Blended Workforce Design
The specification for how Life Without Debt Ltd operates its marketing and business-development function as a blended workforce — two AI agents (a CMO Agent and a Digital Marketing Agent) working under named human supervision, feeding a room-segmented donor CRM that turns awareness into won donors across the seven funding rooms.
Correction · Standing Order #5 · Marketing approval authority
This v0.2 draft (Session 8L) was designed around a mistaken assumption — that Carla Oliver (Board-Appointed CFO Advisor) was also the day-to-day approver for LWD marketing content. The user (Laurence Hugo, CFO / co-founder) issued Standing Order #5 at Sprint 28.13:
"the marketing has to be approved by laurence and lisa hugo and the human CMO corrina mcgowan from yourdigitalteam (YDT), NOT carla oliver, i am the cfo not the CMO."
The corrected marketing approval authority for LWD is:
- Laurence Hugo · Co-founder · ACL 387398 holder · RG 96 compliance reviewer · beneficiary-voice authority.
- Lisa Hugo · Co-founder & Head of Philanthropy · donor-frame authority.
- Corrina McGowan · Human CMO · YourDigitalTeam (YDT) · marketing craft / tone / channel-fit authority.
Carla Oliver remains the CFO (project phase, rolling to BAU CFO on project completion) and reviews the CFO frame only — financial claims, salary bands, tax-time messaging where a financial claim is being made. Carla is not a marketing approver. The v0.3 body refresh was completed in Sprint 29.4: §3 CMO-Agent and Digital-Marketing-Agent human-approver blocks, §4 TOFU owner attribution, §6 humans-table schema and reference-table intro, §8 LWD-placement map, and §8.5 HA4 entry on the Human Agency Scale ladder have all been rewritten to the triad. This correction banner remains controlling; if any historic Carla-as-marketing-approver phrasing surfaces elsewhere in the site tree, it is superseded by Standing Order #5.
Standing Order #6 (Sprint 32, 2026-08-06) extends this. Four operating assumptions banked from user directive: (A) Laurence Hugo + Lisa Hugo are the Business Development / donor-relations team (marketing approval authority under SO #5 retained). (B) Corrina McGowan is commercially engaged (paid, not volunteer/advisor) and now sits at HA4 and HA5 alongside Laurence + Lisa — the triad sits WITH the Board at terminal approval, not just at HA3 sight-check. (C) Carla Oliver (CFO) has set the Anthropic budget cap at $10,000 AUD TOTAL project lifetime (not monthly), with a halt-at-80% threshold of $8,000 AUD cumulative. On project completion (both charity registration set up and the HA1–HA5 ladder has released the CMO + Digital Marketing Agency content already built within project scope) Carla rolls off to BAU CFO. (D) The Board Meeting of Tuesday 25 August 2026 (Australia/Sydney) is the hard deadline for three decisions: Kit-04 HA4 vote, DRP v1.0 adoption, launch-scope resolution. See /board-pack-2026-08-25 and /sprint32 for the live status pages.
1. The problem this design solves
LWD does not have the Year 1 budget to hire a full marketing team. It also has ambitions that would ordinarily require one: a seven-room funding portfolio (Rooms A–G), a channel presence across Instagram, Facebook, LinkedIn, X, Google, YouTube, Spotify and Reddit, a landing-page-per-room search strategy (AEO/SEO/GEO), and a business-development function that has to turn awareness into won donors month after month.
The awareness layer — the top of the marketing funnel — has largely been built. What LWD does not yet have is the sales funnel: the consideration and conversion layer that captures awareness-layer traffic, nurtures it stage by stage, and converts it into signed gifts, grants, partnerships and regular giving. Nor does LWD have the production engine that keeps the awareness layer fed with fresh, on-brand, room-appropriate content across all eight channels every week.
This design specifies a blended workforce — two AI agents supervised by named humans — that closes both gaps at a fraction of the cost of hiring and manages the CRM that Lisa Hugo (Co-founder & Head of Philanthropy) uses day-to-day.
2. Design principles
Every element of this design is subordinate to these principles. They exist to keep the blended workforce lawful, on-brand, and accountable to the Board.
P1. Agents draft. Humans decide.
Nothing produced by an agent goes to an external channel, donor or regulator without a named human approver logged against it. Non-negotiable.
P2. Every output traces to House principles.
Every artefact must be traceable to (a) the Seven Founding Principles in value-proposition.html, (b) a specific room A–G, and (c) the Transparency Pledge.
P3. Different rooms, different languages.
Principle #3 of the House is the design law. The CMO Agent enforces it in briefs. The Digital Marketing Agent enforces it in copy. The CRM enforces it in nurture segmentation.
P4. Data stays inside LWD's own stack.
CRM on Cloudflare D1. No third-party vendor for prospect data. Agent audit logs held on-stack. Consistent with ACNC Governance Standard 3 (compliance with Australian laws) and the Privacy Policy.
P5. Auditable to the Board.
Every agent brief, every output, every human approval, timestamped and searchable. If the Board asks "who approved that LinkedIn post?", there is an answer.
P6. Pre-DGR-safe language.
Until DGR endorsement lands, no artefact makes a "tax-deductible" claim to any donor. The Compliance Guardrails One-Pager (Sprint 0.6) is the reference — the agents' system prompts embed those guardrails.
3. The two agents
The blended workforce has two agent roles. Both agents are, at model level, Anthropic Claude Sonnet 4.5 instances, differentiated by system prompt, tool access and human supervisor. They do not operate autonomously — they operate under briefs and hand outputs to human approvers.
CMO Agent
- Role in one line
- Sets the marketing mix strategy across Rooms A–G, campaign by campaign, and briefs the Digital Marketing Agent to produce it.
- Reports to
- Marketing approval authority: Laurence Hugo (co-founder), Lisa Hugo (Head of Philanthropy), Corrina McGowan (Human CMO · YourDigitalTeam). Board (via monthly Marketing Report). CFO-frame review (financial claims / salary bands only): Carla Oliver (CFO Advisor).
- Grounded in
- value-proposition.html (Seven Principles + Rooms A–G) · theory-of-change.html · growth-model.html (activation sequence) · media-campaign.html (calendar & targets) · funding-applications.html (existing pitches) · media-kit.html (voice & terminology).
- Typical brief in
- “Design the Y1 EOFY campaign for Room C.” · “What’s the corporate-partnership sequence for CBA?” · “Sequence Room A foundation approaches for next quarter.” · “Draft the monthly marketing report to the Board.”
- Typical output out
- A strategy brief: objective, room, target audience, key message per Principle #3, channel mix, budget shape, success metrics, editorial guardrails, and a work order for the Digital Marketing Agent naming the artefacts required.
- Human approver
- Marketing approval authority (SO #5 · extended under SO #6): Corrina McGowan (Human CMO · YourDigitalTeam · commercially engaged · default), Laurence Hugo (co-founder · BD-team lead · beneficiary voice), Lisa Hugo (Co-founder & Head of Philanthropy · BD-team lead). Any single approver can sign off at HA3 for tactical work; the triad co-approves at HA4 and HA5 for strategic decisions — the triad sits WITH the Board at terminal approval (SO #6). Enforced at the endpoint layer by
POST /api/agent-outputs/:id/approve. Carla Oliver reviews CFO-frame only (financial claims / salary bands / remuneration disclosures) and is not a marketing approver; Carla holds the $10,000 AUD Anthropic project-lifetime cap under SO #6. - Escalation triggers
- Any legal / compliance question → Deen. · Any beneficiary-depiction question → Laurence. · Any financial claim → Carla (CFO frame). · Any budget above ceiling → Board.
- Never does
- Never publishes directly. · Never approves its own output. · Never makes a “tax-deductible” claim until DGR issued. · Never invents a case study or statistic — only cites those already in the codebase / evidenced.
Digital Marketing Agent
- Role in one line
- Takes work orders from the CMO Agent and produces channel-ready content — ads, posts, emails, landing-page copy, briefing packs — optimised for AEO / SEO / GEO where relevant.
- Reports to
- CMO Agent (for briefs) · Lisa Hugo (for BD-linked content — prospect nurture, proposal drafts, meeting briefs).
- Grounded in
- Everything the CMO Agent is grounded in, plus: paid-ads.html (creative library) · case-studies-bank.html (narrative proof) · referrals.html (Room F patterns) · how-lwd-can-help.html (beneficiary voice) · media-kit.html §4 (approved terminology and banned phrases).
- Channels produced for
- Instagram, Facebook, LinkedIn, X, Google (Search + Display + YouTube), YouTube long-form, Spotify (podcast pre-rolls / mid-rolls), Reddit, and email (donor lifecycle sequences).
- Typical brief in
- Work order from the CMO Agent, or a direct human ask: “Write 5 LinkedIn posts for Room B this month.” · “Draft the EOFY email sequence for Room C.” · “Write the Room A landing-page copy with AEO structured data.”
- Typical output out
- Channel-native drafts with variants where the platform benefits from A/B testing (e.g. 3 hook variants per Meta ad). Each output includes: the room it targets, which Principle it draws on, which case-study / statistic it cites (with source ID), and a self-check against the pre-DGR compliance guardrails.
- Human approver
- Marketing approval authority (SO #5 · extended under SO #6): Lisa Hugo (BD-team · donor-relations content), Corrina McGowan (Human CMO · YDT · commercially engaged · public-facing marketing), Laurence Hugo (BD-team · any output that speaks in his voice or invokes his clinical background). Any one of the triad can approve public-facing marketing at HA3; the triad sits WITH the Board at HA4/HA5 for terminal approval (SO #6). Carla Oliver reviews CFO-frame only (any financial claim in the copy) and is not otherwise a marketing approver.
- Escalation triggers
- Any beneficiary depiction bordering the Sprint 0.4 guardrails → Laurence. · Any legal claim / statistic not in the citations index → Deen. · Any financial claim in the copy → Carla (CFO frame). · Any output where the target platform's ad policy is ambiguous → Corrina (Human CMO · YDT).
- Never does
- Never publishes directly. · Never invents statistics or case studies. · Never uses images / voices / likenesses of real beneficiaries (only composites already in the case-studies bank). · Never claims “tax-deductible” pre-DGR.
The two agents are deliberately separated. The CMO Agent decides what and why; the Digital Marketing Agent decides how. This separation is what allows humans to intervene meaningfully — the marketing approval authority (Laurence, Lisa, and Corrina McGowan · Human CMO from YourDigitalTeam) can approve the strategy (the CMO Agent's brief) without having to read every LinkedIn post, and any one of them can approve a single post without having to re-litigate the strategy.
4. The funnel — where the agents and the CRM plug in
The marketing funnel is drawn end-to-end below. Every stage names its owner (which agent or human), the nurture touchpoints that fire at that stage, and where the CRM captures the record.
What this is: anonymous traffic on the public website (Door 1), a channel post scrolled past, a paid ad seen, a piece of earned media consumed. No opt-in. Aware of LWD, not yet identified.
Nurture touchpoints: paid-ads library (channel-native), earned PR, blog posts, room-specific SEO landing pages (once built), Google Ads Search + Display, Meta / LinkedIn / X organic, YouTube / Spotify placements, Reddit community engagement, referral-partner mentions.
Owner: Digital Marketing Agent (production) under CMO Agent (strategy). Approval authority (SO #5): any of Laurence Hugo / Lisa Hugo / Corrina McGowan (Human CMO · YDT). Carla Oliver reviews CFO-frame only.
CRM captures: aggregate metrics only (channel, impressions, click-through, cost). No PII yet.
What this is: the suspect has identified themselves. Newsletter signup, donation form filled but abandoned, resource downloaded (bequest guide, foundation prospectus, referral-partner pack), room-page enquiry submitted, event RSVP.
Room segmentation: the moment identity is captured, the CRM stamps the record with a room segment (A–G) based on the source page / campaign / referral partner. Every downstream nurture touchpoint is room-scoped from here forward.
Nurture touchpoints: room-appropriate 5–7 email welcome sequence (foundation officer vs 45–70 caregiver vs CSR officer — three completely different sequences), monthly newsletter (segment-branched), social re-engagement (custom audiences via ad platforms), retargeting.
Owner: Digital Marketing Agent (produces nurture content) → Lisa (BD-facing prospects escalate to her; regular-giving prospects run automated).
CRM captures: full prospect record. Name, contact, room, source, engagement history, giving capacity band (see §5 on how this is set).
What this is: the prospect shows signals of giving capacity + genuine interest. For Room A/B/E: the organisation or individual matches the capacity screen and has engaged twice or more. For Room C: repeat engagement or a first micro-donation. For Room D: a policy officer has attended a briefing. For Room F: a referring clinician has completed the referral form.
Nurture touchpoints: personalised cultivation email from Lisa (or Laurence for major donors / high-profile foundations), briefing pack tailored to the room, invitation to an intro call, case-study match to their sector.
Owner: Lisa (owns the pipeline from this stage onward for BD-relevant rooms) · Digital Marketing Agent still produces the touchpoint drafts, Lisa approves.
CRM captures: qualification date, qualifying signal, next-scheduled touchpoint, capacity estimate.
What this is: two-way conversation started. Intro call held. Board member / donor advisor in the room. Room-specific proof shared (evaluation framework, ESG alignment, tax-deductibility briefing post-DGR, bequest structure). This is where the sale is actually made — the “ask” that follows is confirmatory.
Nurture touchpoints: bespoke case narrative (produced by the Digital Marketing Agent on brief from Lisa), meeting cadence (every 2–4 weeks), tailored impact projection using the growth-model figures, invitation to visit a case (with beneficiary consent) or attend a Board observer session.
Owner: Lisa + Laurence (for major donors / foundations / government — Rooms A, B, D, E). Never automated.
CRM captures: meeting notes, materials shared, decision-makers identified, competing priorities noted, expected decision date.
What this is: formal proposal / gift agreement / partnership tier / EOFY campaign ask made. Amount and instrument specified. Response window open (typically 30–90 days depending on room).
Nurture touchpoints: proposal follow-up sequence at 7d / 21d / 60d, MOU / gift-agreement red-lines, gift-terms clarification, referee introductions if requested.
Owner: Lisa + Board (Board is the signatory for gifts / MOUs above thresholds set in the Direct Relief Policy and Reserves Policy).
CRM captures: proposal document, amount asked, instrument, response due date, red-line status.
What this is: gift received / grant awarded / partnership signed / regular giving begun. Enters the stewardship track. The gift is not the end of the relationship — it is the start of the next one.
Nurture touchpoints: 48-hour personal thank-you (from Laurence or Chair, hand-signed for major donors) → 30-day welcome-to-community pack (case-study preview, next-cohort briefing) → quarterly impact reports (auto-generated by Digital Marketing Agent from CRM + operations data; human approved) → annual renewal / uplift ask.
Owner: Laurence + Lisa. The 48-hour thank-you is a Board-level commitment.
CRM captures: gift record (amount, instrument, dates), stewardship touchpoints scheduled and completed, next renewal date, uplift trajectory.
What this is: declined, or non-response after 3 nurture cycles. Recorded with a reason code (timing / priorities / fit / capacity / relationship-elsewhere / unclear). Returned to long-nurture at 12-month interval.
Nurture touchpoints: annual soft touch (year-end impact report), invitation to public events, sector-news share when directly relevant.
Owner: Digital Marketing Agent (automated) · Lisa reviews annually for re-activation.
CRM captures: loss reason, re-engagement date, any change in circumstances noted (new CSR officer, foundation strategy shift, personal transition).
5. CRM design — on Cloudflare D1
Data lives inside LWD's own Cloudflare stack — the same stack as the intranet. Six tables carry the full pipeline. Every table has created_at, updated_at, and (where relevant) created_by / updated_by columns for audit — suppressed below for readability.
5.1 Table prospects — the master donor record
| Column | Type | Purpose |
|---|---|---|
| id | TEXT PRIMARY KEY | ULID. |
| display_name | TEXT | Person name, foundation name, or corporate name. |
| prospect_type | TEXT | individual · foundation · corporate · government · health_partner · estate. |
| room | TEXT NOT NULL | A / B / C / D / E / F / G — the room this prospect belongs to. Non-nullable by design (Principle #3). |
| stage | TEXT NOT NULL | suspect · prospect · qualified · cultivation · ask · won · lost · steward. |
| capacity_band | TEXT | micro (<$100) · small ($100–$1k) · mid ($1k–$25k) · major ($25k–$250k) · principal (>$250k). |
| email, phone, org, role | TEXT | Contact + affiliation. |
| source | TEXT | Which channel / campaign / referral partner captured this prospect. |
| consent_flags | TEXT (JSON) | Privacy-Act-compliant consent: email marketing, phone contact, case-study use, photograph use. |
| assigned_to | TEXT | Named human FK to humans table: Laurence, Lisa, Corrina (marketing approval authority per SO #5), Carla (CFO-frame only), Deen (legal) — enforces the “named human approver” principle. |
| next_action | TEXT | The single next scheduled touchpoint. |
| next_action_due | DATE | When that touchpoint fires. |
5.2 Table touchpoints — every interaction, ever
| Column | Type | Purpose |
|---|---|---|
| id | TEXT PRIMARY KEY | ULID. |
| prospect_id | TEXT FK | Links to prospects.id. |
| touchpoint_type | TEXT | email_sent · email_opened · email_clicked · meeting_held · proposal_sent · donation_made · event_attended · social_engagement · referral_received. |
| channel | TEXT | email · phone · in_person · linkedin · meta · google · etc. |
| subject | TEXT | One-line label. |
| notes | TEXT | Free text. |
| artefact_id | TEXT FK | Which agent output was used (if any) — links to agent_outputs.id. |
| outcome | TEXT | progressed · no_response · declined · converted. |
5.3 Table gifts — the money side
| Column | Type | Purpose |
|---|---|---|
| id | TEXT PRIMARY KEY | ULID. |
| prospect_id | TEXT FK | Links to prospects.id. |
| room | TEXT | Denormalised for room-revenue reporting. |
| amount_aud | REAL | Gift amount in AUD. |
| instrument | TEXT | one_off · regular · grant · partnership · bequest_pledge · bequest_realised · in_kind. |
| received_date | DATE | When funds cleared. |
| gift_fund | INTEGER | 1 if held in the DGR Gift Fund, 0 otherwise (post-DGR endorsement). |
| restricted_purpose | TEXT | Null if unrestricted; otherwise the restricted purpose per gift agreement. |
5.4 Table agent_outputs — the audit trail
| Column | Type | Purpose |
|---|---|---|
| id | TEXT PRIMARY KEY | ULID. |
| agent | TEXT | cmo · digital_marketing. |
| brief_in | TEXT | The full brief given to the agent. |
| output | TEXT | The full agent response. |
| room | TEXT | Which room the output targets. |
| principle_refs | TEXT (JSON) | Array of Founding Principles cited by the output. |
| citations_used | TEXT (JSON) | Array of case-study IDs / statistic sources. |
| human_approver | TEXT | Named person who approved / edited / rejected. |
| approval_status | TEXT | draft · approved · edited · rejected. |
| published_channels | TEXT (JSON) | Where it went once approved. |
| llm_model, prompt_tokens, completion_tokens, cost_aud | MIXED | Cost tracking for Board reporting. |
5.5 Table nurture_templates — the per-stage-per-room library
| Column | Type | Purpose |
|---|---|---|
| id | TEXT PRIMARY KEY | e.g. C_prospect_welcome_01. |
| room | TEXT | A–G, or ALL for cross-room. |
| stage | TEXT | Which stage this template fires at. |
| channel | TEXT | email · linkedin_dm · sms · letter · meeting_agenda. |
| sequence_position | INTEGER | Position in a sequence (1, 2, 3…). |
| delay_days | INTEGER | Days after previous touchpoint before this fires. |
| subject_line | TEXT | For emails. |
| body_markdown | TEXT | The template body. Supports variables like {{first_name}}, {{room_case_link}}. |
| editable_by_agent | INTEGER | 1 if the Digital Marketing Agent may vary it per prospect; 0 if it must be sent verbatim. |
5.6 Table humans — the approvers
Reference table (5 rows as of Sprint 28.13): id=1 Carla Oliver (CFO Advisor, Board-Appointed · CFO-frame review only, not a marketing approver), id=2 Lisa Hugo (Co-founder & Head of Philanthropy, marketing approval authority), id=3 Deen Van Praag (Legal Advisor), id=4 Laurence Hugo (Co-founder & CFO, ACL 387398, RG 96, marketing approval authority · beneficiary voice), id=5 Corrina McGowan (Human CMO · YourDigitalTeam · marketing approval authority). Extensible as staff / directors are added. Every agent_outputs.human_approver, prospects.assigned_to, touchpoints.created_by is FK'd here. Standing Order #5 (Sprint 28.13) sets marketing approval authority as the triad Laurence + Lisa + Corrina; CFO-Advisor review is a separate lane.
6. Room × stage nurture matrix (Y1 focus)
This is the core mapping the Digital Marketing Agent produces content into and the CRM automates against. Only Y1-activating rooms (A and C) are populated in full below — Y2 rooms (B and F) are outlined; Y3+ rooms (D, E, G) are placeholders until their activation year.
| Prospect (welcome / warm) | Qualified (cultivation begins) | Cultivation (2-way convo) | Ask (proposal) | Won (steward) | |
|---|---|---|---|---|---|
| Room A Foundations |
Foundation-officer welcome email + LWD one-pager + link to Grant Application excerpt. Sequence: 3 emails over 14 days. |
Personal note from Laurence + tailored theory-of-change extract for their programme area. | Board-observer invitation + evaluation-framework briefing + funder-specific budget model. | Full Grant Application + budget + evaluation plan + governance pack (from funding-applications.html). | Quarterly impact report + evaluation data + invitation to programme visit + Y2 renewal 90d before due. |
| Room C General public regular giving |
Welcome + Margaret / David / Kwame case study + $19.25 explainer. Sequence: 5 emails over 21 days. |
Micro-donation ask ($10–$25 one-off) + Transparency Pledge explainer. | Regular giving invitation ($19.25/week) + gift-uplift ladder + community welcome pack preview. | Regular giving signup + tax-deductibility statement (post-DGR) + monthly-supporter badge. | Monthly impact snippet + EOFY special report + annual uplift ask (May of each year). |
| Room B Corporate CSR / ESG |
CSR-officer welcome + Bank Paradox one-pager. | Tailored ESG-alignment brief + partnership tiers. | Intro call + tailored impact projection for their customer base. | Partnership Prospectus + MOU draft + tier selection. | Quarterly co-branded impact report + employee-giving toolkit + Y2 renewal + tier-uplift ask. |
| Room F Health sector |
Clinical-audience welcome + referral pathway explainer + case outcomes. | Introduction to nearest LWD case-manager + MOU template offer. | Site visit / joint case review + service-level agreement drafting. | Signed MOU + fee-for-service pricing (where applicable, post-DGR). | Monthly referral-outcomes report + joint case studies + annual review. |
| Room D·E·G | Activation year not yet reached — templates deferred to that year’s pre-activation sprint. | ||||
7. Room-specific landing pages — AEO / SEO / GEO
Every room gets its own landing page because Principle #3 (Different Rooms, Different Languages) applies as strongly to landing-page copy as to email or ad copy. A foundation officer landing on a page written for a 45–70 caregiver will bounce, and vice versa.
7.1 CX best-practice rules the agents enforce
- Message match: the headline of the landing page repeats the promise of the ad / referral / email that sent traffic there. If the LinkedIn ad said “Fund the antidote to your bank’s own collections practices,” that phrase appears above the fold on the Room B landing page.
- Single primary CTA: one page, one conversion goal. Room A = request grant briefing. Room C = start regular giving. Room F = referral MOU. Secondary CTAs are demoted visually.
- Above-the-fold trust signals: ACNC-registered logo (post-registration), DGR badge (post-DGR), Board photos, Transparency Pledge headline.
- Room-appropriate proof: Room A gets evaluation framework + theory of change. Room C gets Margaret’s case + Transparency Pledge dollar breakdown. Room B gets ESG alignment + partnership tiers. Never mix.
- Form friction proportionate to ask: Room C donation form = 3 fields. Room A grant-briefing form = 8 fields (organisation, focus area, cycle timing, contact).
7.2 AEO / SEO / GEO structured data every landing page embeds
- Schema.org JSON-LD as
<Organization>,<NonprofitOrganization>,<WebPage>, and where relevant<FAQPage>with the room’s most-asked questions answered inline. - Answer-engine-friendly Q&A blocks so ChatGPT / Perplexity / Gemini can cite LWD when someone asks “what Australian charity helps terminally ill people with debt?”. This is the AEO layer — specific, self-contained answers with clear provenance.
- Named-entity clarity — every page names LWD, ACN (post-registration), ABN, charity status, DGR status. This is the GEO (Generative Engine Optimisation) layer — help the AI engines disambiguate.
- Traditional SEO — unique title, meta description, H1–H3 outline that matches search intent for the room, internal linking to relevant deep pages, sitemap + robots directives.
- Page-speed and Core Web Vitals — static-served from Cloudflare edge; images optimised; no client-side heavy JS unless the room's CTA requires it.
7.3 The seven landing pages
- /room-a — Foundations. Headline: the funding gap in evidence-based debt-hardship advocacy. Primary CTA: request grant briefing. Proof: theory of change + evaluation.
- /room-b — Corporate. Headline: fund the antidote to your own industry's collections practices. Primary CTA: partnership conversation. Proof: ESG alignment + Bank Paradox thesis + tiers.
- /room-c — Regular giving. Headline: one hour a week that changes everything. Primary CTA: $19.25 regular gift. Proof: Margaret's case + Transparency Pledge.
- /room-d — Government & health. Headline: financial distress as a palliative-care outcome. Primary CTA: policy briefing. Proof: health-system cost analysis.
- /room-e — Bequests & major donors. Headline: a gift that outlives you and its recipient. Primary CTA: bequest conversation. Proof: perpetuity architecture.
- /room-f — Health sector. Headline: we take the debt off your patient's plate, completely. Primary CTA: referral MOU. Proof: referral system + case outcomes.
- /room-g — International. Headline: an Australian answer to a global end-of-life problem. Primary CTA: diaspora-donor briefing. Proof: universal-benefit framing + precedents.
Rooms A and C build first (Y1 activation). Rooms B and F next (Y2). Rooms D, E, G in their activation years.
8. Human decision rights — who approves what
| Decision | Owner | Escalates to |
|---|---|---|
| Annual marketing mix strategy | CMO Agent drafts → Laurence + Lisa + Corrina (YDT CMO) approve | Board |
| Individual campaign design <$5k spend | CMO Agent drafts → Corrina (YDT CMO) primary approver; Laurence + Lisa co-approve | Laurence if novel positioning |
| Individual campaign design ≥$5k spend | CMO Agent drafts → Laurence approves | Board |
| Public-facing marketing content (ads, social, newsletters) | Digital Marketing Agent drafts → any one of Laurence / Lisa / Corrina (YDT CMO) approves | Laurence if beneficiary depiction; Deen if legal claim; Carla only if a financial claim is being made |
| BD-linked content (proposals, MOUs, cultivation emails) | Digital Marketing Agent drafts → Lisa approves | Laurence for major donors; Board for MOUs over threshold |
| Prospect stage progression in CRM | Lisa | Laurence for Rooms A/B/D/E |
| Won-donor thank-you (major) | Laurence (personal) | Chair after appointment |
| Beneficiary depiction in any artefact | Laurence | Deen if lawful-basis question |
| Any tax-deductibility / DGR claim pre-endorsement | Blocked at agent level | Carla (CFO frame) & Deen (legal) after DGR issued; Corrina (marketing craft) |
| Any statistic / case study not in the citations index | Blocked at agent level | Deen (legal / evidence chain) adds to citations index if verified; Corrina (Human CMO · YDT) approves the phrasing |
8.5 Human Agency Scale — LWD's HA1–HA5 mapping
Section 8 answered who signs off what. This section answers a deeper question the Board will be asked in every governance review: on any given task, how much human agency is preserved? LWD adopts the Human Agency Scale (HAS) proposed by Stanford's Social & Language Technologies Lab as the shared vocabulary for that answer.
The five rungs, applied to LWD
Each rung below shows (a) the canonical Stanford definition and (b) how it maps to LWD's actual marketing and BD workflow — which agent or human sits at that rung, and for what class of task.
AI agent handles the task entirely on its own.
Canonical definition: the AI agent completes the task end-to-end without human intervention.
AI agent needs minimal human input for optimal performance.
Canonical definition: the AI can operate largely on its own but performs better with a brief human steer (a topic, a room, a compliance flag).
AI agent and human form an equal partnership, outperforming either alone.
Canonical definition: neither party could produce the artefact at the same quality independently; the collaboration is the unit of work. Dominant worker preference in 47 of 104 occupations (Stanford data).
AI agent requires human input to successfully complete the task.
Canonical definition: the agent cannot finish the task without specific human judgment on a critical step — usually domain, legal, or values judgment.
POST /api/agent-outputs/:id/approve requiring a human_approver ID that matches an active row in the humans table.AI agent cannot function without continuous human involvement.
Canonical definition: the human is present throughout — setting direction, arbitrating trade-offs, taking accountability for the final artefact.
Roster snapshot — who sits where
The following mirrors the humans table in D1 as at Session 8L. Approval endpoints accept any of these five IDs as human_approver. Roles are placeholder defaults; individuals may correct their titles in-line.
The AI stack
- HA1 · Digital Marketing Agent — Claude Sonnet 4.5 under a CMO-Agent brief. Cold production: organic social, ad variants, email drafts, metadata. Sits at H2 in practice (needs a brief), H1 only for boilerplate.
- HA2 · CMO Agent — Claude Sonnet 4.5 under Corrina's (YDT Human CMO) standing brief. Editorial review, campaign strategy, brief-writing for HA1. Sits at H3 when paired with Corrina + Laurence + Lisa.
- HA3 · SEO / Design / Web-Audit Agent (PROPOSED, Session 8M+) — A third agent role for AEO/SEO/GEO audits, website copy and design improvement drafting. Sits at H3 with a light human reviewer.
The named humans
- HA4 · Marketing approval authority triad (SO #5 · extended under SO #6) — Laurence Hugo (
humans.id = 4, co-founder, BD-team lead, ACL 387398 holder / RG 96 compliance reviewer), Lisa Hugo (humans.id = 2, Co-founder & Head of Philanthropy, BD-team lead), and Corrina McGowan (humans.id = 5, Human CMO · YourDigitalTeam · commercially engaged). The triad co-approves at HA4 for marketing content. Under SO #6 the triad also sits at HA5 (see below) — the triad sits WITH the Board at terminal approval, not just at HA3 sight-check. - HA4 · Carla Oliver (CFO frame only) —
humans.id = 1, CFO (project phase → BAU on project completion). Sits at HA4 for CFO-frame material only (financial claims, salary bands, tax-time framing). Not a marketing approver. Holds the $10,000 AUD Anthropic project-lifetime cap (SO #6C). - HA5 · Marketing-authority triad (SO #6) — Laurence Hugo, Lisa Hugo, Corrina McGowan sit WITH the Board at terminal approval. Marketing content for release requires triad + Board.
- HA5 · Lisa Hugo —
humans.id = 2. Co-founder & Head of Philanthropy. BD-team lead. Donor-relations sign-off. - HA5 · Deen Van Praag —
humans.id = 3. Founding Director. Legal-claim and directorial sign-off. - HA5 · Corrina McGowan —
humans.id = 5. Human CMO · YourDigitalTeam · commercially engaged. Programme, editorial, and marketing-craft sign-off.
9. Build sequence — how this design becomes a running system
- Session 1 (this session) — Door 4 scaffolding on the portal + this design artefact. No live agents. No CRM yet. Board can review the design.
- Session 2 — CRM. Cloudflare D1 database with the six tables in §5. Hono API routes. A working kanban / list UI for Lisa. Nurture-templates library seeded from existing collateral (emails from paid-ads.html, sequences from media-campaign.html). No agents wired. Lisa can log prospects, move them through stages, see per-room pipeline. This is the point at which Lisa becomes productive.
- Session 3 — Agents wired. Anthropic API key held as a Cloudflare secret. Hono routes for the CMO Agent and the Digital Marketing Agent. System prompts embed the Seven Founding Principles + Rooms A–G + pre-DGR guardrails + citations index. Both agents write to
agent_outputswith the audit fields in §5.4. Human approval UI in front of both. - Session 4 — Room A + Room C landing pages. The two Y1 rooms. Full AEO / SEO / GEO structured data. Message-match with existing ad creative. Conversion into the CRM.
- Session 5 — Room B + Room F landing pages. The two Y2 rooms.
- Session 6 — End-to-end wiring. Agent output → nurture-templates → CRM triggers → sent touchpoints → touchpoints table updated → next action scheduled. Complete loop.
- Session 7 — Rooms D, E, G landing pages + KPI dashboards + Go-Live checklist.
10. Open questions for the Board / Laurence / Deen
- Budget ceiling for CMO Agent auto-approved campaigns — draft assumes $5k. Confirm or adjust.
- Named human approver default — RESOLVED at Sprint 28.13: the CFO Advisor (Carla) is not the day-to-day marketing approver. Standing Order #5 sets marketing approval authority as Laurence + Lisa + Corrina (Human CMO · YourDigitalTeam). Carla holds CFO-frame review only. See correction banner at top of this document.
- Data residency — Cloudflare D1 replicates globally at the edge; primary region is configurable. Confirm APRA / Privacy Act requirements are met by default Cloudflare posture (draft assumes yes; Deen to confirm).
- Anthropic API cost governance — RESOLVED at Sprint 32 under SO #6C: Carla Oliver (CFO) has set the cap at $10,000 AUD TOTAL project lifetime (not monthly), with a halt-at-80% threshold of $8,000 AUD cumulative. Baseline reconciled at $1,695 AUD. Stored authoritatively in D1
budget_configandbudget_ledger. Live status: /sprint32. - Beneficiary consent for case-study use in agent outputs — the three composite cases (Margaret / David / Kwame) are already composite, not identifying. If real cases are ever added, the beneficiary-depiction guardrails (Sprint 0.4) apply verbatim.
- Board reporting cadence — monthly Marketing Report from the CMO Agent to the Board. Confirm format (attached to Board pack).
- v0.4 (Sprint 32, 2026-08-06) — Standing Order #6 propagation. Four assumptions banked from user directive: (A) Laurence + Lisa are the BD / donor-relations team; (B) Corrina McGowan commercially engaged (paid, not volunteer/advisor) and moves from HA3 sight-check to HA4 and HA5 alongside Laurence + Lisa — triad sits WITH Board at terminal approval; (C) Carla Oliver is CFO (project phase → BAU on project completion) with a $10,000 AUD Anthropic project-lifetime cap ($8,000 halt) — not a marketing approver; (D) Board Meeting hard deadline is Tuesday 25 August 2026 with three decisions (Kit-04 HA4 vote, DRP v1.0 adoption, launch-scope resolution). Body edits: correction banner extended with the four SO #6 items; §3 CMO-Agent + Digital-Marketing-Agent human-approver blocks reworded to reference SO #6 and BD-team framing; §8 lwd-map H3 / H4 / H5 blocks rewritten (triad sits with Board at H4 and H5; Carla's $10k cap noted); §8.5 HAS ladder HA4/HA5 entries rewritten around the triad (residual “Lisa Newman” legacy phrasing corrected to Lisa Hugo); §10 Anthropic-API cost-governance open question resolved with the $10k figure. New D1 tables banked this sprint (migrations 9341/9342/9343):
budget_config,budget_ledger,governance_calendar,governance_agenda_item. See /sprint32 and /board-pack-2026-08-25 for live status. - v0.3 (Sprint 28.13, 2026-08-06 · body refresh completed Sprint 29.4) — Standing Order #5 correction: marketing approval authority = Laurence Hugo + Lisa Hugo + Corrina McGowan (Human CMO · YourDigitalTeam), NOT Carla Oliver. Correction banner added at top of document. Section 8 open-question resolved. Sprint 29.4 completed the residual body pass: §3 CMO-Agent and Digital-Marketing-Agent human-approver blocks rewritten to the triad with Carla scoped to CFO-frame only; §4 TOFU owner attribution reattributed; §6
humanstable schema description and reference-table intro rewritten to name all 5 rows (post migration 9334); §8 lwd-map rewritten with the triad as marketing approval authority; §8.5 HAS ladder HA4 entry rewritten to the triad plus Carla's CFO-frame lane; §7 fact-check row for citations-index reattributed to Deen (legal) + Corrina (phrasing). - v0.2 (Session 8L, 2026-07-30) — added §8.5 Human Agency Scale citing Stanford SALT Lab (arXiv:2506.06576v2). Mapped LWD's blended workforce to H1–H5: HA1 Digital Marketing Agent, HA2 CMO Agent, HA3 proposed SEO/Design/Web-Audit Agent, HA4 Carla + Laurence, HA5 Lisa + Deen + Corrina. Roster expanded from three to five active human reviewers via D1 migration
0003_reviewers_expansion.sql. Note: the "Carla approves marketing" model in this v0.2 draft is superseded by v0.3 above. - v0.1 — initial DRAFT for review. Two agents specified, funnel mapped, CRM schema defined, seven landing pages outlined, human decision rights tabled, seven-session build sequence set.