# Marketing Measurement and Experiment Decision Record

Editable Markdown working material. Use for a measurement decision whether or not an experiment exists. Complete the experiment branch only for a proposed or actual study, and label its state.

## 1. Question and authority

- Record/version/date:
- Business question and decision horizon:
- Proposed action and accountable decision owner:
- Existing authority, budget/operational limits and conditions:
- Reporting question, predictive question or causal question:
- Approved data access, use and AI-provider disclosure:
- Evidence cut-off, timezone, currency and reporting basis:

An unknown field stays Unknown. A recommendation is separate from an authorised decision.

## 2. Metric and source definitions

| Metric ID | Definition / formula / unit | Numerator and denominator | Population, grain, window and exclusions | Source/version/owner |
|---|---|---|---|---|
| M1 | *Distinguish event, customer, order and revenue* | *Specify each* | *Include maturity and returns where relevant* | *Actual reference* |

| Source ID | Evidence/locator and extraction time | Observed / modelled / attributed | Identity/deduplication and permission limits | Quality or coverage gap |
|---|---|---|---|---|
| E1 | *Actual supplied source* | *Do not combine silently* | *What can/cannot be linked* | *Missing does not mean zero* |

## 3. Baseline and comparison

| Comparison | Baseline value/period | Compared value/period | Definition compatibility / maturity | Calculation and limits |
|---|---|---|---|---|
| *Metric ID* | *Supplied value or Unknown* | *Supplied value or Unknown* | *Compatible / Conditional / Not comparable, with reason* | *Show arithmetic; no unsupported inference* |

Record attribution model, lookback windows, interaction-date versus conversion-date reporting, consent/reporting changes, refunds/cancellations, cohort age and source reconciliation differences. Do not sum credited channel conversions as unique incremental sales. Keep prior definitions and later changes visible.

## 4. Checks: proposed versus performed

| Check | Expected result | Actual supplied result / Not performed | Source, date and version | Consequence / owner |
|---|---|---|---|---|
| *Tracking, duplication, missingness or reconciliation check* | *Agreed expectation* | *No invented test pass* | *Actual evidence* | *Blocked interpretation or next check* |

## 5. Study-design branch

Select: **No experiment / Proposed experiment / Conducted study with supplied evidence / Incomplete record**.

If no experiment exists, record the comparison's limitations and skip claims about randomisation, significance or causal lift. Observational causal inference also needs an explicit credible design and specialist review; it is not obtained by relabelling a dashboard comparison.

| Design element | Prespecified plan/reference | Actual implementation/evidence | Deviation or unresolved issue |
|---|---|---|---|
| Hypothesis, decision and target effect | *Before results where applicable* | *Record* | *Do not backdate* |
| Population, assignment/exposure unit and comparison | *Design rationale* | *Actual counts/logs* | *Selection, spillovers or interference* |
| Primary outcome and guardrails | *Metric definitions, horizon* | *Observed results* | *Metric changes* |
| Sample/information and uncertainty approach | *Inputs and specialist rationale* | *Actual method/results* | *Do not invent power or validity* |
| Duration, stopping rule and exclusions | *Original rule and approval* | *Actual timing and exclusions* | *Peeking, post hoc choices, missingness* |
| Assignment/exposure/analysis integrity | *Required checks* | *Actual checks, including imbalance where relevant* | *No automatic pass from a favourable result* |

## 6. Interpretation and alternatives

| Proposition | Observed result / attribution / causal claim / hypothesis | Evidence and contrary evidence | Uncertainty or assumption | Smallest discriminating check |
|---|---|---|---|---|
| *One proposition per row* | *State category* | *Use source IDs* | *Design limits, intervals only if supported* | *Proposed, not completed* |

Separate average from marginal return; revenue from contribution/profit; credited from incremental outcomes; absence of statistical significance from evidence of no effect. If a causal interpretation is proposed, state identification assumptions and what evidence supports them. Do not choose a universal significance threshold or decide legal compliance.

## 7. Human interpretation and next decision

- Analyst review and actual competence/scope, or Pending:
- What the evidence supports:
- What it does not support:
- Decision boundary / agreed criterion and whether it was set before results:
- Alternatives and trade-offs, including Hold / Investigate / Bounded test:
- Human disposition, rationale, date and evidence reference, or Pending:
- Approved action/version/ceiling and operational owner, or Not authorised:
- Monitoring, stop condition and review trigger to be agreed:
- Actual action and later verification evidence, or Not performed:

## 8. Change and learning record

| Date/version | New evidence or definition change | Effect on interpretation/decision | Owner and re-review status |
|---|---|---|---|
| *Complete* | *Preserve the original decision record* | *Explain* | *Open until evidenced* |

This is a working record, not statistical validation, approval to spend or evidence of causality merely because a form is complete. AI output requires professional review. Apply the organisation's actual access, retention and deletion rules; no period is supplied here. No new licence is implied.