Guardrails define the difference between observation, recommendation, and autonomous execution.
Recommended guardrails
- Daily and weekly spend caps
- Maximum budget movement per action
- Campaigns, channels, countries, or audiences FloKit cannot modify
- Minimum confidence threshold for automated execution
- Required approvals for pricing, paywall, and offer changes
- Rollback triggers for CAC, ROAS, revenue, or trial conversion regressions
Keep write access disabled until your tracking, revenue data, and approval process are verified with historical actions.
Approval model
Start with human approval for every action. After the team trusts the signals and rollback behavior, allow lower-risk actions to execute automatically inside narrow limits.