Marketing mix modeling (MMM) has become faster, cheaper and more accessible in the last few years. But these significant developments raise an important question: as the barriers to building models fall, what does it take to turn their outputs into better business decisions?
Open-source frameworks, cloud-based platforms and AI are changing the economics of measurement. Processes that previously required months of work can now take weeks, and marketers have more opportunities to update models as circumstances change.
The benefits are reflected in Gain Theory’s report, When MMM Moves at the Pace of AI: Speed is Easy. Growth is Everything. Asked which advancement had most improved decision-making over the previous 12–24 months, 62% of respondents chose “AI-driven speed”.
Yet exclusive interviews we conducted with senior marketers who participated in the report reveal an important distinction. Faster access to insights is valuable but it doesn’t guarantee that those insights are robust, relevant or actionable.
Across the conversations, a consistent theme emerged: as technology takes on more of the work involved in producing models, human judgment becomes more important to determining what the outputs mean and what businesses should do next.
“The Cost of the Model Has Collapsed”
Historically, the barriers to adopting MMM were practical rather than philosophical. Marketers wanted to understand what was driving performance, but the resources required to build and maintain a model was prohibitive. Today, that is no longer the case.
“The cost of the model has collapsed,” says a senior marketing leader at Expedia. Open-source frameworks and lightweight modeling tools have, in their view, fundamentally changed the economics of measurement: “They’ve taken the time to first model from multi-month engagements to a few weeks. This is real democratization.”
A senior marketing leader at Levi Strauss describes a similar change. The process is now “faster and more accessible”, they say, making it easier to align data and assumptions.
When modeling becomes less expensive and more accessible marketers can reconsider how they use it. “Instead of getting an MMM model once a year, you can have it run multiple times,” notes a senior marketing leader at Chewy. The benefit of this is that more frequent updates create opportunities to revisit assumptions, investigate changes, and use measurement as an ongoing input to planning.
MMM is also playing a broader role within a company’s overall measurement approach. Senior marketing leaders at BYD and HelloFresh describe combining it with digital attribution, customer journey analysis and CRM data to build a more complete picture of marketing incrementality. Here, the value lies in bringing those perspectives together as different approaches can help answer different questions. The marketer’s task is to understand how findings relate to one another, rather than assume that any single tool provides the whole answer.
More Control Brings Greater Responsibility
Alongside changes in cost, speed and role of MMM, there has been a shift in ownership. Marketers no longer need to rely entirely on external partners to run models and interpret the results.
“I don’t have to put in a request or have a huge RFP to onboard a company,” says the senior marketing leader at Chewy. “It’s something that you’re able to do very quickly without dependence on a lot of other departments.”
This greater autonomy can bring marketers closer to the measurement process. At HelloFresh, the senior marketing leader describes having “more control” and “a better perspective on the various data sources”.
That is useful progress, but control over the software should not be confused with control over every factor that determines the quality of its outputs. Having a clearer view of data sources does not necessarily make the underlying data complete, accurate or timely.
Offline media data can remain difficult to collect and reconcile, privacy regulations can continue to reduce the signals available, and insufficient granularity can limit what a model is able to explain.
Automation doesn’t remove these constraints. A model can be built more quickly while still relying on incomplete information.
As the senior marketing leader at Expedia points out, the real value of greater control comes from owning the priors – the assumptions brought into the modeling process – and developing holdout and geo-split testing capabilities to verify what the models are saying.
Ownership, in this sense, is not just about access to a platform. It is taking responsibility for the reasoning that connects data, analysis and decisions.
Business Context Remains Essential
The distinction between producing an output faster and making a better decision becomes clearest when recommendations meet the realities of a business. As the senior marketing leader at BYD warns: “AI can make modeling faster, but the real value still comes from asking the right questions and understanding what each methodology can and cannot explain.”
That observation gets to the heart of effective measurement. A model is useful when it helps answer a relevant business question. Its technical capabilities do not remove the need to define that question carefully or understand the limits of the answer.
Consider a recommendation to increase investment in a particular channel. Historical performance may support the recommendation, but the marketer still needs to establish whether it is feasible.
“AI can tell me that we should invest more money in brand, but [I know] that’s not actually a possibility,” says the senior marketing leader at Chewy.
The recommendation and the business constraint are not necessarily in conflict. They are different parts of the same decision. Understanding the potential return matters but so does understanding the budget, priorities and options available.
AI models are becoming better at capturing and incorporating business context, and further improvements should make it easier to bring information such as competitor activity, supply constraints and changes in demand into the analysis.
The distinction is therefore not that people understand context while models cannot. It is that someone still needs to assess whether the relevant context has been captured, how reliably it has been interpreted and what may be missing.
To make this a reality at Expedia, human oversight is a structural requirement rather than an optional check. The position at Levi Strauss is similarly clear: “AI can help us as a support tool, but it’s definitely not the final decision maker.”
Preserve the Ability to Challenge an Answer
As recommendations become easier to generate, there is also a risk that organizations may lose the habit of questioning them. “The failure mode I’d watch isn’t AI overruling people, it’s atrophy,” warns the senior marketing leader at Expedia. “If every recommendation arrives pre-written, the muscle of interrogating a number weakens. It’s fastest in those who never built it.”
This is important because the ability to challenge an output and assess whether it provides a sound basis for action is key part of what makes measurement valuable. Asking questions, such as “Why has a result changed?”, “Is the explanation consistent with what is happening in the business?”, “Could another factor account for the pattern?”, “What further evidence would give us confidence to act?”, should not be treated as obstacles to a faster process. They should be used to distinguish a plausible recommendation from a well-supported decision.
The challenge is to capture the benefits of automation without weakening the capabilities needed to use it effectively. Less time spent on manual production should create more room for interpretation, testing and discussion — not less scrutiny of the result.
Speed Is an Advantage, Not the Objective
The changes taking place in MMM are genuinely encouraging. Lower costs can broaden access, faster modeling can make measurement more responsive, and greater autonomy can bring marketers closer to the data and assumptions informing their plans. But these developments are means to an end.
The objective is not simply to run more models or generate recommendations more frequently. It is to help businesses make better choices about where to invest, what to change and how to grow.
That requires technology and human judgment to work together. Models provide evidence and help explore options. People bring business context, challenge assumptions and determine what action is appropriate ensure a business achieves its objectives.
Download When MMM Moves at the Pace of AI: Speed is Easy. Growth is Everything.
