Skip to main content
Goal: Enable FloKit to execute approved actions on your ad platforms, starting with the lowest-risk action types and expanding as trust is established. This recipe describes a four-phase progression. Each phase builds the confidence and operational baseline required for the next. Skipping phases increases rollback rates and slows trust recovery when regressions occur.

​
Before you start

  • At least 30 days of payback data in FloKit, validated against your source of truth (see Validate CAC payback).
  • At least 5 actions that your team reviewed, approved manually, and executed on the ad platforms — with outcomes recorded in Actions → Outcomes.
  • Guardrails configured in FloKit → Settings → Guardrails.

​
Phases

1

Phase 1 — Read-only (current state)

FloKit generates recommendations and populates the action queue. Your team reviews each recommendation and executes any approved changes directly on the ad platform — FloKit does not touch ad platforms in this phase.What this phase teaches you:
  • Whether FloKit’s recommendations align with your team’s intuition.
  • Which action types are most consistently correct.
  • Whether your payback data is stable enough to act on.
Guardrail configuration for Phase 1:
  • No write access to any ad platform.
  • No guardrails required yet — no actions are being executed.
Duration: Until you have 5 manually executed actions with recorded outcomes and confirmed payback data accuracy.
2

Phase 2 — Approval-based execution

Connect write access to your ad platforms. FloKit now prepares changes and executes them after your team approves — but every action still requires explicit approval before anything changes on the platform.Start with the lowest-risk action type: creative pauses. Creative pauses are reversible, have limited blast radius, and produce clear outcome signals within 48–72 hours.What this phase teaches you:
  • That FloKit’s execution matches what the recommendation described.
  • How long actions take to produce measurable outcomes.
  • Which rollback triggers make sense for your account size and volatility.
Guardrail configuration for Phase 2:
  • Daily spend cap: ≤20% of current daily budget.
  • Exclude any campaign in an active manual A/B test.
  • Confidence threshold: 85%.
  • Require manual approval for all action types.
  • Enable rollback trigger: CAC >25% regression within 72 hours.
  • Enable only creative_pause action type initially. Expand to budget_shift after 5 successful creative pauses.
Duration: Until you have 5 completed creative_pause actions, all within expected outcome range.
3

Phase 3 — Selective automation

For action types where you have seen consistent positive outcomes across multiple cycles, configure auto-approval below a spend threshold. Actions above the threshold still require manual approval.Example configuration for creative pauses:
  • Auto-approve creative_pause actions with under $500/day impact and >85% confidence.
  • Require manual approval for creative_pause actions with over $500/day impact.
  • Manual approval still required for all other action types.
Monitor the Action outcomes report weekly to verify auto-approved actions are performing within expected ranges.Guardrail configuration for Phase 3 (example):
  • Auto-approval threshold: creative_pause under $500/day, confidence >85%.
  • Budget shift: still manual approval only.
  • Rollback trigger: CAC >20% regression within 72 hours, auto-rollback enabled.
  • Spend cap: ≤25% of daily budget (slightly expanded after Phase 2 validation).
Duration: Minimum 30 days. Expand to additional action types only after consistent positive outcomes.
4

Phase 4 — Monitored automation

Broader auto-approval coverage with tighter rollback monitoring. FloKit operates autonomously within your guardrail boundaries and rolls back on regression signals without waiting for manual review.What changes in Phase 4:
  • Auto-approval extended to budget_shift actions below a spend threshold.
  • Rollback is automatic, not flagged-for-review.
  • FloKit sends a Slack or webhook notification for every auto-approved action and every rollback.
What doesn’t change:
  • Spend caps remain in place.
  • Excluded entities remain excluded.
  • Actions above the per-action spend cap still require manual approval.
  • FloKit never modifies pricing, paywalls, or subscription offers.
Guardrail configuration for Phase 4 (example):
  • Auto-approve creative_pause up to $2,000/day impact.
  • Auto-approve budget_shift up to $5,000/week, confidence >85%.
  • All other action types: manual approval.
  • Rollback trigger: CAC >20% or ROAS below guardrail floor, auto-rollback within 4 hours.
  • Weekly review of action outcomes is still expected — automation is not a substitute for oversight.
Never skip Phase 1 and 2. Teams that move to execution without validated payback data and a manual approval baseline tend to experience higher rollback rates and take significantly longer to build confidence in FloKit’s recommendations. The 30-day manual review period is the fastest path to safe automation — not a detour around it.

​
Guardrail summary by phase

SettingPhase 1Phase 2Phase 3Phase 4
Write accessNoneConnectedConnectedConnected
Auto-approvalNoneNoneSelectiveBroad
Spend capN/A≤20% daily budget≤25% daily budget≤30% daily budget
Confidence thresholdN/A85%85%85%
Rollback triggerN/A25% CAC regression20% CAC regression20% CAC / ROAS floor
Auto-rollbackN/ANo (flag only)OptionalYes