Command Center
Aria chat_bubble Talk to Aria
Briefing

EFD — HighLevel Build Runbook + Voice AI Prompt

Phased 14-21 day build sequence sourced from Circa's 470+ HighLevel helpdesk markdown files. Master snapshot creation, A2P registration, Voice AI agent setup, 5 workflows. Plus the master Voice AI system prompt + tone-of-voice guide.

Updated April 27, 2026

📅

Four Phases — Zero to First Client Live

Phase 0 — Pre-build (Day 0-2: Apr 27-29)

  • Register second A2P Standard Brand under Aspire's EIN (Fast Track $3, ~3 day approval)
  • File campaign use case = Mixed (transactional + low-volume marketing)
  • Phone strategy decision: LC Phone, per-sub-account numbers (skip Twilio rebill, skip shared pool)
  • Open Stripe in EFD's new EIN (separate from Aspire's Stripe)

Phase 1 — Master Snapshot Build (Day 3-7: Apr 30-May 4)

  • Create EFD-MASTER sub-account (never billable)
  • Internal white-label branding + brand boards
  • 10 custom fields (location_legal_name, on_call_director_phone_1/2/3, etc.)
  • Voice AI agent — variables wired to {{custom_values.*}} for cross-client portability
  • Five workflows: after-hours warm transfer, missed-call text-back, voicemail-to-text, first-call alert, preneed 5-touch nurture
  • Round-robin arrangement-meeting calendar
  • Family Services pipeline (First-Call → Arrangement Booked → Service Scheduled → Aftercare)
  • Forms (preneed inquiry, online arrangement starter) with consent checkbox
  • Funnels (preneed capture, after-hours holding page)
  • SMS/email templates (first-call, arrangement reminders, preneed nurture, aftercare)
  • Snapshot creation with Voice AI included; IP-protected

Phase 2 — Pre-client Testing (Day 8-10: May 5-7)

  • 100-call simulation against EFD-MASTER (all required scenarios + edge cases)
  • Pass criterion: 95%+ of 100 calls pass
  • Call-recording compliance configuration (Phone Numbers → Edit → Enable + Disclosure for two-party states)
  • Funeral-director consultant review of master prompt ($500-1,500 one-time)
  • Re-run 100 simulated calls after consultant feedback → v1.0 prompt

Phase 3 — First Client Onboarding (Day 11-14: May 8-11)

  • Clone snapshot → new sub-account in client's legal DBA
  • Per-client customization (custom values, brand board, KB FAQ corpus)
  • Buy + assign LC Phone number in client's area code; reduce inbound timeout
  • A2P campaign submission for client (5-10 business day approval — plan timeline)
  • Reputation engine: connect Google Business Profile + 7-day-post-service review trigger
  • Director training (30-min Loom) + 1-page handoff doc

Phase 4 — Steady-state Ops (Day 15+: May 12+)

  • Weekly QA: sample 10 calls/client/week, score against rubric
  • Monthly reporting deck per client (PDF, emailed)
  • Monthly preneed nurture content drop (snapshot refresh + push)
  • Pricing trigger: 2 consecutive months of >100 calls answered → bump from pilot to standard
  • Churn monitoring: drop in answered-call rate, zero opportunities in 14 days, AI-related negative review
⚠️

Eight Funeral-Vertical Gotchas (Circa-Documented)

  1. Voice AI snapshots do NOT carry phone numbers. Every clone needs fresh number purchase + manual assignment + recording disclosure toggle.
  2. Inbound call timeout trap. Local voicemail can pick up before GHL timeout — Voice AI never answers. Reduce inbound timeout in phone config.
  3. A2P campaign is per-sub-account, not per-brand. Brand approval = one-time; each client still needs campaign vetted (5-10 days). Plan timeline.
  4. Snapshot refresh deletes contacts mid-workflow. Yanks contacts at removed wait steps. Test in sandbox before pushing to live preneed nurture sequences.
  5. Geo permissions can block international callers. Snowbird clientele calling from Mexico get blocked unless enabled.
  6. Default sub-account number ≠ Voice AI number. Always provision a second number for the AI; never assign to default.
  7. Call recording disclosure is on the phone number, not the agent. Easy to forget at clone; legal exposure in two-party states (CA/FL/PA/MA/IL/MD/WA).
  8. A2P standard brand has Fast Track fee ($3). Pay it. Without it brand approval drifts past 7 business days.
🎙️

Voice AI Master Prompt — Highlights

Full prompt + per-client variables + 100-call simulation set in memory/drawers/aspire-ventures/efd/voice_ai_master_prompt.md. Below: the 10 hard rules + the 4 intents.

Four intents the AI must identify:

  1. DEATH_NOTIFICATION — someone has died → warm-transfer to on-call director within 30 sec
  2. PRICING_INQUIRY — capture name + callback number, schedule director call-back
  3. PRENEED_INQUIRY — book appointment OR start 5-touch nurture sequence
  4. OTHER — vendor, solicitor, wrong number, or service-in-progress

Ten hard rules (never violate):

  1. NEVER say "I'm sorry for your loss." That's for the human director.
  2. NEVER quote a specific price.
  3. NEVER promise a service the funeral home doesn't offer.
  4. NEVER use the word "death" before the caller does. Use "passed," "passing," "what happened."
  5. NEVER say "I'm an AI" unless directly asked.
  6. If asked: "I'm an automated assistant for {business}. A real director will be on the line in under a minute."
  7. NEVER attempt to comfort beyond "Take your time. I'm here when you're ready."
  8. NEVER share any caller's information with another caller.
  9. NEVER claim to know what happened, the deceased's wishes, or anything beyond what the current caller said.
  10. If caller is in distress and unable to communicate, transfer immediately without further questions.

The prompt MUST be reviewed by a retired funeral director consultant ($500-1,500 one-time engagement) BEFORE any client goes live. This replaces Jaime's UX-research seat — load-bearing for grief-adjacent reliability.

Full runbook with every step + Circa doc citations: memory/drawers/aspire-ventures/efd/build_runbook.md

Voice AI master prompt + 100-call simulation set: memory/drawers/aspire-ventures/efd/voice_ai_master_prompt.md