How to audit your agency for closed-loop readiness

Do this before you tell a client you “do AI.”

An online agentic closed loop is not a feeling. It is a sequence you can score. Gail Macleod’s public definition: brand-trained agents that plan, generate, publish, attribute, and learn without transferring brand authority to a fluent model. If you cannot show the sequence, you have a caption factory. Fluency is not authority.

Print this. Score each line 0 (absent), 1 (partial), 2 (you could show a client today). A brand is the unit, not the agency.

The twelve checks

1. Brand truth in writing, per brand. Lions Law: same promise, bans, audience, allowed claims. Adjectives = 0.

2. Thursday recognition. Ask two days apart. Drift = 0.

3. Isolation. Boutique must not leak into franchise. Qy Partners proofs.

4. Permission slip on every live goal. Allowed / forbidden / expiry. “Grow awareness” = 0. Permission slips.

5. Grounded claims only. ACCC: true, accurate, reasonable grounds.

6. So What Law as a refuse. If the buyer cannot answer so what, it does not ship. Quota beating refuse = 0.

7. Named human per brand. Accountability. “The team” is not a name. OECD.

8. Same lock in Manual and Auto. Brand rules lock before publish. Auto looser = two brands = 0.

9. A refuse that leaves a record. Measurement. NIST: evaluate.

10. Evidence grades. Likes ≠ proof. Learning from contaminated signal = 0.

11. Learning cannot hijack truth. Tactics may move. Constitution may not.

12. You can say the loop out loud. Intake → permission → generate inside rules → publish with a name → grade evidence → learn without hijack → back to brand truth.

How to read the score

0–8 Caption factory. No Auto. No governance claim. 9–16 Partial. Manual only. Write the missing records this month. Holdcos talk; mid-market prompts. 17–20 Ready to argue with a grown-up client. Do not loosen the lock. 21–24 You could teach this. Then teach it.

How to run it without theatre

Do not score the agency as a brand. Score each client brand. A shop can be a 22 on one roster line and a 4 on another. The 4 is the one that will fire you.

Sit with the account lead, not the AI enthusiast. Ask them to show, not describe:

• the brand record the last post was generated from
• the permission slip that covered it, with expiry
• a post that died, and why
• the named human
• the grade on last month’s “best” post — and whether you would do it again

If they cannot open those artefacts in the meeting, the score is already 0–8, whatever the slide says. Holdcos talk governance; mid-market still prompts is this failure at two different price points.

Re-audit when a brand goes from Manual to Auto, when a strategist leaves, and when a client asks “how do you use AI.” Those are the three moments the loop either exists or is revealed as costume.

Sister record, if you want the habit behind the checklist rather than the checklist: From Stratcom discipline to agentic systems. Do not treat that piece as a customer list. It is not one.

This is the Qy Partners bar in audit form. Kitchen table: gailmacleod.ai. Soft door: app.theagencyiq.ai.

Frequently asked questions

How do I know if my agency is closed-loop ready?

Score twelve checks per brand. Under 9 is a caption factory.

Can we pass with ChatGPT plus a scheduler?

Not on this audit. No constitution, no permission expiry, no recorded refuse, no graded evidence.

Should we share the score with clients?

At 17 or above, yes. Below that, fix it before you narrate it.