Strategy
Part of Building a marketing mix model that survives scrutiny, step by step
The marketing mix modeling questions worth asking before anything else
Common marketing mix modeling questions cover data, causality, channels, ROI, experiments, refresh timing, ownership, uncertainty, and decision limits.
What to take away
- MMM estimates effects under a stated model; it does not read channel truth directly from a ledger.
- Business teams should ask about the decision, data, assumptions, uncertainty, and action together.
- A channel ranking without feasible ranges and claim limits is incomplete.
Common marketing mix modeling questions usually surface when a model reaches a budget meeting. Clear answers keep an aggregate statistical result from being mistaken for a platform report or a guaranteed plan.
Ten questions leaders should ask
| Question | Short answer |
|---|---|
| What is MMM? | Aggregate modeling of marketing and other outcome drivers |
| What does it estimate? | Conditional channel effects, curves, and scenarios |
| Is it causal? | Only under explicit causal assumptions |
| Does it need user-level data? | No, designs commonly use aggregated time and geo data |
| Can it include offline media? | Yes, when exposure and spend are defined consistently |
| Why include controls? | To represent relevant outside drivers and confounding |
| Is ROI exact? | No, it is an estimate with economic and statistical uncertainty |
| Does fit select the winner? | No, predictive fit alone cannot validate causal effects |
| Can experiments help? | Yes, when their estimand and setting are aligned |
| Who owns action? | A named business owner, supported by model and data stewards |
Google's introduction to Meridian covers three stages of its open-source MMM: input, modeling, and post-modeling. It sits at the modeling end of the marketing analytics tools market, where the product is a method, not a report.
Ten MMM questions to ask
- What is MMM?
- What does it estimate?
- Is it causal?
- Does it need user-level data?
- Can it include offline media?
- Why include controls?
- Is ROI exact?
- Does fit select the winner?
The page lists historical ROI, response curves, and budget allocation as questions the product addresses. Those capabilities depend on the user's model and data.
What MMM does not settle
- The single correct channel taxonomy
- Whether every important confounder was observed
- A guaranteed future return
- The value of a lead or customer without an economic rule
- Whether a vendor's model fits a buyer's risk
- Which operational constraint can be ignored
Meta's Robyn project introduction identifies ridge regression, time-series decomposition, automated parameter search, and budget allocation as parts of that package.
What MMM does not settle
- The single correct channel taxonomy
- Whether every important confounder was observed
- A guaranteed future return
- The value of a lead or customer without an economic rule
- Whether a vendor's model fits a buyer's risk
- Which operational constraint can be ignored
Ask the modeler to show the outcome series, spend range, controls, uncertainty, and at least one sensitivity result before discussing allocation. Ask finance to confirm the value and cost basis. Ask channel owners whether proposed movements can be delivered.
End the review with a dated decision record. If evidence is too weak, choose a smaller change, a test, or no change. A holdout or geo test belongs to campaign measurement, which settles in one period what a model can only estimate. Uncertainty is a planning input, not a reason to disguise a guess as precision.
Resolve the question with a decision record
A useful answer identifies the decision owner, eligible population, evidence period, method, and action. It also states what the result cannot establish. If the same question returns, compare the records before changing the metric.
The GAO evaluation design guide explains how evaluation questions, evidence needs, and design choices fit together. Use its design discipline as a check on the method, not as proof that a marketing result is causal.
The W3C Privacy Principles statement gives system designers a shared vocabulary for privacy and warns against shifting privacy work onto individuals. Apply that principle to the data flow behind these questions. It does not replace the law, contract terms, or consent analysis.
Common questions
Can MMM measure brand advertising?
It can estimate aggregate effects when the outcome, variation, controls, lag, and assumptions provide enough information. Weak signal should be reported, not forced.
Is MMM only for large companies?
No, but the scope must match the available data and decision. Fewer modeled treatments may produce a more honest result.
Why do two MMMs disagree?
They may use different data, outcomes, controls, priors, transformations, periods, geographies, estimands, or model-selection rules. Fixing those choices in advance is most of the work in measuring marketing mix modeling.
