The Hard Truth About MMM and Incrementality Testing
MMM and incrementality testing can't tell you what to do next. Learn why measurement without orchestration leaves money on the table.
Why Measurement Alone Will Never Fix Marketing and What Comes Next
Marketing mix modeling and incrementality testing are the two most rigorous tools a marketing team can deploy. They are also, for most teams, the point where decisions stop rather than start.
Both are necessary. Neither can tell a team what to do next. The gap between measurement and action is where most organizations stall, running 2 to 4 incrementality tests per year against a marketing mix that changes weekly.
You already know the pattern. The MMM says one thing. The platform says another. Last week's incrementality result contradicts both. Tuesday morning starts with three numbers that disagree, and a budget meeting where someone has to pick one.
Two methodologies dominate the modern measurement stack:
What follows is why neither is sufficient on its own, and what the next evolution looks like: an orchestration layer that unifies both with real time data and business context.
Why MMM Is Powerful and Fundamentally Constrained
Marketing Mix Modeling is surging back because it offers a holistic view of how marketing drives business outcomes. It uses aggregated data, not user-level identifiers, to estimate how channels and other business drivers contribute to revenue.
What MMM Does Well
Accounts for all channels and drivers, including offline spend, promotions, seasonality, and macro effects
Produces ROI estimates that reflect economic impact rather than platform-reported conversions
Is privacy safe by design and independent of cookies or user tracking
Today, about half of US brand and agency marketers use or plan to invest in MMM to guide strategic budget allocation.
Where MMM Breaks Down
MMM lacks granularity by design
MMM operates on aggregated time series data. You will never get user level, ad level, or campaign level truth out of it. That means you know which channels matter, but not which tactics inside them, and you cannot translate the output into day-to-day platform actions.
MMM moves slower than marketing platforms
Even modern automated MMMs that refresh weekly are still lagging indicators. Platforms change daily. Auctions shift hourly. Consumer behavior can turn in days. MMM will always validate decisions after the fact, not guide them in the moment.
MMM explains the past, not the next move
MMM answers the question:
Given everything that happened, what drove impact?
It does not answer:
Given what is happening right now, what should I do next?
That gap matters more than most teams admit.
Incrementality Testing: Causal Truth with Narrow Scope
Incrementality testing is one of the most causally sound ways to measure marketing impact. By comparing exposed and unexposed groups, it isolates the true incremental effect of a campaign or tactic.
Where Incrementality Shines
Cuts through platform bias and over attribution
Validates specific hypotheses with high confidence
Reaches channels no pixel can see. A geo holdout is the only way to establish whether out-of-home, transit, or print caused anything
Where Incrementality Falls Short
It's slow by necessity
Well designed incrementality experiments typically require 4 to 8 weeks to reach statistical significance. During that time, the tested campaigns must remain largely unchanged. That makes incrementality unsuitable as a real time decision engine.
You can only test a fraction of your system
Incrementality is inherently narrow. Each test isolates a single lever. Most marketing organizations run only 2 to 4 incrementality tests per year; even the most mature programmes aim for 20 or more. That is nowhere near enough to continuously guide all channels, audiences, creatives, and budget decisions.
It validates impact but does not orchestrate decisions
Incrementality tells you whether a change caused lift. It does not tell you how to rebalance the rest of the system once that insight exists.
Measurement Alone Does Not Drive Better Decisions
Here is the uncomfortable truth.
Having an MMM provider or an incrementality provider does not mean you are making better marketing decisions. It means you are better informed after the fact. Cann was spending $480K per year on AppLovin while the dashboard reported positive ROAS. A 19-day geo holdout proved zero incremental lift. Measurement without orchestration let that spend run unchecked.
Most teams still lack a system that connects:
Slow but statistically strong signals from MMM
Narrow but causal insights from incrementality
Fast and noisy signals from platforms like Google and Meta
Contextual but unstructured signals: goals, constraints, risk tolerance, channel beliefs
Without that connective layer, measurement remains passive.
A Concrete Example: When Demand Surges, Measurement Lags
Imagine a consumer brand sees a sudden spike in organic demand.
Brand search volume jumps
Direct traffic increases
Social mentions accelerate
What Happens Today
MMM will not reflect this until the next refresh. Incrementality is not running on this specific moment. The team hesitates or reacts based on gut feel or platform dashboards.
What Should Happen Instead
An orchestration layer detects:
Real-time signals: search demand, auction pressure, bid-landscape shifts
Historical learnings from MMM
The company's stated objectives, for example efficient growth versus aggressive capture
The system recommends shifting budget from Meta prospecting into branded search to capture existing demand while watching marginal returns.
A week later, MMM updates and validates whether that decision improved overall ROI. That learning feeds back into future recommendations.
This is the loop measurement alone cannot close.
Why MMM Still Matters in an Orchestrated System
If decisions are happening faster, why keep MMM?
Because MMM provides:
Long term validation
Structural understanding of channel effects
A stable baseline against short term noise
MMM becomes the memory of the system. Incrementality becomes the truth check. Together they form the inputs to a decision layer that any orchestrated system depends on. Platform data provides speed. Orchestration turns all of it into decisions.
What Comes Next
What comes next is not another dashboard or another static model.
It is a decision layer that:
Ingests MMM and incrementality outputs as signals, not verdicts
Combines them with real time platform data
Applies business goals, constraints, and strategic preferences
Makes recommendations while learning from outcomes
This is not about replacing human judgment. It is about augmenting it at the speed marketing actually operates.
Close the Loop or Keep Guessing
MMM and incrementality are foundational. Without an orchestration layer on top, they will always leave value on the table.
The teams that win next will not be the ones with the best models. They will be the ones that close the loop between what happened and what to do about it. As one growth lead put it after a year of running both MMM and incrementality without an orchestration layer: "'Cause if we're not gonna use it to make changes in how we plan, then what's the point?" He was right.
See which of your marketing dollars are actually working.
FAQ
Can MMM tell you which specific ads or campaigns to change?
No. Marketing mix modeling operates on aggregated time-series data and produces channel-level estimates, not ad-level or campaign-level direction. MMM identifies which channels drive impact but cannot specify which tactics inside those channels to adjust. Translating MMM outputs into platform-level actions requires a system that combines the model's signal with real-time platform data.
How long does an incrementality test take to produce results?
Well-designed incrementality experiments typically require 4 to 8 weeks to reach statistical significance. During the test window, the campaigns being tested must remain largely unchanged, which limits how many tests a team can run in a year and makes incrementality unsuitable as a real-time decision engine.
Why is having both MMM and incrementality testing not enough?
Each solves a different problem at a different speed. MMM provides a strategic, backward-looking view of channel contribution. Incrementality testing provides causal proof of a single lever's impact. Neither connects those signals to real-time platform data, business constraints, or each other. Teams end up better informed but still making decisions on gut feel or stale dashboards.
What does an orchestration layer add on top of measurement?
An orchestration layer ingests MMM outputs and incrementality results as inputs, combines them with real-time platform signals and business context (goals, constraints, risk tolerance), and surfaces specific recommendations. It closes the gap between knowing what happened and deciding what to do next.
How many incrementality tests can a marketing team realistically run per year?
Most marketing organizations run only 2 to 4 incrementality tests per year. Even the most mature programmes aim for 20 or more, which still falls short of covering all channels, audiences, creatives, and budget decisions continuously.
