Step 1: record the account's baseline and set the cap below it
Record the ad account's last two months: spend, the platform's reported results, any independent attribution you trust, how often the account was actually changed, and who changed it. Then set the hard cap for the pilot below the baseline spend, so that the campaign cannot cost more than the account already did and the comparison is conservative by construction.
The cap lives at the account level in the Performance Ads Digital FTE's Setup and cannot be breached by any proposal; the FTE plans within it and the morning brief shows spend against it daily.
Step 2: connect the account and keep the two budgets apart
Connect the ad account with a dedicated user scoped to the campaigns in the pilot, able to read, propose, and, once a class is promoted, adjust within caps, but not to create campaigns or raise budgets without approval. The FTE's own token budget for analysis, drafting, and reporting sits in the organization's wallet; the ad spend sits on the platform. The brief shows both, and the two never mix.
Connect the conversion data your site can supply honestly, because attribution is only as good as the events it can see.
Step 3: write the creative rules and the permanent gates into the remit
In the remit, write the creative rules from the Brand Brain: voice, prohibited claims, protected audiences, and any regulated statement that always needs a named approver. Write the permanent gates: every new campaign launch, every budget increase above a threshold you set, and any creative touching a regulated claim wait for a person at every level. Write the detections that always escalate: a spend spike, a platform policy flag, a performance collapse on one creative.
The fact gate then checks every creative against the prohibited claims before it reaches Approvals.
- Creative rules: voice, prohibited claims, protected audiences
- Permanent gates: launches, budget increases above threshold, regulated claims
- Always-escalate detections: spend spike, policy flag, collapse
- Cap: at the account level, unbreachable
Step 4: approve the launch with its full package, then read the brief daily
The launch arrives in Approvals as one package: the audience, the creative that passed the fact gate, the projected spend against the cap, and the reasoning. Approve, edit, or send back. Once live, the brief each morning shows spend against the cap, the platform's reported results, the independent attribution beside them, and any detection with a proposed response, such as pausing a creative, that waits for approval.
Optimisation proposals inside the cap, such as shifting budget between ad sets, are gated in the first cycle and build the record.
Step 5: review after a full cycle and decide what optimisation can earn
At the review, compare against the baseline on the same measures, and read what the baseline rarely showed: how many days the account was touched, in the audit trail, and what dual attribution said when the two numbers disagreed. Then decide whether in-cap optimisation of a specific kind has earned autonomy from a clean record. Launches and budget increases are not candidates at any review.
Revisit the cap for the next cycle deliberately; a cap set for a quiet quarter is the first thing to reconsider before a launch.
What does done look like?
Done is a campaign that ran inside a cap it could not breach, launched with a full package a person approved, optimised with proposals a person saw, reported with two attribution numbers rather than one, and reviewed against a baseline recorded before it started. The marketing lead read the brief in minutes each morning and spent an hour on the review, and knows exactly how often the account was touched and by what decision.
The performance view will mark unmeasured what it cannot measure, which for ads is usually the part of attribution that no platform can see; that honesty is the result, not a limitation.
What should you refuse to do in the first campaign?
Refuse to raise the cap mid-cycle to chase a result; the comparison depends on it. Refuse to promote any launch or budget class; they are gates, not ladders. Refuse to read the platform's attribution alone. And refuse a creative the fact gate blocked by approving it with a source added afterwards; send it back and let the FTE revise.
Each refusal is what makes the first campaign a controlled experiment rather than a spend.