Data & Analytics Practical insights
Use brand signals in MMM without double counting
Review how brand search enters a marketing mix model, distinguish demand controls from mediated effects, and avoid counting the same contribution twice.
Brand search can reflect both existing customer demand and the effects of marketing. That makes it useful for measurement and easy to misinterpret. Google’s September Meridian update describes support for relevant brand signals. [1] Choosing where those signals belong requires a causal question, not simply another column in the dataset.
Draw the proposed relationship
Consider a video campaign that increases awareness, followed by more branded searches and purchases. In that proposed pathway, branded search partly sits between advertising and sales. If an analyst automatically controls away all changes in brand search, the model may remove part of the effect the team wanted to estimate. Conversely, existing demand can influence both searches and purchases. A measurement specialist must distinguish these roles.
Define total and direct effects
Decide whether the business question concerns the campaign’s full contribution or a contribution excluding a particular intermediate route. Document the timing assumptions and alternative explanations. Do not report the full video effect and then add the same search-mediated effect again as independent value. A neat spreadsheet total can conceal that double counting.
Run a structured sensitivity review
Ask the modeler to compare plausible specifications and explain why results move. Check lag choices, signal construction, geographic coverage and correlation with other media. Branded queries can also change after publicity, service problems or distribution changes; their movement is not automatically evidence of stronger brand preference. AI can summarize the model notes, but it should not choose causal assumptions from column names.
Use evidence that can challenge the story
Bring experiments, campaign timing and independent brand research into the review where available. Preserve uncertainty instead of selecting the version with the largest brand contribution. Before your next MMM update, prepare a one-page causal diagram and ask the measurement owner which paths the model estimates, which it blocks, and which remain unidentified. That discussion is more valuable than an unexplained extra feature.
Sources and evidence
Sources checked on 4 October 2026. Proposed workflows and hypothetical examples are editorial analysis.
From insight to practice