An old marketing industry technique is seeing a resurgence in popularity these days: MMM. Marketing mix modeling, sometimes referred to as media mix modeling, measures the impact of a company’s advertising in various marketing channels—TV, magazines, podcasts and more—while factoring in other variables such as budget spending and economic forces.
Though the concept has been around for years, MMM is back in vogue thanks to advancements in machine learning and data analytics. When you consider the death of third-party cookies, and the fact that MMMs take a top-down approach to market analysis while placing less emphasis on user-level data, it’s little surprise that MMM is having a renaissance.
“If you’re just reporting on media, then it’s only of interest to the CMO,” said Michael Cross, global senior vice president of measurement at Media.Monks, a virtual talent hub for data, digital media and technology services. “But if you’re actually putting media in context with all the other drivers—like price, promotions—you then get the attention of the finance team and then you get your opportunity to take this to the board.”
MMMs are undoubtedly gaining in recognition in today’s landscape, but what do you—a marketer, not a data scientist—really need to know to interpret results and make better decisions? MMMs are created using a wealth of past data, statistical wizardry and some very important assumptions. Let’s talk about each of these.
The anatomy of an MMM
There are multiple phases of an MMM, and different producers of them—including econometricians, in-house data scientists and other data consultants—will generate their own unique twists to the form while catering to a company’s unique data needs.
One of the most important parts of an MMM report is what’s known as the decomposition chart, which is essentially a layered tracking grid that plots measurements of categorical data sets collected over a period of time. Imagine measurements devoted to the market’s impact on sales atop media-spend numbers, price changes, social media investment and a competitor’s revenue. How has your company’s branded pay-per-click campaign been going? You’ll see its ups and downs on the decomposition chart, as well as data about seasonality effects on sales and other outside forces, such as global pandemic or economic recession.
Each MMM chart—with its colored bars and line graphs—helps an analyst observe potential trend relativity. Interactivity capabilities of today’s MMM dashboards give analysts seemingly limitless options of how to approach the data sets, from comparing measurements by category to assessing time threshold dimensions.
An MMM may contain additional slides charting “spending” versus “sales” versus “profit,” with breakdowns of each by advertising type (e.g., outdoor or display). Based on these outcomes, the model will suggest how to spend in the future, project profit change, ROI and more.
“Market mix models can be leveraged to measure anything,” noted William Parker, vice president and head of client strategy at Leavened, a marketing mix optimization platform. “When we look at things such as if the metaverse ever comes to fruition or if virtual reality becomes a major space for marketers, we can measure that right now with the tools we have today. Today we’re just measuring impressions and spends.”
The evolution of MMM
MMMs have been used by companies for decades. Because MMM requires a lot of data, the process of building an actionable report was incredibly labor intensive in the pre-digital age. They took months to cultivate and, when they were analyzed, the data was often dated and not as helpful to marketing decision-makers. If marketers ordered an MMM to be prepared at all, they usually did so on a quarterly or annual basis. That’s no longer the case.
“A process that used to be on a six-month cadence can now be reduced down to days,” said Myles Younger, head of innovation and insights at U of Digital, an organization that provides education and training for digital marketing teams. “You’re getting the reports more frequently, and the reports have less latency, so they may be only a few weeks or a few days out of date.”
Today, MMM vendors can develop a report for a marketing team in as few as four weeks. This kind of turnaround time has primarily been made possible by the integration of artificial intelligence and machine learning tools that automate key elements of MMM production, like data collection and forecasting. MMMs have also become cheaper to generate and increasingly rely on only aggregate data. Because MMMs report from a bird’s-eye view of factors, including TV spending, search spending, web and mobile display spending, with each broken down by market, an analyst can secure the necessary data by emailing the company’s finance department. Oftentimes, they can even include publicly available information.
The more modest price tags attached to MMM reports these days give smaller-sized brands the opportunity to use the solution for a broad range of tests, such as whether a partnership with a particular influencer is making an impact.
How to avoid bias in MMM
As with any solution that uses AI, an overreliance on machine-generated insights creates risk. But any good econometrician who takes the time to analyze MMM data can mitigate such outcomes. In many cases, simple settings can help.
For example, in the retail or consumer packaged goods industries, marketers know that a TV spend uplift should be roughly between 1% and 4%. “It shouldn’t be 20%,” Media.Monks’ Cross said, “because we’ve never seen that.” So, whoever’s building the MMM can program such boundaries into the model, so it emerges “more stable” and “doesn’t go haywire.”
Another way to avoid biased influence on decisions tied to MMMs is to triangulate all measurement solutions. “If you’re getting two different answers from two different solutions, it doesn’t mean one of them is wrong,” Parker said. “They both might be kind of right.”
In this case, multitouch attribution analyses, as well as A/B or geofencing testing and brand lift studies, can create a more detailed a picture of your marketing outcomes.
Because it’s a macro form of measurement, MMMs also provide what Younger says is “a really blurry view of what’s going on compared to what your digital marketer might be used to seeing.” And while the price of MMM delivery might have gone down, that fuzzy view is still going to be costly.
“MMM is expensive because it takes a long time to do [and] you need experienced people to do it,” says Cross. “If you’re a global brand with a lot of products, then it can get very, very expensive to do, but in that case, what we recommend is that the clients try and build the capabilities in-house.”
The future for MMM
If the price of MMM is a problem, there’s a chance that won’t be the case at some point in the future. The automation power will only become more substantial and the speed will only increase, helping to cut down resource requirements.
As data sets change, so too will the insights within a typical MMM. When Google kills off third-party cookies completely, the hope is that ad agencies put their eggs in its Privacy Sandbox basket. Apple already has its SKAdNetwork up and running, which maintains user privacy while attributing engagement to app installs.
With these and other new forms of data collection on the horizon, said Younger, the question becomes “How do you factor that into MMM?”
