What should companies doing (or thinking about doing) media mix modeling beware of?
At Recast, we’ve seen the biggest MMM challenges fall into three major categories:
- Diminishing Returns
- Data Limitations
- Time Shifts
Let’s break down why these issues are so difficult—and how they can be solved.
1. Diminishing Returns: Why More Spend ≠ More Results
One of the biggest pitfalls in MMM is assuming that every marketing dollar is equally effective and suggesting dumping more budget into a channel that has already hit its saturation point.
Why Diminishing Returns Happen
- Every marketing channel has a saturation point. Spending $10,000 on Facebook ads might yield a 5X ROI, but spending $1M won’t necessarily scale that ROI.
- Some channels saturate quickly (generally Meta, Google Search), while others take longer (TV, Out-of-Home).
- If an MMM treats all dollars as equally effective, it will overestimate your channel’s impact and waste your spend.
How Recast Models Diminishing Returns Correctly
At Recast, we address diminishing returns through a multi-pronged approach. Our saturation modeling estimates the incremental lift each additional dollar provides, rather than simply assuming a flat ROI.
By leveraging Bayesian nonlinearity, we capture how marketing effectiveness evolves at different spend levels, avoiding the pitfalls of linear assumptions. To ensure the reliability of these insights, our models undergo rigorous real-world validation against actual marketing experiments, confirming that the recommended budget allocations remain realistic and effective.
2. Data Limitations: Why MMM Needs the Right Inputs
MMM is not a big data problem—it’s a small data problem. Unlike digital attribution, which relies on massive real-time datasets, MMM has to extract insights from a relatively small historical dataset—which makes data quality incredibly important.
Common Data Problems:
- Missing Spend Data
This seems obvious, but many brands don’t properly track historical spend, especially for offline channels like TV or Out-of-Home. If you don’t know when and where you spent your budget, MMM just can’t measure impact accurately.
- Lack of Granularity
MMM works best with daily data, but some brands only track monthly or quarterly spend, which makes modeling less precise.
For example, if your model only sees “Facebook: $500,000 in Q3,” it misses day-to-day changes in performance and doesn’t get enough signal.
- Seasonality
Marketing results are highly seasonal. If your MMM only has one year of data, it can’t tell the difference between a good July vs. a normal seasonal spike. That’s why Recast requires 27 months of historical data to properly estimate seasonality.
3. The Time Shift Problem
A common pitfall in media mix modeling is the assumption that marketing impact is immediate. In reality, many channels produce delayed effects, and failing to account for these time shifts can lead to misattributed revenue and misguided budget decisions.
How Time Shifts Cause Bad MMM Decisions
Incorrectly attributing revenue to the wrong channels
For example, let’s say a customer sees a TV ad today but doesn’t purchase until two weeks later.
If the MMM doesn’t account for this delay, it might wrongly attribute the sale to a Facebook ad the customer clicked on right before purchasing.
Overestimating short-term channels
Some channels (e.g., Google Search Ads) drive instant conversions. Others (e.g., TV, influencers, email campaigns) have delayed effects—people might buy days, weeks, or even months later.
This leads to over-investing in “last-click” channels while ignoring the longer-term impact of branding and awareness efforts. Channels like paid search look great in an MMM if delayed effects aren’t modeled properly.
How Recast Handles Time Shifts
- Burn-In Period (First 90 Days)
Recast excludes the first 90 days of data from direct modeling, allowing marketing effects to settle. This prevents the model from overfitting to short-term noise.
- Dynamic Time-Shift Modeling
Instead of assuming fixed delays, Recast estimates time-shift effects for each channel based on real-world data. This allows the model to capture true channel performance over time.
4. Why Validation Matters More Than the Model Itself
Here’s a hard truth: Most MMMs are wrong.
If an MMM doesn’t account for diminishing returns, data limitations, and time shifts, it will waste millions in marketing spend by recommending bad budget allocations.
How to Validate an MMM Properly
- Out-of-Sample Forecasting: can the MMM predict future results when budgets change?
- Alignment with Experiments: does the MMM match lift test results from real-world experiments?
- Backtesting: if we train the model on past data, can it predict the next 3 months?
- Stability Checks: does the model give consistent insights week over week?
If an MMM fails these tests, it shouldn’t be trusted.
