Recommendation types
| Type | Example | Primary review question |
|---|---|---|
| Budget shift | Move spend from a weak cohort to a stronger one | Is the stronger cohort reliable enough to scale? |
| Creative pause | Pause fatigued creative with declining conversion | Is the decline real or temporary noise? |
| Audience shift | Exclude a low-LTV segment | Will exclusion reduce waste without harming scale? |
| Offer test | Test a different trial length or plan emphasis | Is the test safe for revenue and brand? |
| Paywall experiment | Test paywall messaging or layout | Is product approval required? |
| Bid adjustment | Change target CPA or bid rules | Does the new target respect payback and budget limits? |
Review checklist
Before approving, confirm:- The recommendation has a clear reason.
- The expected impact is tied to payback, ROAS, CAC, LTV, conversion, or retention.
- Cohort size is large enough to act on.
- Attribution and spend data are current.
- Guardrail checks passed.
- The affected campaign, market, audience, creative, or offer is in scope.
- Rollback conditions are defined.
- The owner and review window are clear.
Review workflow
Read the reason
Start with the signal behind the recommendation. Look for the cohort, metric, date range, and comparison baseline.
Check data freshness
Confirm the relevant revenue, attribution, spend, and event data has synced recently enough for the decision.
Check confidence
Review cohort size, historical consistency, and whether the signal repeats across more than one reporting window.
Confirm guardrails
Make sure the action stays inside spend caps, campaign exclusions, approval rules, action type settings, and rollback triggers.
Choose the decision
Approve, reject, defer, or convert the recommendation into a manual test. Do not approve just because the action looks plausible.
Record context
Add a note explaining why the decision was made. This helps the team learn from outcomes later.
Decision criteria
| Decision | Use when |
|---|---|
| Approve | Data is current, guardrails pass, confidence is acceptable, and the action is in scope |
| Reject | The recommendation conflicts with strategy, data quality is poor, or the action is unsafe |
| Defer | More data is needed or another campaign/event is about to change the baseline |
| Manual test | The idea is useful but should be executed as a controlled experiment |
Approval notes
Good approval notes are short but specific:- “Approved for Meta annual cohort test. 90-day ROAS is above target and spend cap remains under $12k/day.”
- “Deferred until App Store revenue sync completes. Current renewal data is incomplete.”
- “Rejected because campaign is part of brand holdout and excluded from automation.”
What to monitor after approval
| Signal | Why it matters |
|---|---|
| Spend | Confirms the action did not exceed budget caps |
| CAC | Shows whether acquisition cost improved or worsened |
| Trial conversion | Captures near-term funnel impact |
| Renewal and refund behavior | Confirms revenue quality |
| ROAS and payback | Shows whether the action improved durable growth |
| Rollback trigger | Protects against unexpected regression |
First-review recommendation
For the first two weeks, review recommendations manually. Do not enable automatic execution until:- At least one payback report has been validated.
- Guardrails have been reviewed by the growth owner.
- The team has approved or rejected several recommendations.
- Rollback triggers have been tested or reviewed.