Trust in an automation tool shouldn't be a checkbox you tick on day one. With AdsPlane, how much the system is allowed to do is a ladder you climb deliberately — and you always control the pace.
Every seller starts at the bottom. Promotion between stages is explicit and earned, never automatic.
The five stages
Stage 0 — Shadow Mode. No execution at all. The engine runs the full Runbook and shows you the Manifest it would propose, complete with Data Confidence Scores. This is where new sellers evaluate AdsPlane and where you test new Guardrail Policies safely.
Stage 1 — Manual Gate. The default operating mode. Execution happens only after you approve — via the dashboard or a Telegram Command. You can approve the whole Manifest or just a high-confidence subset. Plenty of cautious sellers stay here long-term, and that's a perfectly good place to be.
Stage 2 — Guided Execution. Low-risk actions within your Guardrail caps can auto-approve, while high-risk categories — large bid increases, budget changes, campaign pauses — still require your explicit approval. You're notified of everything that executes automatically.
Stage 3 — Trusted Automation. The engine auto-executes all actions within
your Guardrail Policy, with circuit breakers watching for spend spikes, drift,
and API error rates — any of which trigger an automatic pause and a Telegram
alert. A morning Reconciliation digest tells you what happened, and /pause is
always one message away.
Stage 4 — Intraday Autonomy (Rally-Control). Optimization runs across multiple cycles a day, within tighter intraday spend caps and cooldowns. It's opted into separately and carries its own Guardrail Policy.
You set the pace
Moving up a stage always requires an explicit opt-in, and the higher stages require a track record of clean Reconciliation runs first. You can stop at any rung. You can step back down. And no matter the stage, your Guardrails bound what's possible and the emergency stop is always available.
That's the point: autonomy is something you grant as trust stabilizes — not something the tool assumes.
Read more about how the Runbook works, or see the full Autonomy Ladder.