Automation & Connector Map

How the whole machine connects — for J&M’s own business (finding and closing clients) and for client builds (what J&M installs). The principle behind every row: repeatable work becomes saved, deterministic automation with a single source of truth; humans keep the judgment calls (Prime Directive; field-tested human-in-the-loop).


Part 1 — J&M’s own sales-to-delivery pipeline

 FIND ─────────► QUALIFY ────────► PITCH ──────────► CLOSE ──────────► DELIVER ────────► EXPAND
 outbound +      discovery call    audit report      walkthrough +     Concierge         referrals +
 content         (SPIN)            (Claude Design)           proposal     upsell
Stage Task Automated with Human keeps
Find Cold email sequence Smartlead/Instantly + warmed domain Writing the observed-gap hook
Find LinkedIn touches Heyreach The reply conversation
Find Prospect logging (no dupes) prospect system / knowledge.db Deciding who’s ICP
Qualify Booking cal.com (public discovery link) + text reminders Running the call
Qualify Call capture Zoom + Fathom (auto-transcribe) Asking SPIN questions, listening 80%
Pitch Findings → analysis Claude AI-analysis skill on transcript Filtering bad-fit recs, verifying tool prices
Pitch Analysis → proposal the internal playbook (JSON → MD) + Claude Design The numbers being honest
Close Proposal → shareable doc ~/google/md_to_gdoc.py The walkthrough + objection handling
Deliver Post-call admin Claude skills (call log, action items, follow-up email) → Notion The building itself
Deliver Async support Voxer/Slack, 12-hr SLA The actual answers
Expand Referral ask Templated intro drafts Asking at peak value

Connectors: cal.com → Zoom/Fathom (calendar) → transcript → Claude skill → proposal-generator (JSON hand-off, field names match the notes template) → Claude Design/GDoc → Notion hub. The one data spine is the findings JSON (discovery notes → proposal → build log all speak it).


Part 2 — Client build connector map (the two stacks)

Universal shape (stack-agnostic)

        INBOUND                    BRAIN                    ACTION                 SYSTEM OF RECORD
  calls · forms · email ──►  Claude (draft/analyze) ──►  SMS · email · book ──►  CRM · Notion · dashboards
  DMs · site chat            (human approves early)      (A2P 10DLC + consent)

Connector matrix — what wires to what

From To Via Carries
Missed call SMS text-back GHL native / Twilio+n8n “sorry we missed you” + booking thread
Web form Speed-to-lead skill Zapier/n8n lead details → personalized draft
Inbound call (unanswered) AI receptionist Retell.ai qualify → book / message
Job complete (CRM event) Review request GHL / n8n+Twilio delayed SMS w/ review link
Job notes Quote skill Claude + price book notes → priced quote for review
New lead (any channel) CRM record GHL / n8n → EspoCRM contact + source + stage
Booking / change Team notify cal.com/GHL + SMS who/when → assigned staff
Any client data Dashboards Metabase KPIs (leads, response time, reviews)

The three invariants (every client, every stack)

  1. Brain = Claude. Drafting, analysis, summarization all route through Claude skills/projects.
  2. Human-in-the-loop first. AI drafts, human approves. Loosen guardrails only as trust builds (start with full review).
  3. SMS = A2P 10DLC + consent, always. No exceptions. TCPA is not optional.

Part 3 — Automation candidate checklist (what to look for on the audit)

A task is a good automation candidate when it’s repeatable, rule-based, and currently manual. Score each against the Impact-Effort matrix:

Quick Wins (ship < 1 week) go first — they fund the guarantee and build trust for the bigger projects. Never automate a process before you’ve optimized it.

← Client One-PagerLaunch Checklist →