AI ad ops for agencies — many client accounts, one goal, per-account approval
The agency problem isn't doing the work once — it's doing it forty times without either drowning or making the same mistake across every client at once. Agent Planners fans one plain-language goal across a manager account with a per-account dry-run, keeps per-client guardrails in per-account memory, isolates each client in its own organization, and produces individually reviewable proposals rather than one opaque bulk action.
One goal, every sub-account, reviewed per account
Connect at the Google Ads manager-account level and a single goal — "flag any ad group over $200 spend with zero conversions" — runs the same logic across every selected sub-account, returning a consolidated per-account breakdown rather than one blended number that hides which clients actually have the problem.
Crucially, a goal fanning out to fifty accounts produces fifty individually reviewable proposals, each with its own dry-run. That's the difference between scaling the work and scaling the blast radius — see the full MCC fan-out walkthrough.
Client isolation is structural, not a convention
| What's isolated per organization | Why it matters for an agency |
|---|---|
| Connected accounts & integrations | One client's credentials are never in scope for another client's task |
| Account memory (targets, protected campaigns, rules) | Client A's "never pause brand" rule can't leak into Client B's run |
| Skills library | A method built for one client stays scoped to them unless you deliberately share it |
| Audit log | A per-client record of what was proposed, approved and applied |
| Billing | Spend attributable per organization |
Roles that match how an agency actually works
| Role | Can do |
|---|---|
| Admin | Everything — members, integrations, approvals, billing |
| Editor | Authorize integrations, create tasks, approve mutations — no member or billing control |
| Member | Create tasks and view results — cannot authorize integrations or approve writes |
Standardize your method across the roster
The audit you run for your best client should be the audit every client gets. Save it as a Skill and the agent follows that method on every account instead of re-deriving one — which is how process quality stays consistent as the roster grows past what one person can personally review. Skills, memory & teams covers the mechanics.
Pricing at agency volume
Credits pool per organization. The Agency plan is $1,000/month for 1,200,000 credits; Scale is $500/month for 550,000. Because credits map to real model and tool cost, a heavier month costs more and a lighter one costs less rather than paying per seat or per account — the pricing page has the full ladder.
Frequently asked questions
- Can an AI agent manage multiple client Google Ads accounts?
- Yes — connecting at the manager-account (MCC) level lets one plain-language goal fan out across every selected sub-account, with a consolidated per-account breakdown and a dry-run preview before any write applies.
- How do you stop one mistake from hitting every client at once?
- A goal fanning out to fifty accounts produces fifty individually reviewable proposals, each with its own dry-run and its own approval — not one bulk action. Each sub-account's state is also re-validated immediately before its write applies.
- Are client accounts kept separate?
- Yes — each organization is an isolated workspace with its own connected accounts, integrations, account memory, skills library, audit log and billing. Nothing crosses between them.
- Can I stop junior staff from approving changes to live client accounts?
- Yes — the Member role can create tasks and view results but cannot authorize integrations or approve writes. Editors can approve; Admins additionally control members and billing.
- What does it cost for an agency?
- The Agency plan is $1,000/month for 1,200,000 credits, and Scale is $500/month for 550,000. Credits pool across the organization rather than being charged per seat or per client account.