Insights
Short perspectives on the assumptions, limits, and choices that make a measurement system useful.
Attribution observes conversions associated with an interaction; incrementality estimates what would have happened without the spend. Treating them as the same can overvalue channels that capture existing demand. The practical answer is not to discard attribution, but to combine it with tests, aggregated models, and a clear decision definition.
When a cohort is still young, two plausible curves can lead to very different budget choices. A single expected ROAS hides that risk. Ranges, scenarios, and decision rules let teams act earlier without pretending uncertainty has disappeared.
A marketing mix model cannot repair inconsistent definitions, unrecorded changes, or granularity that does not match the decision. A prior diagnostic establishes whether the model will be actionable, which data is missing, and whether another form of evidence could answer the problem sooner.
Earlier posts on data engineering, visualization, tennis, and civic data remain available, separate from the current focus.
Explore the archiveA focused first conversation
In an introductory conversation, we can frame the problem, review the available evidence, and decide whether there is a useful next step.
Discuss your challenge