Simulated marketing benchmark: learn to construct a comparable interval.
Answer:
This page presents eleven entirely simulated profiles. They observe no sector and cannot be used as a target or evidence of performance. Their sole educational use is to show how to document a metric, filter a comparable distribution, weight proximity to an internal case and update a prior. A real benchmark requires verifiable population, period, provenance, licence, definitions and selection bias.
Method and decision:
Audit formula, unit, horizon, currency, date, source and exclusions, keeping irreconcilable observations separate. Filter on factors that structurally change the metric and disclose sample size, median, interquartile range and extremes. Weight each reference by similarity of margin, channel, maturity and market, and test whether excluding dominant references changes the result. Combine the external prior with internal evidence according to their precision. Use a benchmark to bound a hypothesis or spot an anomaly, then decide on internal economics and local evidence.
Worked example — simulated data:
Eleven fictitious profiles have a weighted mean ROAS of 2.19 and population weighted standard deviation 0.49. An equally fictitious internal measurement is 1.70 with standard deviation 0.40. External precision is 1 ÷ 0.49² = 4.16; internal precision is 1 ÷ 0.40² = 6.25. The combined estimate is (2.19 × 4.16 + 1.70 × 6.25) ÷ 10.41 = 1.90. Its standard deviation is √(1 ÷ 10.41) = 0.31, giving an approximate 95% interval of 1.29 to 2.50. More precise internal evidence moves the prior from 2.19 to 1.90. The decision uses the internal contribution interval; simulated profiles never become a performance target.
Limits:
The eleven profiles are simulated, not a sector benchmark. Published data often selects exceptional results. Similarity weighting remains a judgement to test, and a changing market can quickly make even real distributions obsolete.