Marketing response analysis
Map the CSV once, then run only the trend, cannibalization, MMM contribution, or regression forecast you need.
End of analysisNext steps and reference material below. Not part of the analysis result.
How is channel contribution estimated?
Fit a regression to a weekly spend and outcome panel. Not causal.
Method and FAQOpen only when you need the methodology
See diagnostic criteria and next steps
When to use it
Use this before scaling brand or retargeting spend, or when paid outcomes rise while organic or branded outcomes weaken. The signals prioritize a holdout or on/off test; they do not prove causality by themselves.
What to inspect
Review delayed response after a spend increase, movement against organic and branded outcomes, overlap between channels, and model fit. Read the four signals together rather than labelling one correlation as cannibalization.
How to act
Do not turn off a suspect channel immediately. Keep a controlled slice by geography, audience, or time, then compare the incremental outcome. Use budget allocation only after confidence and marginal efficiency support a move.
Frequently asked questions
Does a large regression coefficient prove the channel drove results?
No. Regression shows association, not causation. Confirming an effect requires an experimental design such as a holdout or incrementality study.
What happens when channel spends move together?
Multicollinearity makes channel-level contribution unidentifiable. The tool refuses the decomposition and explains why instead of printing a confident-looking number. Check this in advance with the VIF diagnostic.
Is the forecast band a confidence interval?
No. It is a reference range from historical residuals, not a causal confidence interval. Structural change widens the real error.