning infrastructure as OpenAI, the makers of ChatGPT and DALL-E.
The future of MMM
While advances in artificial intelligence may seem like the future is now for some marketers, the Adtriba team still sees big things in store for marketing mix modeling.
“I think advances in automation will open up new possibilities,” said Tim. “MMM used to be done annually or semi-annually , which often meant results were already obsolete.”
Now, Tim is providing daily reports to clients.
Your wish is my command
With such powerful technologies like machine learning and MMM at their fingertips (not to mention their multi-touch attribution tools), you might think Janos and Tim have it all. But it’s the human factor they wish they could still influence a bit further.
They both expressed that, while MMM can show you the way, marketers must still act on those insights — which sometimes requires a bit of courage.
“I sometimes wonder if we could feed the MMM results back into some sort of automated budgeting system,” said Tim, “but at the end of the day, clients still need to maintain control and responsibility.”
Indeed. While these advanced AI-driven tools are incredibly powerful, they almost always need the human touch in the end.
Disclaimer: The featured image for this article was created using generative AI.