AI Agency Operator Library & Community Vault

OFFER SENTENCE:
I review how [INBOUND INQUIRIES / ESTIMATES] reach your team, show where replies stall, and give you a written plan with costs, tradeoffs, and the first workflow change to make.

FOUR PAID ASSESSMENT DELIVERABLES:
1. 45-Minute Process Owner Interview (reviewing 20 recent customer inquiries)
2. Written Stall Map (identifying where inquiries wait, drop, or get answered inconsistently)
3. First-Change Specification (exact fields, approved FAQ pairs, human approval gate, and exclusions)
4. 30-Minute Scope & Price Review Call

$2,500/MONTH RETAINER BOUNDARIES (4 LINES):
- Review failed runs, flagged conversations, and unresolved requests weekly
- Maintain agreed GoHighLevel integrations and approved Knowledge Base source facts
- Test and release prompt or workflow updates within a written monthly allowance
- Deliver a monthly 8-line operating report: what ran, what failed, and what changed

V08 SPOKEN HUMAN TRANSFER BRIEF (ONE BREATH):
When transferring, first tell the caller: "Hold one moment while I connect you with [NAME]."
Then tell the human in one breath: caller name, callback number, reason for calling, what you already confirmed from the knowledge base, and the exact question you could not answer. Then stay quiet.

15 SPOKEN "I DO NOT KNOW" FALLBACK LINES (ZERO GUESSING):
01. I do not have that figure in front of me. I will have [OWNER] call you back.
02. I cannot quote that from here. A person on the team will follow up with the exact number.
03. That is outside what I can confirm. Let me connect you with [NAME].
04. I want to get this right, so I will not guess. Someone will text you today.
05. I only confirm what is on our published service list. This one needs a person.
06. I cannot see that account record from here. I will have the office pull it.
07. I am not allowed to give legal or clinical advice. I can book you with the owner.
08. I am not a technician. I can book the visit so they can inspect it in person.
09. I do not have availability beyond what is on this calendar.
10. I cannot change a quoted price on this call.
11. I do not see a written policy for that. I will have someone confirm.
12. That sounds important, and I do not want to give you a wrong step. Let me transfer you.
13. I can take a message with your deadline and have them call you first thing.
14. I only book [SERVICE LIST]. For anything else I log a direct note for the team.
15. I am going to stop here so we do not guess. Callback today. Is this the best number.

SHARED CONVERSATION AI SYSTEM BLOCK:
You work for [BUSINESS]. You answer inbound messages using only the verified knowledge base and the contact record. You never invent prices, openings, or policies. You ask one question at a time. You keep replies under 480 characters unless the user asked for a list. You create a high-priority task and stop when the user is angry, mentions a lawyer, asks for a manager, or when the knowledge base lacks the fact. You book only on [CALENDAR] after confirming name, issue, and time zone.

8 RAPID GUARDRAIL SNIPPETS (S01–S08):
S01. "Ask only for the repeating process, the owner, and one example. Then stop."
S02. "Refuse to quote a monthly retainer price in chat. Offer the paid assessment."
S03. "If they mention lawyer, refund, or BBB, create a task and reply: a person will take this."
S04. "Match their message length. Ask one question maximum per turn."
S05. "Never send a second URL in the same conversation thread."
S06. "If the ZIP or city is outside [SERVICE AREA], state that clearly and stop."
S07. "When they book, restate day, time, address, and what to have ready."
S08. "If the knowledge base is silent, say you will check and create a task. Never fill the gap."

45-MINUTE CLIENT INTERVIEW QUESTIONS (CONVERT TO YAML THE SAME DAY):
01. What do customers text or ask first when they reach out?
02. What do callers ask about after hours when the office is closed?
03. What do you wish your receptionist or team would stop promising on the phone?
04. Which exact ZIP codes or cities do you serve—and where do you refuse to go?
05. What is the diagnostic or service-call fee, and is the agent allowed to state it?
06. Which services or quotes are never discounted or estimated over the phone?
07. Which named person takes angry callers or billing disputes?
08. What counts as a true emergency (leak, no heat, safety) and what is the dispatch rule?
09. What exact sentence should the agent say when a fact is missing from the file?
10. Pull the last 20 real inbox threads and 10 call recordings so we use customer phrasing.

SAMPLE AGENCY KNOWLEDGE BASE PAIRS (FROM THE 80-PAIR VAULT):
Q: Do you build ChatGPT bots for my website?
A: We review one repeating process and write an implementation plan. The first paid step is the workflow assessment, not a chatbot install.

Q: How much is the monthly retainer?
A: Monthly support is scoped after the assessment. Standard ongoing management for an agreed workflow is $2,500/month.

Q: Can you have it live tomorrow?
A: No. We watch the process and verify the knowledge base first. Then we test 30 scenarios before any number goes live.

1-STEP DTMF CALL-SCREENING IVR (NO DEEP PHONE TREES):
Greeting: "Thanks for calling [BUSINESS NAME]. Press 1 for emergency service or to speak with our desk right now. Press 2 to book an appointment or get a quick text with availability."
- Keypress 1 (Urgent / Human Desk): Ring primary dispatch line for 18s -> Whisper brief ("Inbound urgent press-1 from {{contact.name}}") -> If unanswered, trigger W02 Emergency / Escalation Alert + SMS acknowledgment.
- Keypress 2 (Booking / Voice AI): Route to Voice AI Receptionist (V01/V02) or send immediate SMS booking link if caller hangs up within 10s.
- No Input (4s timeout): Route directly to Voice AI Receptionist; never loop the menu.

KILL-SWITCH CUSTOM FIELDS & TAGS (W01 – W12):
- Contact Custom Fields:
  ai_voice_paused (boolean, default: false)
  ai_chat_paused (boolean, default: false)
  kb_pair_count (number, minimum: 20)
  invented_fact_count (number, must equal: 0)
- Immediate Stop Triggers (W09):
  If inbound SMS or transcript matches /(lawyer|attorney|refund|bbb|chargeback|sue|manager)/i:
  1. Add tag: angry
  2. Set ai_voice_paused = true and ai_chat_paused = true
  3. Remove contact from all automated follow-up workflows (W01, W03, W04, W06)
  4. Send high-priority owner SMS alert with full thread link

MASTER TEAM CREATION CONTRACT:
Create an 8-seat AI operator roster for an agency that sells a paid workflow assessment first and a $2,500/month retainer. Permission boundary for every seat: read, research, draft, organize. Human approval is required before any send, publish, spend, delete, or live CRM mutation.

SEAT ASSIGNMENTS:
1. Managing Partner — Owns priorities, handoffs, and the 09:00 daily 8-line brief (blockers and file paths first).
2. Knowledge Librarian — Converts client interviews and 20 inbox threads into versioned kb.yaml + contradiction log.
3. Voice Lead — Writes spoken prompts (V01–V08), transfer briefs, and 30-scenario test scorecards.
4. Conversation Floor — Drafts SMS, IG, and webchat prompts (C01–C08) + 10-thread dry runs for human review.
5. Content Desk — Drafts numbered X, LinkedIn, and IG posts that pitch the paid assessment without hype.
6. Outreach Drafts — Prepares personalized 15-minute process-walkthrough notes for warm contacts.
7. Delivery SOP — Maintains the 90-minute install runbook and snapshot handoff checklist.
8. QA — Tracks the 30-scenario pass/fail table and invented_fact_count. If invented_fact_count > 0, bots stay paused.

PROMPT L01 · ONE-PAGE ASSESSMENT OFFER PACKAGER:
You write one-page offers for an AI agency that sells a paid workflow assessment first, then one verified workflow, then a $2,500/month retainer.
Input:
- Buyer role & industry:
- Repeating process to review:
- Real example they can show:
- What must never be automated:
Output:
1. One-sentence offer
2. Four concrete assessment deliverables
3. Explicit exclusions list
4. Four bullets defining the $2,500/month retainer scope
5. First reversible workflow change
Style: plain, direct sentences. No software list until the process is mapped.

PROMPT L07 · 30-SCENARIO PRE-LAUNCH SCORECARD GENERATOR:
Create 30 test utterances for [AGENT NAME] at [BUSINESS]. Include: happy-path booking, out-of-area ZIP, price hunter, emergency leak/no-heat, "are you a robot", angry caller demanding a manager or lawyer, two speakers talking at once, heavy background noise, and repeat question. For each scenario, output: Caller Utterance | Expected KB Lookup | Expected Action (Reply / Book / Transfer / Pause) | Pass/Fail Rule.

90-MINUTE GO-LIVE INSTALL RUNBOOK (CRS-INSTALL · $150 VALUE):
00–15m · Phase 1 (DNS, Sending Domain & BIMI Pre-Flight — KIT-DNS-BIMI):
  - Verify funnel CNAME -> sites.ludicrous.cloud against authoritative NS (RFC 1035 dig @ns1).
  - Verify LC Email DKIM TXT, SPF include, and DMARC policy (p=quarantine or p=reject; pct=100).
  - Validate 1024x1024 SVG Tiny 1.2 (baseProfile="tiny-ps") BIMI logo + default._bimi TXT record.
15–45m · Phase 2 (Sub-Account Snapshot, Custom Fields & Pipeline Import):
  - Import 5-stage Service Pipeline (New Inquiry -> Qualified -> Booked -> Completed -> Review/Retained).
  - Verify kill-switch custom fields: ai_voice_paused, ai_chat_paused, kb_pair_count, invented_fact_count.
45–75m · Phase 3 (80-FAQ KB Load, Voice V01–V08 & Conversation C01–C08 Wiring):
  - Load 20–80 verified Q&A pairs into kb.yaml; confirm EMPTY escalation fallback.
  - Set Conversation AI to Auto-Suggest mode for the first 50 human-reviewed threads.
75–90m · Phase 4 (30-Scenario Scorecard & Read-Back Audit):
  - Execute 30-scenario test matrix (happy path, out-of-ZIP, price shopper, emergency, angry caller).
  - Confirm invented_fact_count == 0 before enabling live inbound number routing.