Client
Growth / ads team
Service
Implement
Status
Shipped
Stack
Multi-agent system · validator · governance rules in code
A growth team's ad audit agent.
A team of specialized agents that audits an ad account and writes a safe, ready-to-run optimization plan. An independent validator checks every output.
The brief
A growth team managed ad spend across many accounts. A proper manual audit took a senior analyst hours per account, and the depth varied with whoever ran it. They wanted it automated, without the risk that an AI confidently suggests a change that breaks a live campaign.
What we built
A team of specialized agents rather than one model doing everything. Analysis is kept separate from recommendation, recommendation from safety review, and an independent validator checks every hand-off between them, so no single model’s judgment goes unchecked. The output is a ready-to-run optimization plan, plus the report that justifies it.
Account data
Step 1Analyze
Step 2Propose
Step 3Safety check
Plan + report
Independent validator · structural and semantic checks on every hand-off · repair, retry, or escalate
What makes it hold up
Real platform governance is encoded as hard rules in code, not left to the model. The kinds of changes that can destabilize a live campaign are blocked outright, and anything borderline is routed to a person for sign-off. When an output fails a check, the system repairs or regenerates it, and escalates instead of guessing.
Where it landed
The audit that took a senior analyst hours per account now runs on demand, at the same depth every time. The team reviews a finished plan instead of assembling one, and nothing goes live without clearing the same checks, no matter who runs it.
AgentsOne job eachvalidated at every hand-off
CoverageFull accountcampaign structure to creative
Self-correctionBuilt inrepair, retry, or escalate
Read another case
All work → 
