Who this is for. All four humans read this end-to-end once so the model is shared. Corrina reads it as her working playbook (she runs the weekly rhythm and moves the review pack onto WhatsApp). Laurence, Lisa, and Carla read section 3 (the shared gate), section 4 (the domain lines they sign), section 5 (the quorum rules), and section 8 (red flags & crisis) — those are the sections needed to sign a domain line confidently.
Parent hub. /ai-marketing-agency (agency org chart, 6 charter pages, decision-rights matrix, rhythms). This manual sits above the charters and describes how the humans run them.
Companion pages. /marketing-review-checklist (the 7-checkbox card the AI CMO runs automatically) and /marketing-escalation-map (the domain-line assignment table — which artefact categories trigger which sign-off lines).
Contents
- 1. Governance note — external contractors and founder accountability
- 2. Who's who (4 humans, 2 external, 2 co-founders)
- 3. The shared gate — AI CMO pre-check → review pack → WhatsApp group → named sign-offs
- 4. The four named domain sign-off lines
- 5. Quorum rules — how many sign-offs decide the ship
- 6. The weekly rhythm (Mon · Tue · Wed · Thu · Fri)
- 7. Session-launch prompts — exactly what to paste, in what order
- 8. Red flags — stop, don't ship, ask everyone
- 9. Crisis protocol (journalist call · complaint · beneficiary incident)
- 10. Hours budget — the 4-hour/week math, honestly
- 11. Known failure modes & guards
- 12. Change control
1. Governance note — external contractors and founder accountability
Two of the four humans running LWD's marketing sit outside LWD: Corrina (Your Digital Team, external headless marketing agency) and Carla (COSAI CFO Services, external PMO director then ongoing CFO). Two are LWD co-founders: Laurence (Co-founder & CEO) and Lisa (Co-founder & Head of Philanthropy). This arrangement is efficient for a lean charity but it carries a governance rule that everyone should be explicit about.
The rule, in one sentence. An external contractor's sign-off on a domain line is a domain opinion — expert, contracted, and material — but never a founder-accountability decision. Under FIA Code and ACL 387398 conduct rules, the accountable persons are the co-founders. Corrina and Carla sign their domain lines with full weight in normal operation; on the constitutional-hard-case items (real-beneficiary anything, hard-constraint page, material narrative shift) the co-founders' sign-offs are the decisive ones and the external contractors' sign-offs are advisory to that decision.
Everyone signs their own line. Nobody rubber-stamps someone else's line. The AI CMO does not sign any line — it produces the review pack, runs the 7-checkbox card, and reports. Humans decide.
This isn't bureaucracy. It's what makes the arrangement safe for everyone in it — the external contractors are protected from carrying founder-level accountability they weren't hired to carry, and the co-founders retain the authority regulators expect them to exercise.
2. Who's who
Laurence
Co-founder & CEO of LWD. Signs the Evidence & Brand line: evidence integrity (Y1-number use), spokesperson quotes, brand voice, operating narrative, anything touching cosai-proposal.html, journalist-response substance.
Lisa
Co-founder & Head of Philanthropy at LWD. Signs the Philanthropic line: donor, funder, trust/foundation, and corporate-partner messaging; all philanthropic-network communication; funder-facing landing pages.
LWD co-founderCarla
PMO Director then ongoing CFO (COSAI CFO Services, external). Signs the Financial & Regulatory line: $ figures and cost ratios, ACL / ABN / ACN / DGR / PBI / ACNC copy accuracy, tech-that-runs-LWD integrity, hard-constraint infrastructure, Google Ads account admin.
COSAI · externalCorrina
CMO (Your Digital Team, external headless marketing agency). Signs the Marketing-Craft line: voice, calendar-fit, channel-fit, ad-copy quality. Also runs the weekly rhythm and acts as courier — moves the AI CMO's review packs into the shared WhatsApp group.
Your Digital Team · external3. The shared gate
Every artefact the AI agency produces (ad copy, landing page, press release, LinkedIn post, email, spokesperson response) flows through one shared review gate. Four humans see everything. Each signs the line they own. The AI CMO does not vote — it produces the review pack and reports the 7-checkbox card result.
Step 1 — AI produces. One of the 5 specialist agents drafts the artefact against its Sprint 53 charter (Y1 numbers verbatim, 4 compliance markers present, dignity floor honoured, voice consistent).
Step 2 — AI CMO pre-checks. The AI CMO runs the 7-checkbox card against the artefact automatically. Result recorded on the review pack: 7/7 clean, or the specific box(es) that failed with a one-line fix note.
Step 3 — AI CMO produces the review pack. A structured 6-item pack: (i) the artefact itself, (ii) the 7-checkbox card result, (iii) which domain lines are triggered per the domain-line assignment table, (iv) the AI CMO's recommendation, (v) reversibility note, (vi) deadline. The AI CMO does not judge whether to ship — it reports.
Step 4 — Corrina moves it onto WhatsApp. Corrina copy-pastes short review packs (short artefact, card result inline) into the shared LWD marketing WhatsApp group. For long artefacts (a landing page, press release, or proposal), she drops a link to the shared doc plus a one-line summary + card result. Every review pack lives in one WhatsApp thread. All four humans are in the group.
Step 5 — Humans sign the lines that apply. Each human whose domain line is triggered replies in the thread with a signed line: ✓ Evidence & Brand — Laurence, ✓ Philanthropic — Lisa, ✓ Financial & Regulatory — Carla, or ✓ Marketing-Craft — Corrina. If any human wants a change before signing, they reply with a specific revision request. Corrina rebriefs the AI CMO; new draft; back to step 1.
Step 6 — Publish once quorum is met. Quorum is defined per artefact category (section 5). When the required lines are signed, the artefact ships. Corrina records the sign-offs in the Friday CMO Summary as evidence the gate was met.
Transport — WhatsApp group, Corrina as courier. The AI CMO cannot post to WhatsApp directly. Corrina copy-pastes short review packs; drops links for long ones. Her role in the transport is a courier, not a router — she does not filter, decide, or delay any review pack. Every pack the AI CMO produces this week reaches the WhatsApp group this week. If Corrina is on leave, Laurence or Lisa runs the courier step; the manual is not a Corrina-dependency.
4. The four named domain sign-off lines
Under Option B, the four sign-off lines are the substance of the gate. Each is named, each has an owner, each signs their own line only. Nobody signs a line they don't own. The AI CMO does not sign any line.
Evidence & Brand line — Laurence. Covers: Y1 audited numbers, spokesperson quotes, operating narrative, brand voice, hard-constraint items (cosai-proposal.html), substantive journalist response.
Philanthropic line — Lisa. Covers: donor / funder / trust / foundation / corporate-partner messaging, funder-facing landing pages, grant/trust submission copy.
Financial & Regulatory line — Carla. Covers: $ figures and cost ratios, ACL/ABN/ACN/DGR/PBI/ACNC copy accuracy, tech-that-runs-LWD integrity, hard-constraint infrastructure verification, Google Ads account admin, paid-media budget above thresholds.
Marketing-Craft line — Corrina. Covers: voice fit (Box 07 of the checklist as read by a human), calendar-fit, channel-fit, ad-copy quality, execution-level judgement.
Domain-line assignment for each artefact category is spelled out in the Domain-Line Assignment Table. The table is category-based, not judgement-based — if a row says a line is triggered, the owner signs; nobody negotiates whether a line was really needed.
5. Quorum rules — how many sign-offs decide the ship
Quorum is the number of sign-off lines an artefact needs before it publishes. Four levels, based on how many domain lines the artefact touches (per the assignment table).
cosai-proposal.html; any material shift in the LWD operating narrative that departs from the constitutional docs. All four sign, regardless of whether their domain line is nominally triggered.
Why second-eye on single-domain items. Diffusion of responsibility is the failure mode a shared gate has to guard against. "Everyone reviews = nobody reviews carefully" is a real risk. The second-eye rule means at least two humans have read every non-routine artefact — not to override the domain owner's judgement, but to ensure someone else has actually looked. The second-eye human signs a brief acknowledgement (✓ Second-eye — [name]), not a domain line they don't own.
Who counts as second-eye? Any of the other three humans. In practice: Corrina second-eyes the founders' domain-touching items (evidence, philanthropic, financial claims that don't touch marketing-craft); a founder second-eyes Corrina's marketing-craft items when they touch the founder audience. The AI CMO's checkbox result is NOT a second-eye — it's the pre-check that runs before humans see anything.
6. The weekly rhythm
Five days, ~2 hours of Corrina's time, ~30 min each for Laurence / Lisa / Carla, roughly 3.5 hours across 4 humans total. The AI CMO carries the pre-check and pack-production load, which is the reason the human hours stay small.
Monday 45 min · Corrina
PLAN. Corrina opens a fresh Claude/Genspark session and briefs the AI CMO (see section 7 for the exact paste order). The AI CMO returns: this week's 3 priorities, a specialist brief for each of the 5 agents that has work this week, and a list of artefacts expected to require multi-domain or constitutional quorum. Corrina reviews the plan and posts a Monday-plan message to the WhatsApp group so the other three humans know what's coming this week and can flag load conflicts early.
Output: 1 CMO weekly plan + up to 5 specialist briefs in-flight by Monday afternoon. WhatsApp Monday message posted so the founders know the week's shape.
Tuesday 30 min · Corrina + async on WhatsApp
REVIEW packs land. Specialists have produced first drafts. The AI CMO runs the 7-checkbox pre-check on each and produces a review pack. Corrina posts each pack to the WhatsApp group (copy-paste short, link-drop long). Humans read and sign the lines that apply to their domain, in their own time through Tuesday and Wednesday.
Output: All Tuesday-morning artefacts have a review pack in the WhatsApp group by Tuesday afternoon. Sign-off replies start rolling in Tue-Wed.
Wednesday 30 min · Corrina · humans finish signing
UNBLOCK. Some artefacts will be waiting on one more sign-off. Corrina scans the WhatsApp threads, chases any human whose line is still open, and rebriefs the AI CMO on anything a human sent back with REVISE. If quorum is met on any artefact, it publishes. If a specialist is stuck (missing context, ambiguous brief), Corrina rebriefs.
Output: Wednesday unblock note in the WhatsApp group documenting what shipped, what's held, and any sign-off still open.
Thursday 30 min · Corrina · quorum closes out
FINISH. Anything that came back with REVISE gets its revised version posted to the WhatsApp group. Humans re-sign (or not). By close of Thursday, quorum has closed on ~90% of the week's planned artefacts and they've shipped. Anything still open goes into Friday's summary as in-flight or is withdrawn.
Output: Thursday publishes the week's approved artefacts. Sign-off record kept in the WhatsApp thread — that's the audit trail.
Friday 45 min · Corrina + ≤5 min each human
CLOSE. Corrina asks the AI CMO to consolidate: what shipped this week (with the quorum reached and sign-offs recorded), what got held and why, what the analytics say, what next Monday needs to know. That becomes the Friday CMO Summary. It goes to the WhatsApp group by close of business Friday. Each human takes ≤5 min to read it and acknowledge (a single 👍 on WhatsApp is fine).
Output: Friday CMO Summary in the WhatsApp group. Sections: Shipped this week (with sign-offs) · In-flight · Sign-offs still open · Numbers · Next-week priorities.
7. Session-launch prompts — what to paste, in what order
Every AI session Corrina opens starts with the relevant charter's 4 paste-blocks. Then a task-specific prompt on top. Below are the two Corrina uses most often. Both are pasted into a fresh session — never into a session that's been running for other work, because context drift makes the AI CMO forget the LWD constitution.
Monday plan — the AI CMO briefing
Step 1. Open Claude or Genspark. Fresh session.
Step 2. Paste the four blocks from /ai-cmo-charter (Block 1 Identity, Block 2 Context, Block 3 CMO Knowledge, Block 4 Triggers & Outputs). Paste as-is. Do not paraphrase.
Step 3. Paste the following Monday-plan prompt:
Review pack — how the AI CMO produces one per artefact
After each specialist produces an artefact, Corrina asks the AI CMO to run the pre-check and produce a review pack. The prompt is short because Blocks 1–4 already carry the standing instructions.
Friday close-out
Step 1. Same session Corrina used for Monday (if it's still live) or a fresh session with the 4 CMO blocks re-pasted.
Step 2. Paste the following:
8. Red flags — stop, don't ship, ask everyone
These are the reflexes everyone in the marketing operation shares. If any of these appear, the current draft does not ship until it is fixed. No exceptions. No "close enough". A red flag on any human's read triggers a REVISE reply on WhatsApp; the AI CMO regenerates; the review pack cycles again.
Stop-signals (any one of these = do not publish):
- A Y1 number that isn't in the canonical set. Only these: 31 households · $1.82M · 23-day average · 52% K10 reduction · $1,840 average cost per case · 87c per dollar to direct service. Never invent, never round, never extrapolate.
- "Debt wipe" or "debt relief" language. LWD does mediation and settlement. It does not wipe debt. ASIC RG 96 prohibits "debt relief" phrasing anyway.
- Tax-deductible / DGR / PBI claim while DGR status is pending. Even one word.
- Urgency-appeal language. "Save a life today", "One dollar could change everything", any manufactured urgency. LWD's voice is qualified, quiet, evidence-led.
- Any real LWD beneficiary named, quoted, or shown in imagery — without written consent AND all-four-human sign-off (constitutional hard case). Stock imagery with faces is fine and does NOT trigger this rule.
- Hero-framing of carers, clinicians, or LWD staff. Not "these brave clinicians"; not "our founders' incredible journey".
- Any change to
cosai-proposal.html— the hard constraint page. Constitutional hard case: all 4 humans sign. - Missing any of the 4 compliance markers (ABN, ACN, ACL 387398, mediation disclosure) or missing "complaints" or "privacy" links.
- An outcome claim without the caveat. "Y1 outcomes describe past pilot results for 31 households and are not a guarantee of individual future outcomes." Attached whenever Y1 numbers appear in donor-facing copy.
9. Crisis protocol
Three categories of event that break the normal weekly rhythm. Each has its own reflex. In crisis the WhatsApp group remains the shared surface, but the shape of the reply changes — sign-offs become quick verbal "yes / no / hold" from the founders while Corrina holds the operational line.
Journalist call
Immediate (within 30 min): Corrina acknowledges receipt to the journalist with a holding line: "Thank you — noting your query. Someone from LWD will be in touch within X hours." Do not answer the substantive question in the holding line. Do not commit to a timeline shorter than 4 business hours unless the journalist's deadline is tighter and you have already reached Laurence.
Within 4 hours: Corrina posts to the WhatsApp group with the journalist's query, the holding line sent, and requests the Evidence & Brand line signature on the substantive response draft. Laurence signs the substance (Evidence & Brand line); Carla signs Financial & Regulatory if the query touches numbers or registrations. Response goes to the journalist from Laurence or on Laurence's authority.
Never: answer a journalist off-the-cuff. Never let a specialist agent respond to a journalist — only Corrina holds the operational line and only Laurence decides the substance.
Complaint received
Same day: acknowledgement logged (Corrina). FIA Code requires same-day acknowledgement of marketing complaints.
Within 5 business days: substantive response drafted (AI CMO on Corrina's brief), review pack posted to WhatsApp, Laurence signs Evidence & Brand (all complaints), Lisa signs Philanthropic if donor-related, Carla signs Financial & Regulatory if regulatory. Sent under Laurence's authority.
Always logged: to /complaints per FIA Code discipline.
Beneficiary incident
Immediately: Corrina stops any live marketing activity that could touch the incident. All four humans are notified in the WhatsApp group within the hour. Marketing does not respond publicly until Laurence has determined the appropriate response with clinical/operational leads. This is a constitutional-hard case; all four humans sign; marketing follows, marketing does not lead.
10. Hours budget — the 4-hour/week math, honestly
The whole point of this design is that 4 humans can run a professional marketing operation in ~4 hours a week total. Here is the actual math under Option B:
| Person | Sign-off line | Weekly time | What they spend it on |
|---|---|---|---|
| Corrina | Marketing-Craft (external) | ~2 hrs | Mon plan 45m · Tue courier + review 30m · Wed unblock 30m · Thu finish 30m · Fri close 45m. Sign the Marketing-Craft line on ~10-12 artefacts/week; second-eye 3-5 more. |
| Laurence | Evidence & Brand (co-founder) | ~30 min | 2-4 sign-offs/week (spokesperson quotes, Y1-number use, brand-voice calls, hard-constraint items). Consulted on crisis. Reads all Friday summaries. |
| Lisa | Philanthropic (co-founder) | ~20 min | 1-3 sign-offs/week pre-DGR (funder-page copy, philanthropic-network messaging). Load increases significantly post-DGR. Reads all Friday summaries. |
| Carla | Financial & Regulatory (external) | ~30 min | 2-4 sign-offs/week (financial figures, registration/regulatory copy, tech-integrity claims, Google Ads account admin). Reads all Friday summaries. |
| Total across 4 humans | ~3.5 hrs | Inside the 4-hour budget with a small buffer for exceptional weeks. Reading the WhatsApp thread is included in each human's line total — the shared gate saves time because no human reconstructs context that another human already has. | |
When the math breaks down. If Corrina finds herself couriering more than ~15 packs in one week, something is wrong — either the AI agency is drifting (rebrief needed) or the calendar was overloaded on Monday. If any human's sign-off load runs over 45 min in a week, that is a signal too — the assignment table may be too broad. The four-hour budget is a design constraint, not an aspiration.
When the math expands (planned). Post-DGR, Lisa's load will grow (donor pipeline live). When a paid-media budget starts flowing, Carla's admin load grows. Both are expected and both stay inside a 6-hour total budget in normal steady-state operations. Above 6 hours/week total, LWD needs an internal hire — not more contractors.
11. Known failure modes & guards
| Failure mode | Guard |
|---|---|
| Diffusion of responsibility — "everyone reviews, nobody reviews carefully" | Named domain sign-off lines. Each human signs their own line by name. The second-eye rule on single-domain items means at least two humans confirm they've read every non-routine artefact. Un-signed lines are visible in the WhatsApp thread — there is no "silent pass". |
| A human signs a line they don't own ("I'll approve this for Carla, she's busy") | Not allowed. Each line is signed only by its named owner. If the owner is unavailable, the artefact is held until they sign or Corrina rebriefs to remove the domain-line trigger. |
| The AI CMO's pre-check misses a compliance marker and humans trust the check instead of reading | The 7-checkbox card is a floor, not a ceiling. Humans read the artefact itself, not just the card result. If a human catches something the pre-check missed, that's a signal — add to the card's calibration in section 12 (change control). |
| The AI CMO drifts context across a session — forgets Y1 numbers, misuses boilerplates | Every new AI CMO session starts fresh with the 4 paste-blocks. Nothing "remembered" — only what's pasted. Session-per-week discipline (Monday session used all week; fresh session next Monday). |
| A review pack sits in the WhatsApp thread with no sign-offs for >48 hours (quorum stalling) | Corrina @-mentions the un-signed line owner at 48h. At 72h with no response the artefact is held (default) or withdrawn (Corrina's call). Sitting in a queue indefinitely is worse than deciding not to ship. |
| Corrina publishes on Friday afternoon and no other human is available for a same-day fix if something breaks | Fridays after 3pm AEST are read-only for new publishes. Anything time-critical publishes Mon-Thu. Weekend publishes require pre-Friday quorum sign-offs and same-day-Monday retro. |
| Weekly plan expands to 6+ priorities | Monday-plan prompt enforces max 3. If more than 3 things feel urgent, one of them isn't. Force the choice. |
| An external event (DGR granted, journalist story lands) breaks the weekly rhythm | Documented in section 9 (crisis protocol) and in each specialist charter's Block 4. Trigger-response is prescribed — not improvised. |
| The AI CMO starts sounding like it's voting ("I recommend BLOCK on this") | Rebrief. The AI CMO's job is to run the pre-check and produce the report. Recommendations on quorum level are welcome. Verdicts on ship/hold are for humans. Reset the session if the pattern persists. |
12. Change control
This manual is reviewed monthly by all 4 humans (30 min in the WhatsApp group, one message each). Changes are logged with reasons in git. Two events trigger an out-of-cycle review:
- DGR granted → whole manual is re-read; Lisa's Philanthropic-line load steps up; funder-facing landing pages activate; new domain-line triggers apply.
- A red-flag event that made it past quorum to publish → the failure is analysed, the failure-mode row is added to section 11, the manual is patched, the fix is retro-audited across the last 4 weeks of shipped artefacts to check whether it slipped elsewhere.
Manual changes never happen silently. Corrina notes them in her Friday summary the week they land, and highlights them in the WhatsApp group. The manual is versioned by Sprint number.