One-Page Executive Summary Template

Fill-in readout: headline with interval, channel table with evidence tiers, what we verified, what would change our mind — and no naked point estimates.

Headline result

Total KPI (window)total kpi

Marketing-attributed KPIattributed80% CI: [ low – high ]

Blended ROAS [interval]roas80% CI: [ low – high ]

Headline, in one sentence — with its interval:

headline result incl. credible interval

Channel summary

Every ROAS with its interval and its evidence tier — never a bare number.

Channel	ROAS [80% interval]	Evidence tier	Recommended action
________	____ [ ____ – ____ ]	________	________
________	____ [ ____ – ____ ]	________	________
________	____ [ ____ – ____ ]	________	________
________	____ [ ____ – ____ ]	________	________
________	____ [ ____ – ____ ]	________	________
________	____ [ ____ – ____ ]	________	________

Evidence tiers — Validated: consistent with experimental results (geo lift, RCT); act with confidence. Narrow, unvalidated: precise but never tested; reasonable for moderate budget changes — design validation tests. Wide, unvalidated: uncertain and untested; avoid major budget moves — prioritize experiments here, where testing has the highest information value.

Action vocabulary — Strong, consider increasing · Maintain current levels · Uncertain, recommend geo-test · Likely unprofitable, review · Reduce or eliminate.

What we verified

Convergence diagnostics clean — r-hat ≤ 1.01, zero divergences across chains.

Posterior predictive check passed — replicated data matches mean, variance, and autocorrelation of the observed KPI.

Out-of-sample backtest — holdout forecast error within the pre-agreed tolerance.

Sensitivity analysis — the headline holds across all pre-specified alternative specifications.

Data informativeness — posteriors contracted from priors for every channel we recommend acting on.

What would change our mind

Honest attribution names the evidence that would revise it.

The next experiment, already designed (see its pre-registration memo):

channel to testdesign — geo lift / did / flightingcostduration (weeks)

Decision it updates: if the test reads readout a we will action a; if it reads readout b we will action b.

No naked point estimates

Every number on this page carries its interval. A point estimate without uncertainty is not a decision input: an optimizer fed point estimates alone will pile budget onto whichever channel is most flattered by noise. Where an interval is too wide to act on, the action is an experiment, not a guess.