Guarded Autopilot
Automation you can hand to a client without losing sleep
Autopilot is not a rules engine with API keys. Every change begins as a typed proposal, passes a policy check, is previewed where the provider allows it, executes against a bounded object set and is verified by a follow-up read.
Risk tiers
What may run unattended
Spend exposure caps
Daily and per workspace
Blackout windows
Launches and client freezes
Verify and roll back
Follow-up read, then reversal
Decorative scene description: connected terrain zones for paid media, SEO and GEO, conversion rate optimisation and revenue outcomes send data streams into a central ABP.Marketing intelligence core.
Value
What changes for your team in the first month
Each claim below maps to a surface in the product and to a record you can open.
Automate the safe work, escalate the rest
Risk tiering separates reversible, bounded changes from material ones, so routine hygiene runs while judgement calls still reach a human.
Tier and route shown on every proposal before it runs
Give clients a control surface, not a promise
Caps, permitted action types, execution hours, approvers and blackout windows are configured per workspace and visible to the client in their portal.
Policy changes recorded in the audit trail
Prove what happened, every time
Each execution stores the exact request, the provider response, the verification read and the actor, so a change can always be explained after the fact.
Receipts are immutable and exportable
Governance
What the system actually does
- Risk tiers decide what can run unattended and what always needs a human.
- Spend exposure caps limit how much budget any automated change can move per day and per workspace.
- Approval matrices route material actions to the right internal or client approver.
- Blackout windows stop changes during launches, sales periods or client freezes.
- Every execution produces an immutable receipt: request, response, verification and actor.
- Reversible actions roll back automatically when verification fails or a performance threshold is breached.
Capabilities
Built as a governed workflow, not a dashboard
Every capability produces an auditable record: what was observed, what it was compared against, what was proposed and what happened next.
Policy
Workspace policy engine
Per-workspace policy defines eligible action types, tiers, caps, approvers, expiry windows and required evidence thresholds.
Safety
Preview and validate
Where the provider supports it, ABP.Marketing runs a preview or validate-only request and shows the projected provider response before execution.
Recovery
Verification and rollback
After execution the provider state is read back. Mismatches and threshold breaches trigger reversal and an alert to the owner.
Demo
A receipt that survives the awkward client question
Approval, bounds, request, response, verification and outcome are stored as one immutable record per execution.
Execution receipt
Illustrative interface- Action type
- Pause ad group · reversible
- Bounds
- 1 object · no budget change · expires in 24h
- Policy check
- Passed — within tier 1 and daily exposure cap
- Approved by
- Named approver · client role · 09:42
- Provider response
- Accepted · request ID stored
- Verification read
- State confirmed paused at 09:47
- Rollback
- Available · prior state recorded
Evidence model
Every finding shows its working
- Authorisation
- Who approved, when, on what evidence, and for how long.
- Bounds
- Object set, permitted values, spend exposure and expiry timestamp.
- Provider response
- Raw accepted response stored with the receipt, secrets redacted.
- Verification
- Follow-up read result and pass or fail status.
Boundaries
What we will not claim
- No external action can bypass the proposal, policy and audit layers.
- The language model has no unrestricted database access and no generic provider call tool.
- Nothing shows as executed until the provider accepted the request.
- Autopilot can be paused instantly per action type, workspace, tenant or provider.
Objection handling
Objections, answered directly
See it against your own accounts
Connect a read-only workspace and we will walk the Outcome Graph, the reconciliation and your first findings with you.