Guardrail one: what should the hard cap be?
The daily spend cap is the guardrail that defines the worst realistic outcome, so set it as the amount you could lose in a day without a difficult conversation. On ftes.ai it is a hard limit the Performance Ads FTE cannot exceed, enforced by the platform.
Set two levels: a per-campaign cap and an account cap the campaign caps sit inside. Budget changes are proposals the manager approves, so raising either cap is a deliberate act with a logged reason.
Guardrail two: which actions keep their gate permanently?
New creatives and new audiences keep their launch gate permanently, at every autonomy level, because a launch is a first-of-its-kind action with asymmetric downside. A wrong creative is a public mistake; a wrong audience is wasted spend and, occasionally, a policy problem.
What can earn autonomy is optimisation of campaigns already running: pausing losers and scaling winners inside caps. That distinction, launches gated and optimisation promotable, is the shape of the autonomy ladder for this role.
- Always gated: new creatives, new audiences, new platforms, budget increases
- Promotable: pausing underperformers, scaling winners inside caps, bid adjustments
- Always escalated: policy rejections, account warnings, anything reputational
Guardrail three: how do you write creative rules?
Write creative rules as prohibitions rather than preferences, because prohibitions are enforced and preferences are interpreted. Claims it may never make, comparisons it may never draw, imagery it may never use, and regulated language it must avoid.
The Performance Ads FTE runs a policy pre-check against platform advertising rules before proposing a launch, but platform policy is a floor. Your brand rules are the ceiling, and they only apply if they are written down.
Guardrail four: what must always escalate?
Policy rejections, account warnings, and anything with reputational weight must always escalate to a named person rather than being handled automatically. A rejected ad can be resubmitted; an account restriction is a different order of problem and needs a human immediately.
Add anything specific to your business: a competitor mention, a pricing claim, a regulated product category. The escalation list is short to write and prevents the incidents that damage trust in the first month.
How do you promote autonomy safely?
Promote optimisation autonomy from the approval history rather than from confidence in the system. If pause proposals on underperforming ad sets have been approved repeatedly without amendment, that class is ready to run inside caps. Scale proposals usually earn promotion later, because their downside is larger.
Keep every promotion reversible. A period of unusual sensitivity, a new product, or a run of send-backs is reason enough to step a class back to propose-only, and doing so should be a single uneventful action.
How do the guardrails appear in the workspace day to day?
The guardrails are not a policy document; they are what the workspace shows. The hard cap appears in the morning brief as spend against the cap for each account, with a warning as it approaches, and the FTE cannot propose a change that would breach it. The permanently gated actions appear in the Approvals tab as launches and budget increases waiting for a named person, each with the creative, the audience, the projected spend, and the reasoning attached, and they never earn auto-approval on the ladder.
The creative rules live in the Brand Brain and the remit, and a draft that breaches a prohibited claim fails the fact gate before it reaches you. The always-escalate conditions, such as a sudden spend spike, a platform policy flag, or a performance collapse, appear as detections in the brief with the evidence and a proposed response. Every one of these is logged in the audit trail with who decided what and when. A manager checking the guardrails does not read a document; they open the brief, the Approvals tab, and Setup, and the guardrails are what they see.
Review the four guardrails at the monthly performance review rather than only after a scare. Read spend against the cap for the month, the launches and budget changes approved and by whom, any creative that failed the fact gate, and the detections that fired, and ask whether each guardrail is set where the business now wants it. A cap set for a quiet quarter is usually the first one that needs revisiting before a launch.