n8n Pipeline Spec

n8n Pipeline Spec — AI Opportunity Audit Automation

Intake → Claude analysis → Branded PDF + shareable graphic + client portal → GHL delivery

Goal: Turn a paid audit purchase into a delivered, impressive, mostly-automated package with minimal human time per engagement. Target: 3–5 hours of your hands-on time per audit (interviews + final review + Loom), the rest automated.

Stack: n8n (orchestration), Claude Sonnet + Haiku (analysis), Supabase (data store), Carbone or Gotenberg (PDF, self-hosted on Hostinger VPS), Placid or Bannerbear (shareable graphic), Metabase + Softr (client portal), GHL (delivery + CRM + nurture), Loom (walkthrough), Stripe or GHL payments (checkout).

This doc covers two pipelines:

1. Lead-magnet pipeline (free scorecard → nurture → paid audit) — lightweight, always-on.
2. Fulfillment pipeline (paid audit → delivered package) — the main event.

---

PIPELINE 1 — LEAD MAGNET (ScoreApp → GHL nurture)

ScoreApp webhook ──► n8n ──┬──► GHL: upsert contact + add tag "scorecard_complete"
                           ├──► Supabase: insert scorecard_result row
                           ├──► Placid: generate branded score graphic → upload to GHL media
                           └──► GHL: trigger nurture workflow (email 1: score + audit offer)

Nodes:

1. Webhook node — receives ScoreApp payload (see Asset 2 for JSON shape). URL: https://[your-n8n]/webhook/scoreapp.
2. Function/Set node — normalize: extract contact, score.overall, score.dimensions, score.binding_constraints, answers.
3. GHL node — Upsert Contact with custom fields: ai_readiness_score, ai_readiness_band, binding_constraint_1, binding_constraint_2. Add tag: lead_scorecard. Add to workflow Audit Nurture v1.
4. Supabase node — INSERT into scorecard_results (full payload + timestamp) for analytics and later benchmarking.
5. Placid node (HTTP or native) — POST score + band + dimension radar data to a Placid template → returns branded share-image URL.
6. GHL node — send Email 1: “Your AI Readiness Score: [X]/100” with the graphic attached + link to the sales page pricing section.
7. IF node — if score.overall between 40 and 79 (prime buyer band) → add tag hot_audit_lead → trigger a Slack/email internal alert so you can do personal outreach.

Nurture sequence (in GHL, not n8n):

• Email 1 (immediate): score + graphic + audit offer.
• Email 2 (+2 days): case study / testimonial.
• Email 3 (+5 days): “only 4 slots this month” scarcity.
• Email 4 (+10 days): “what an audit actually looks like” (sample report teaser).

---

PIPELINE 2 — FULFILLMENT (paid audit → delivered package)

This is the core. Steps map to the delivery workflow in the blueprint. [Auto] = automated, [Semi] = automated draft + you review, [Manual] = you.

Step 0 — Checkout & kickoff [Auto]

Stripe/GHL checkout (success) ──► n8n ──► GHL: move contact to "Audit In Progress"
                                      ──► Supabase: create engagement row (status: kickoff)
                                      ──► GHL email: "You're in" + Fillout intake link + Calendly booking

Step 1 — Deep intake [Auto → Semi]

Tool: Fillout (or Typeform). 7-dimension deep questionnaire — more granular than the lead-magnet scorecard: 5–7 questions per dimension + a “use-case inventory” section where they list their top 10 most time-consuming tasks (impact, effort, data source, owner, frequency).

Fillout webhook ──► n8n ──► Supabase: store intake_responses (engagement_id)
                          ──► GHL: tag "intake_complete", update custom fields
                          ──► (IF node) if Tier = Deep Dive → send Fillout "stakeholder roster" form to collect 5–6 interviewee names

Step 2 — Discovery / stakeholder interviews [Manual]

You run 1 (Snapshot) to 6 (Deep Dive) interviews on GHL-calendar bookings. Record each (with consent).

Automation assist: Each interview recording → n8n via Loom/Drive webhook → Claude Haiku node transcribes + extracts structured notes against the 7 dimensions (prompt template below) → stores structured notes in Supabase under the engagement.

Loom/Drive new video ──► n8n ──► Haiku: transcribe + extract dimension notes ──► Supabase

Step 3 — Analysis (AI-assisted) [Semi — the brain]

This is the most valuable node in the whole pipeline. n8n orchestrates a multi-step Claude Sonnet analysis.

n8n (manual trigger "Run Analysis")
  ──► gather: intake_responses + all interview notes (from Supabase)
  ──► Claude Sonnet node: "Score the 7 dimensions"
  ──► Claude Sonnet node: "Generate opportunity matrix (15–20 use cases)"
  ──► Claude Sonnet node: "Draft 30/60/90 roadmap"
  ──► Claude Sonnet node: "Draft ROI projection"
  ──► Claude Haiku node: "Format to style-guide HTML sections"
  ──► Supabase: store all drafts (status: review)
  ──► notify you (Slack/email): "Analysis ready for review"

Each Claude node = one focused prompt. See PROMPT LIBRARY below for exact templates.

Step 4 — Your review + edits [Manual]

You open a simple admin view (Softr on Supabase, or just the Supabase dashboard) → review the drafted scores, opportunities, roadmap, ROI → edit any you disagree with → mark status: approved.

Supabase status: review → approved (you flip it) ──► n8n trigger "Generate Deliverables"

Step 5 — Deliverable production [Auto]

n8n "Generate Deliverables" trigger (on approval)
  ──► assemble final JSON (approved scores, opportunities, roadmap, ROI, client branding)
  ──► BRANCH A: Carbone/Gotenberg node → branded PDF report
  ──► BRANCH B: Placid node → shareable score graphic (radar + headline number + client logo)
  ──► BRANCH C: Metabase/Softr → populate portal (radar chart, heatmap, roadmap tiles)
  ──► n8n waits for all 3 ──► Supabase: store asset URLs (status: deliverables_ready)

Branch A — PDF (Carbone on Hostinger VPS):

• Template: a branded HTML/DOCX template in Carbone. Tags: {{client_name}}, {{score}}, {{band}}, {{#each dimensions}}, {{#each opportunities}}, {{#each roadmap_phases}}, {{roi_summary}}.
• n8n POSTs the final JSON to Carbone’s API → returns PDF → n8n uploads to GHL media + Supabase.

Branch B — Shareable graphic (Placid):

• Template: a square 1080×1080 + a LinkedIn-landscape 1200×627, both with: client logo, score number, band label, mini radar.
• Auto-generated, so the client can post immediately.

Branch C — Interactive portal (Metabase + Softr):

• Metabase reads from Supabase engagement_results → radar chart + dimension bars + opportunity heatmap + roadmap list.
• Softr page (gated, per-engagement) embeds the Metabase dashboards → looks like a custom client portal.
• n8n writes the Softr row + generates a magic link.

Step 6 — Loom walkthrough [Semi → Manual]

n8n: status deliverables_ready ──► notify you: "Record Loom for [client]"

You record a 5–8 min Loom: one key message (their binding constraints), walk the radar, highlight top 3 opportunities, point to the roadmap, end with CTA (proceed to build? book debrief?).

Automation assist: n8n can pre-draft the Loom script from the approved analysis (Claude Haiku node) so you just read/refine it.

After Loom upload ──► Loom webhook ──► n8n ──► Supabase: store loom_url (status: loom_ready)

Step 7 — Delivery [Auto]

n8n: status loom_ready
  ──► GHL email: "Your AI Opportunity Audit is ready"
       - PDF attached / linked
       - Portal magic link
       - Loom embed
       - "Book your debrief call" link (GHL calendar)
  ──► GHL: tag "audit_delivered", move to "Debrief" stage
  ──► Supabase: status: delivered
  ──► (delay 24h) GHL email: "Share your AI Readiness Score" with the Placid graphic + referral link

Step 8 — Debrief call [Manual]

You run the 60-min recorded debrief. n8n logs attendance + sends a “recap + next steps” email via GHL after.

Step 9 — Referral loop activation [Auto]

Audit delivered ──► GHL: create affiliate record for the client (GHL Affiliate Manager)
                ──► generate their unique referral link
                ──► email: "Earn $400 for every business you refer" with their link + shareable score graphic
                ──► if anyone books via their link → GHL attributes → 30 days after they pay → Wise payout (you trigger manually or automate via Wise API)

---

PROMPT LIBRARY (Claude nodes)

Each prompt is a separate Claude node. Feed structured input, get structured output (JSON).

Prompt 1 — Score the 7 dimensions (Sonnet)

SYSTEM: You are a fractional Chief AI Officer conducting an AI Opportunity Audit.
You are rigorous, honest, and avoid hype. You score on evidence, not optimism.

INPUT: {intake_responses JSON} + {interview_notes JSON}

TASK: Score each of the 7 dimensions 1–5 (use 0.5 increments if needed),
with a 1-paragraph justification per dimension citing specific evidence from the input.

DIMENSIONS: strategy_leadership, data_readiness, technology_tooling,
process_automation, talent_literacy, governance_compliance, value_roi.

RULES:
- Score by the binding-constraint method: the overall maturity is gated by the
  two lowest dimensions, not the average. Note which two are lowest.
- If evidence is missing for a dimension, score conservatively (don't assume
  capability that isn't stated).
- 1 = nothing in place, 5 = leading-edge. Be demanding; most businesses score 2–3.

OUTPUT: JSON { "dimensions": [{"name","score","justification"}],
"binding_constraints": ["dim1","dim2"], "overall_maturity_note": "..." }

Prompt 2 — Opportunity matrix (Sonnet)

SYSTEM: You are an AI automation strategist. You rank real opportunities by
impact × effort, grounded in the client's actual processes and data.

INPUT: {intake + interview notes} + {dimension scores}

TASK: Identify 15–20 concrete AI/automation opportunities for this business.
For each, provide:
- name
- description (1–2 sentences)
- impact_score (1–5)
- effort_score (1–5)
- data_readiness (ready / partial / not ready)
- estimated_hours_saved_per_week (range)
- estimated_annual_savings_usd (range, show math: hours × loaded cost × 48 wks)
- time_to_value (weeks)
- owner_suggestion
- quadrant (quick_win / strategic_bet / fill_in / park)

RULES:
- Only include opportunities grounded in the client's stated processes.
  No generic "implement a chatbot" unless their intake supports it.
- Rank by (impact ÷ effort) descending.
- Be honest about data_readiness — if a use case needs data they don't have,
  flag it as a prerequisite, not a quick win.

OUTPUT: JSON array of 15–20 opportunity objects.

Prompt 3 — 30/60/90 roadmap (Sonnet)

SYSTEM: You are a fractional CAIO building an implementation roadmap.

INPUT: {dimension scores} + {opportunity matrix} + {binding_constraints}

TASK: Create a 30/60/90-day roadmap.

STRUCTURE:
- Days 1–30 (Foundations): fix the binding constraints. 2–4 initiatives.
- Days 31–60 (Quick Wins): ship the highest impact/effort opportunities. 3–5 initiatives.
- Days 61–90 (Scale): strategic bets + governance hardening. 2–4 initiatives.

For each initiative: name, what, why (ties to which dimension/opportunity),
owner, dependency, success metric.

RULES:
- The 30-day phase MUST address the binding constraints first.
- Be specific enough that a competent operator could start Monday.

OUTPUT: JSON { "phases": [{"phase","initiatives":[...]}] }

Prompt 4 — ROI projection (Sonnet)

SYSTEM: You are a financial analyst building a defensible ROI projection.

INPUT: {opportunity matrix}

TASK: Produce a 12-month and 24-month cumulative ROI projection.

For each: sum the estimated_annual_savings across implemented opportunities,
apply a realistic ramp (30% year-1 adoption, 70% year-2), subtract an
implementation cost estimate (your hourly × hours, or a flat $X per initiative),
and present cumulative net savings.

Include a "money slide" headline: "Estimated 24-month net savings: $X
(YYY% ROI on a $3,000 audit)."

RULES: be conservative. Round down. Show the math. If data is thin,
give a range with a stated assumption.

OUTPUT: JSON { "12mo": {...}, "24mo": {...}, "money_slide": "...", "assumptions": [...] }

Prompt 5 — Loom script draft (Haiku, cheap)

INPUT: {approved scores, top 3 opportunities, binding constraints, money_slide}

TASK: Write a 5–8 minute Loom script for the consultant to read.
Tone: confident, warm, no hype. Structure:
- Hook (their binding constraint, in plain English)
- The score + what it means
- Top 3 opportunities (one sentence each, with the savings number)
- The 30-day priority
- Close + CTA (book debrief / proceed to build)

OUTPUT: a markdown script with [SLIDE/CURSOR] cues.

---

SUPABASE SCHEMA (minimum viable)

CREATE TABLE engagements (
  id UUID PRIMARY KEY,
  ghl_contact_id TEXT,
  client_name TEXT,
  company TEXT,
  tier TEXT, -- snapshot | opportunity | deep_dive
  status TEXT, -- kickoff|intake|analysis|review|deliverables|loom|delivered
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE TABLE scorecard_results (
  id UUID PRIMARY KEY,
  engagement_id UUID REFERENCES engagements(id),
  overall_score INT,
  band TEXT,
  dimensions JSONB,
  binding_constraints JSONB,
  answers JSONB,
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE TABLE intake_responses (
  id UUID PRIMARY KEY,
  engagement_id UUID REFERENCES engagements(id),
  responses JSONB,
  use_case_inventory JSONB,
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE TABLE interview_notes (
  id UUID PRIMARY KEY,
  engagement_id UUID REFERENCES engagements(id),
  interviewee_name TEXT,
  transcript TEXT,
  structured_notes JSONB, -- extracted by Haiku
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE TABLE analysis_drafts (
  id UUID PRIMARY KEY,
  engagement_id UUID REFERENCES engagements(id),
  dimension_scores JSONB,
  opportunities JSONB,
  roadmap JSONB,
  roi_projection JSONB,
  loom_script TEXT,
  status TEXT, -- draft|approved
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE TABLE deliverables (
  id UUID PRIMARY KEY,
  engagement_id UUID REFERENCES engagements(id),
  pdf_url TEXT,
  graphic_url TEXT,
  portal_url TEXT,
  loom_url TEXT,
  created_at TIMESTAMPTZ DEFAULT now()
);

Metabase points at these tables; Softr builds the client-facing portal on top.

---

HOSTINGER VPS DEPLOYMENT

Self-host the heavy lifters to avoid per-render fees:

• n8n — Docker on your VPS. Use the official n8n Docker image + Postgres for n8n’s own DB.
• Carbone or Gotenberg — Docker container alongside n8n. Carbone for DOCX/HTML→PDF, Gotenberg for pure HTML→PDF (preserves your CSS). Expose on localhost only; n8n calls it internally.
• Supabase — use Supabase Cloud (free tier is fine to start); only self-host Postgres if you want full control.

This keeps recurring costs near $0 beyond your existing Hostinger VPS.

---

WHAT YOU AUTOMATE vs. KEEP HUMAN

Automated (n8n + Claude)	You (manual)
Intake capture & validation	Discovery & stakeholder interviews
Transcription + note extraction	Push back / probe in interviews
Dimension scoring (draft)	Final score judgment + edits
Opportunity matrix (draft)	Final priority judgment
Roadmap + ROI (draft)	Final review of roadmap/ROI
PDF / graphic / portal generation	Loom narration (your voice = trust)
Delivery email + nurture	Debrief call
Referral link generation	Relationship / sales follow-up

Time per audit after the pipeline is built: ~3–5 hours of your time (mostly interviews + review + Loom), vs. 15–20 hours fully manual. That’s what makes a $3K audit profitable for a solo.

---

BUILD ORDER (recommended sequence)

1. Week 1 — Supabase schema + GHL custom fields + Fillout intake form. Get data flowing into storage.
2. Week 2 — n8n intake → Supabase → GHL (Pipeline 2 Steps 0–1). Verify data lands.
3. Week 3 — Claude analysis nodes (Step 3) with the prompt library. Test on your LMS data (you already have it). Iterate prompts until output is genuinely good.
4. Week 4 — Carbone PDF template + Placid graphic template. The “looks impressive” layer.
5. Week 5 — Softr/Metabase portal + Loom integration + delivery email (Steps 5–7).
6. Week 6 — GHL affiliate + referral email (Step 9). Run 2 pilot audits end-to-end.

Critical: Run the full pipeline on yourself first (audit your own LMS engagement) — you’ll find every broken node before a paying client does.