![](https://mobiledevmemo.com/wp-content/uploads/2024/06/mdm_media_mix_modeling_fails-1024x585.jpg)
I’ve written extensively about media mix modeling (MMM) and its critical role in marketing measurement in the new privacy landscape. As the signals available for marketing optimization diminish or disappear, marketing teams should utilize probabilistic measurement approaches like MMM to fill the information void. Probabilistic measurement frameworks allow marketing teams to understand the effects of their marketing spend using the most reliable data available: ad spend and revenue.
However, while MMM can be a valuable tool in a suite of measurement solutions, it can’t be the exclusive tool used for measuring the effectiveness of ad spend in promoting digital products. I’ve seen MMM be rejected by host organizations like an unsuccessful organ transplant. Integrating an MMM into a marketing workflow requires a tremendous amount of effort: in adapting that workflow to the limitations of MMM, in properly setting expectations across the broader organization, and in supplementing the MMM with other tools to service the use cases for which it is not suitable. In this piece, I’ll outline four reasons why I’ve seen media mix modeling fail.
Misalignment across the organization.
As I detail in How to train your CFO (about probabilistic measurement), incorporating probabilistic measurement into a marketing workflow requires a socialization of the output with the entire company, particularly the finance team. MMM is a budgeting and forecasting tool that is calibrated on historical data and run intermittently (often bi-yearly, quarterly, or monthly). When advertising spend represents a substantial amount of a company’s expenses, as generally it is for companies selling digital products, finance teams expect a near-immediate cadence of reactivity to marketing expenditure that isn’t directly supported with MMM.
![](https://mobiledevmemo.com/wp-content/uploads/2024/06/image-5-1024x882-1.png)
The executive team within an organization needs to understand (and approve of) the limitations of MMM-based budgeting before the tool is incorporated into a marketing workflow. And MMM alone is often seen as insufficient for providing the executive team with the information it needs to approve marketing budgets and strategy. MMM isn’t a component that can be swapped out of the existing marketing measurement stack; more appropriately, it should sit atop a different measurement solution to guide budget allocation, as I describe in Integrating a Media Mix Model into a digital marketing workflow. If the executive team either isn’t prepared for the limitations of MMM or doesn’t view it as an appropriate solution, on its own, for guiding marketing spend, then the solution will fail.
Failure to achieve operational adaptation.
There are two realities about deterministic attribution that marketing teams need to accept:
- It never existed, at least not to the degree that most marketers assumed;
- It provided for a cadence of decision-making that is impossible in its absence.
Resetting expectations about the immediacy of decision-making in the probabilistic environment invokes a wholesale overhaul of the marketing workflow to achieve greater certainty around the correlations between ad spend and revenue. This generally demands higher levels of budget concentration within a portfolio of channels and more moderated adjustments to changes in campaign settings. As I write in The Jenga Tower principle in modern performance marketing:
The new privacy environment engenders what I call a Jenga Tower principle for performance marketing: as with a Jenga tower, changes to performance marketing campaigns must be made slowly, systematically, and with care to not disrupt the very fragile architectural integrity of the structure — which, in the case of performance marketing, is the reporting and analytics apparatus that guides bid and budgeting decisions on the basis of return on ad spend (ROAS).
Using an MMM for measurement necessitates these changes to workflow — by extension, neglecting to institute these changes ensures that an MMM-based approach will fail. It’s not possible to use an MMM in the same manner as a deterministic attribution system; these are two fundamentally different approaches to marketing measurement that are incompatible conceptually.
Lack of traction with scaling smaller channels.
MMM tools attempt to estimate the optimal level of spend across a known portfolio of channels given historical performance. This can work well when a product has reached a steady state of marketing channel utilization and is no longer onboarding or appreciably scaling new channels.
But this profile describes a very specific type of product: often a “legacy” product with known monetization characteristics, with audience relevancy having been established. At the opposite end of this spectrum are new products for which very little data exists around historical channel performance. This describes start-ups.
It’s possible to segment measurement approaches for scaled, existing channels and new channels into separate workflows. This creates additional operational complexity for the marketing team. Yet integrating a new or rapidly scaling channel into an MMM will almost certainly credit any effects from that channel to the larger, existing channels, resulting in budget allocations that aren’t based on underlying performance, starving the new channel of ad spend. This dynamic puts a ceiling on growth: when newer and smaller channels aren’t credited with their performance, they aren’t scaled, and the existing channels are given more budget, which ultimately is reverted because their own performance doesn’t support it.
Inability to accommodate digital-specific, intermediate conversion signals.
MMM as a concept is decades old and was built for a marketing environment in which very little data specific to ad engagement was available. This is not the case with digital marketing, even in the more restrictive privacy reality: intermediate signals like e.g., clicks, SKAdNetwork postback, ad interactions, and other signals can be used to optimize campaigns.
![](https://mobiledevmemo.com/wp-content/uploads/2024/06/image-12-1024x576-1.png)
I describe a dual-track workflow in The emerging marketing economist that utilizes these intermediate signals in an “operational model” to optimize campaigns on a relatively short timeline and higher-level signals like revenue and total ad spend with an “econometric model” on a longer timeline. Exclusively relying on an econometric model like an MMM for marketing decision-making neglects valuable operational information. Not only are these signals available and dependable when promoting digital products on digital channels, but they are essential inputs to proper marketing measurement. Measurement methodologies predicated exclusively on MMM can’t accommodate these and are liable to fail.