Priors

In Bayesian modeling, evidence-based beliefs about how a variable is likely to behave before new data is analyzed. In MMM, priors might encode expected ROAS ranges for a given channel, informed by past experiments, industry benchmarks, or validated test results.