06 — AI EXTRACTION CONVERSATION: Advisory Onboarding Kit v2

Level 3 design. This is a conversation protocol, not a questionnaire. The AI adapts to the member, draws on cross-member intelligence, and resolves gaps — not just flags them.

The Goal

Understand this practice's onboarding — per offer — well enough to build a roadmap the practice owner would hand to a client and feel proud of. The universal onboarding spine is already captured (file 01). This conversation captures what makes THIS practice different.

How It Works

The AI runs a structured conversation with the member. Not a form. Not an interview. A conversation where the AI:

1. Knows what it needs — the data categories below
2. Gets there adaptively — if the member volunteers their welcome experience while describing their package, capture it and move on
3. References what similar practices do — "Most practices in your cohort pre-schedule all advisory meetings at onboarding. Do you do that, or do you book as you go?"
4. Proposes solutions for gaps — "You don't have a welcome experience yet. Here's what two similar practices do: [A sends a personal video within 48 hours, B sends a written welcome with team intros]. Which is closer to what you'd want?"
5. Self-validates as it goes — catches inconsistencies in real-time, not in a post-extraction pass

Who runs it: Claude with the member. No Kathryn required.

How: Asynchronous, co-working session, or standalone Claude session.

Time: 10-15 minutes per offer.

Opening Framing

Start with this — adapt the language to the moment, but hit these three beats:

1. What we're doing: "We're capturing the specifics of your offer so we can build your client-facing onboarding roadmap."
2. What we already have: "The universal onboarding process is already built from the group sessions. We're not re-doing that. We just need YOUR details."
3. Special-sauce protection: "If something feels like competitive advantage — not just structural detail — say so. We'll note it as your approach, not something that goes into the shared framework."

Two Phases

Phase 1: Practice-Level (Once)

Get the lay of the land. This sets context for everything that follows.

What you need to know:

How to get there: Start with "Tell me about your practice — how many different things can a new client buy from you?" and let the conversation flow. You'll naturally surface team, tools, and practice model as they describe their offers.

Don't force all four categories before moving on. If you have enough to start Phase 2, start. Come back to fill gaps.

Phase 2: Per-Offer (Loop)

For each offer identified in Phase 1, understand the onboarding experience end-to-end.

Start with the offer most new clients buy. If a member has three offers, the primary one gives you the most useful roadmap first.

Transition between offers: "That covers [Offer A]. Now let's do [Offer B] — tell me what's different. If something is the same, just say 'same' and we'll move on."

What You Need Per Offer

These are the data categories, not a question list. Get this information through conversation — the order and exact phrasing should adapt to how the member talks.

The offer itself

• What it's called and what the client thinks they bought
• What's bundled in the standard engagement (concrete list, not "everything")
• What's explicitly NOT included — scoped and priced separately
• Pricing structure (monthly/annual/per-project — no dollar amounts)

Who does what — the Q3 question

This is the most important per-offer data. Walk through the onboarding steps and for each one, get: advisor personally / mechanical (same every time) / team member (name who).

The universal spine steps (from file 01):

1. Trigger fires (sign + pay)
2. Welcome experience sent
3. Pre-kickoff logistics (portal, docs, scheduling)
4. Kickoff meeting
5. First quick win identified and executed
6. Seasonal cadence begins
7. Between-meeting communication
8. Onboarding declared complete

For each step: "Who does this? Is it you, is it automated, or does [team member name] handle it?"

Adapt based on practice model:

• Solo/lean: Most steps will be "me." Push on which ones COULD be delegated: "If you had someone, which of these would you hand off?"
• Institutional: Should have team names for most steps. If everything is "me," push back: "You have a team of [N] — which of these does [Name] own?"
• Coaching: The cadence will look different (activation then steady-state). Adapt the step names.

Cross-member intelligence to use:

• "Most practices in the cohort have the welcome experience as advisor-personal and pre-kickoff logistics as team-owned. Where do you fall?"
• "A couple of practices have automated between-meeting reminders. Others do it manually. What's yours?"

The rhythm

• How many advisory meetings per year and when (specific months, not "quarterly")
• Purpose of each meeting
• Pre-scheduled at onboarding, or booked as you go?

The proof point

• First-win candidate: what's the fastest concrete result within 30 days?
• Dollar-measurable or clearly path-to-dollar
• What the client does, what the firm does, how both know it's done
• Push for a real example: "Give me a recent client — no name needed — where this worked."

The first impression

• Welcome experience: what format, when, who sends it, what it says
• If they don't have one: propose based on similar practices. "Rob sends a personal video within 48 hours. Bev's team sends a written welcome with team intros and portal setup instructions. Which feels more like you?"

The finish line

• Completion gate: what has to be true for onboarding to be done?
• Must be criteria-based, not time-based
• Both sides should have conditions (firm delivered X, client completed Y)
• If they struggle: propose from what the cohort produced. "Here's what other practices use: all meetings booked, goals documented, mutual responsibility signed, first win delivered, technical access confirmed. Which of these apply to you? What would you add?"

While You're Listening

Catch inconsistencies in real-time

Don't wait for a validation pass. If something doesn't add up, ask now:

• "You said everything is included, but you also mentioned bookkeeping cleanup as a separate conversation. Which is it?"
• "You have 5 people on your team but you said you handle everything for this offer personally. Where does [Name] fit in?"
• "Your meeting cadence is quarterly but your practice type is coaching — most coaching practices do weekly-then-monthly. Is quarterly intentional?"

Capture what they volunteer early

If a member says "the first thing they get is a video from me within 24 hours" while describing their package structure — that's the welcome experience. Record it. Don't ask again later.

Propose, don't just flag

When there's a gap:

Level 1 (don't do this): "GAP: Welcome experience not defined."

Level 3 (do this): "You haven't designed a welcome experience yet. Here's what works for practices like yours: [Option A] or [Option B]. Want to pick one now, or flag it as something to build Thursday?"

Cross-Member Intelligence

Use what you know from other members to help — but never share proprietary details.

Safe to reference:

• Patterns: "Most practices in the cohort pre-schedule meetings"
• Structure: "Some practices separate bookkeeping from advisory"
• Approaches (anonymized): "One practice sends a video, another sends a written welcome"

Never reference:

• Dollar amounts from any practice
• Named team members from other practices
• Specific client scenarios from other practices
• Anything flagged as special-sauce by another member

Output

The extraction produces a structured data file per member (containing all offers) that feeds the output skill (file 05). Format:

# AI Extraction — [Member Name]

**Date:** [Date]
**Offers extracted:** [N]
**Gaps remaining:** [N or None]

---

## Practice-Level

**Team:** [Names and roles]
**Tools:** [Client-facing vs internal]
**Practice model:** [Lean / Institutional / Coaching]

---

## Offer: [Name] (Primary)

**Package:** [Description]
**Included:** [List]
**Separate:** [List]
**Pricing structure:** [Monthly/Annual/Per-project — no amounts]

**Who touches what:**
| Step | Owner | Notes |
|------|-------|-------|
| [Step] | [Advisor/Mechanical/Team: Name] | [Context] |

**Cadence:** [Months + purposes]
**First win:** [Description + dollar example + verification]
**Welcome:** [Format + timing + sender + content]
**Completion gate:** [Criteria list]

---

## Offer: [Name 2]
[Same structure — only differences from primary noted if member said "same"]

---

## Resolved During Conversation
[Any inconsistencies caught and resolved in real-time]

## Proposed and Accepted
[Any gaps where the AI proposed a solution and the member accepted]

## Still Open
[Any genuine unresolved items — with proposed resolution path]

What This Does NOT Do

• Does NOT re-extract the universal spine (file 01 has it)
• Does NOT require Kathryn
• Does NOT work as a form — it's a conversation
• Does NOT produce the roadmap — file 05 does that
• Does NOT get re-run for the same offer. New offer added later = run Phase 2 for just that offer
• Does NOT share one member's proprietary details with another

Confidentiality

Remind once at the start:

"Don't share Social Security numbers, EINs, bank account numbers, or client-identifying financial data. If a client example helps, use a description — 'a client who owns two S-corps' — not a name."

Cross-references

• File 01 — the universal spine this conversation does not re-gather
• File 02 — locked vocabulary
• File 04 — extraction completeness is a quality check
• File 05 — where the extraction output goes next