A 10–12 week engagement is one pass around the measurement loop: fit (T₀) → prioritize what to test (T₁) → pre-registered experiment (T₂) → calibrated refit (T₃) → allocate (T₄) → re-evaluate (T₅). Each row names the deliverable that gates the next week.
| Week | Phase | Activities | Deliverable | Framework feature |
|---|---|---|---|---|
| W1–2 | Data onboarding + EDA gate | Data inventory, MFF assembly, hygiene checks; EDA pre-flight (outlier, CoV, collinearity, demand-chasing screens; causal-role classification) | Signed Data Onboarding Checklist; EDA gate readout | MFF loader validation; mmm_framework.eda |
| W3 | Pre-specification | Define estimand and decision questions; declare causal roles; pre-register adstock, saturation, priors, trend/seasonality; identification-assumptions sign-off | Pre-registered model specification | Modeling-guide Phase 1 (Plan); the identification contract |
| W4–5 | Fit + validation battery (T₀) | Fit; convergence diagnostics (non-negotiable); posterior predictive check; out-of-sample backtest; pre-specified sensitivity analysis | Validation battery report | BayesianMMM; modeling-guide Phase 3 (Validate); parameter-learning diagnostic |
| W6 | Readout 1 — attribution with tiers | Channel ROAS with intervals; evidence-tier classification (validated / narrow-unvalidated / wide-unvalidated); recommended actions per tier | One-Page Executive Summary | MMMReportGenerator; interpreting-results conventions |
| W7 | Experiment design + pre-registration (T₁–T₂) | EIG/EVOI prioritization of what to test; geo-lift / matched-market DiD / flighting design with power calculation and power-ceiling check; memo signed by analyst, practice lead, client | Signed Pre-Registration Memo | Experiment priorities (EIG/EVOI); design studio (measurement-calibration) |
| W8–10 | Experiment in field | Launch per memo; monitor delivery, not results — interim looks only if pre-specified; deviations logged; intention-to-treat by default | Field log + amendments (if any) | Experiment lifecycle registry (planned → running → completed) |
| W11 | Calibrated refit + final readout (T₃) | Fold the experiment readout into the model as a calibration; refit; compare pre/post-calibration intervals on the tested channel | Final readout — calibrated attribution | Experiment calibration at refit (readout → narrowed posterior) |
| W12 | Allocation + next-cycle plan (T₄–T₅) | Budget-allocation recommendation under uncertainty; re-evaluation: information decay schedule, next test priorities, refresh cadence | Allocation memo + next-cycle test plan | T₄ allocate / T₅ re-evaluate; information decay triggers re-tests |
Week 12 is not the end: experimental information decays as markets drift, so T₅ re-evaluation feeds the next cycle's T₁ priorities — each engagement leaves the client with narrower intervals and a standing test plan, not a one-off report.