Measurement science

Advertising measurement method selector

Published May 29, 2026. Updated June 3, 2026. Status: evergreen source page.

Most measurement mistakes begin before the analysis starts. A team asks one question, buys a method built for another question, and then treats the output as if it settles the decision.

This selector is a practical way to slow that down. Start with the decision, name the counterfactual, choose the method that can answer it, and keep every result inside the boundary of what the design can support.

Budget follow-up: if the budget decision worksheet shows the current evidence is below the needed rung, use the next-method explainer to choose between MMM calibration, geo lift, randomized holdouts, and brand studies.

Start with the decision

Decision question	Usually stronger method	What it still cannot prove alone
Should next quarter's budget move between channels?	MMM calibrated with credible experiments where available.	That every channel estimate is causally identified or stable across future conditions.
Did this campaign create incremental conversions?	Randomized conversion lift test with protected holdout and a fixed outcome window.	That the same lift will hold for a different audience, creative, season, or bid strategy.
Did a market-level launch or retail push work?	Geo, store, or matched-market test with balanced pre-period trends.	That one strong region represents the national average.
Did the campaign change awareness or consideration?	Brand study with comparable exposed and control respondents.	That survey movement caused profitable sales movement.
Which impressions were more likely to be noticed?	Attention measurement used as an exposure-quality diagnostic.	That attention is the same thing as incremental business value.
Which touchpoints appeared in observed paths?	Attribution or path reporting, labeled as descriptive.	That the last or weighted touch caused the conversion.

The method matrix

MMM

Best for budget planning when spend, outcomes, prices, seasonality, distribution, and other business drivers can be modeled over time. It needs transparent controls, priors, response curves, uncertainty, and calibration. Predictive fit is not enough for causal confidence.

Randomized lift test

Best when assignment can be controlled and treatment and holdout groups can be kept meaningfully separate. The readout should name the estimand, exposure compliance, primary outcome, confidence or credible interval, and generalization limits.

Geo or store test

Best when treatment is applied by location rather than person. The core check is whether treatment and control areas had similar pre-period levels and trends before the campaign changed anything.

Brand study

Best for directional perception outcomes such as awareness, recall, favorability, and consideration. It becomes weaker when response bias, small samples, proxy outcomes, and sales claims are hidden behind a single lift number.

Attention measurement

Best for diagnosing exposure quality across placements, formats, creative, and environments. It is useful when paired with a decision rule, but it should not be treated as a universal currency for business outcomes.

Attribution reporting

Best for describing observed paths, handoffs, and operational tracking gaps. It can guide QA and tactics, but it should not be used as the main proof that media caused demand.

Minimum disclosure checklist

Every serious report should state	Why it matters
The decision the result is meant to inform.	A method is only useful if it changes a real planning, buying, creative, or product choice.
The counterfactual being estimated.	Readers need to know what the campaign is being compared against.
The unit of analysis and unit of assignment.	User, household, market, store, and time-window designs have different leakage risks.
The primary outcome and readout rule.	Prevents a positive-looking secondary metric from replacing the planned decision threshold.
Important exclusions, model assumptions, and uncertainty.	Lets readers distinguish a robust result from a fragile point estimate.
Where the result should not be generalized.	Keeps a narrow but useful finding from becoming an inflated planning claim.

How methods work together

The strongest measurement programs do not force every question through one tool. Attribution can reveal broken tagging. Lift tests can calibrate causal expectations. MMM can turn multiple signals into planning ranges. Brand studies can explain whether a campaign changed memory or perception before sales data is mature.

The discipline is to keep each signal in its lane. A brand lift study should not become proof of profitable sales lift. A path report should not become incrementality proof. An MMM should not hide the experiment evidence that would make its causal claims stronger or weaker.

Fast diagnostic

If the report answers "what happened after launch," ask for the comparison that shows what would have happened anyway.
If the report answers "who converted," ask whether those people were already more likely to convert before exposure.
If the report answers "which channel ranked highest," ask how uncertainty and calibration changed the ranking.
If the report answers "what people remembered," ask whether the survey population is comparable and whether the outcome matches the decision.
If the report answers "which placement got attention," ask what business decision attention is supposed to improve.

Takeaway

Choose the method after the decision is clear, not after the dashboard is available. A weaker method honestly labeled is more useful than a stronger-looking method used outside its lane.