Marketing Experiment Review
Turn marketing activity into a durable evidence and decision loop:
```text hypothesis -> asset or campaign -> measured response -> qualified learning -> next test ```
This method prepares readouts and ledger-ready entries. It does not silently publish assets, change spend, alter measurement, or write to a CRM, project system, or marketing ledger without the host's explicit approval and capability.
Context and evidence
Use the current plan, prior experiments, asset or campaign details, time window, and available evidence: analytics, platform reports, qualified responses, sales notes, revenue, or direct feedback. State evidence gaps clearly. A weak signal is a hypothesis, not a strategy verdict.
Read `references/learning-ledger-schema.md` and `references/operating-cadence.md`. Read `references/host-operations.md` only when a user asks for a host-system read or write.
Evidence quality
| Level | Evidence | Permitted decision | |---|---|---| | Anecdotal | One comment or internal impression | Record as hypothesis | | Directional | Small-sample metric or recurring qualitative signal | Cautious iteration | | Operational | Repeated tracked pattern | Adjust the plan | | Economic | Qualified opportunity, revenue, or margin evidence | Inform investment | | Causal | Controlled or strong incrementality evidence | Scale or stop confidently |
Workflow
1. **Recover the hypothesis.** Identify buyer, channel, asset, intended action, metric, owner, and prior decision. 2. **Collect and qualify evidence.** Separate instrument status, platform reporting, behavior, qualified response, and economic evidence. 3. **Locate the first material blocker.** For paid or funnel work, distinguish tracking, reach, creative, offer, landing path, audience, follow-up, and budget before proposing a change. Record health/readiness, tracking status, creative diversity, landing status, the decision metric, and review date. 4. **Classify the learning.** Record what changed for audience, message, channel, offer, experience, follow-up, measurement, or economics. 5. **Make one decision.** Continue, change, stop, fix measurement, or escalate for financial analysis. State confidence and what would reverse it. 6. **Prepare the next test.** Specify one change, owner, measurement, review window, and stop/continue condition. 7. **Prepare the durable record.** Produce a ledger-ready entry and a host handoff. Do not claim that it was saved unless the authorized host operation completed.
Output format
```markdown ## Learning Readout - Period: - Current proof level: - Decision:
## What we learned | Learning | Evidence | Quality | Confidence |
## First blocker and one change next - Blocker: - Change: - Metric: - Review window:
## Financial implications - What is proven: - What remains a proxy: - What evidence is needed:
## Ledger-ready entry and host handoff ```
Quality rules
- Preserve failed tests; a negative result can still be useful evidence. - Do not let clicks, impressions, or anecdotes alone prove a message, audience, or investment decision. - If measurement is unreliable, the decision is to repair measurement before changing the audience or bid. - For paid work, repair a broken tracking, landing, or follow-up path before changing creative or audience; otherwise the test cannot distinguish causes. - Keep historic records append-only; distinguish a proposal from a completed host write.