Expert Clone JV Boarding

Overview

How the System Works

We build AI that thinks like an expert — not just mimics their words. Our moat is the decision architecture: we extract how an expert chooses their words, not just what words they choose. This is a joint venture, not a service. We build together, we win together.

Ecosystem

Brand Architecture

Three brands serve different purposes in the pipeline. Understanding which brand the expert and their audience interact with at each stage is critical.

Phase 1

Discovery & Qualification

The universe of potential experts is large. Our job is to filter ruthlessly. Maybe 10% of the people Derek talks to are worth considering, and those 10% sort into different buckets based on timing, audience alignment, and revenue potential.

Expert Buckets

Qualification Questions

These questions need to be answerable with low cognitive load — multiple choice where possible, conversational flow preferred over a 15-question form. Delivered via chatbot, voice agent, live call with Derek, or voice-recorded intake.

1. Revenue range — Annual income from expertise (MCQ brackets)
2. Audience size — Platform reach across channels (MCQ brackets)
3. Current offer stack — What they sell today (MCQ categories)
4. Client lifetime value — Revenue per client relationship
5. Framework or methodology — Do they have a named system people pay for? (Y/N + describe)
6. Current affiliate program — Do they have one, and what does it look like?
7. Primary constraint — Why JV? What's the real motivation? (MCQ: scale reach / save time / new product / monetize differently / other)
8. Client pain points — What does their audience struggle with most? (MCQ categories)
9. Content volume — How much existing content do they have? (podcasts, videos, courses, books)
10. AI readiness — Have they used AI tools? What's their comfort level?

Intake Channels

Multiple paths to the same scoring system. Don't require one method — meet the expert where they are.

• Live call with Derek — He asks the questions conversationally, records via Tactiq/Fireflies, transcript gets extracted
• Chat agent — AI-driven conversational intake on athio.ai that asks questions naturally. If they answer 3 in one response, it adapts.
• Voice agent — Phone-style AI interview. The wow factor alone sells the partnership. (Roadmap)
• Voice memo from Derek — After a call, Derek records a brain dump: "Just talked to John, he's got X, might be a hot lead for bucket Y." Transcript extracted and scored.
• Form fallback — Email them a lightweight version if nothing else works

Phase 2

Scoring & Approval

All roads lead to a standardized score. Whether the data came from a live call, a chatbot session, or Derek's voice memo — it all feeds the same rubric.

AI Scoring Rubric

The LLM takes the open-ended answers and maps them to a standardized scoring system. Categories scored 1-5:

1. Audience alignment — Does their audience overlap with our current offering and 1-2 iteration roadmap?
2. Revenue signal — Evidence of real paying customers, not aspirational numbers
3. Framework maturity — Do they have a named, teachable system, or just general expertise?
4. Content depth — Enough raw material to build a quality clone (podcasts, courses, books, calls)
5. Partnership temperament — Win-win mindset vs. transactional. Are they rowing the same direction?
6. Time-to-value — How quickly can we get their audience using the product?
7. Constraint clarity — Do they know why they want this? Clear motivation = better partner

Decision Gate

After scoring, three outcomes:

• Approved for JV — Moves to Phase 3 boarding. Gets into the queue based on capacity (not all at once).
• Redirected to pay-to-play — Warm handoff to MasteryOS brand. Different landing page, different expectations. They can bid for a spot — leave it open-ended, not a sticker price.
• Waitlisted — Captured in CRM, revisited quarterly. Not wasted effort if the contact is preserved.

Phase 3

Expert Onboarding

Once approved, we need to extract everything required to build their clone and set up their platform. This is the "boarding" — and it has hard dependencies. Miss a step here and everything downstream breaks.

The Boarding Checklist

Phase 4

Clone Architecture & Build

This is the secret sauce. We don't pattern-match words. We extract decision heuristics — how the expert chooses their words, approaches problems, and structures their thinking. Everyone else gives you an actor reading a script. We give you the writer's room.

The Three-Track Extraction

Raw expert documents are processed through three parallel pipelines:

Validation Process

Before going live, the clone must pass the tequila test.

1. Extract test cases — Segment some expert documents as holdouts. Generate questions from them as if a real user was asking.
2. Run the clone — Feed those questions to the clone. Capture its responses.
3. Compare against ground truth — Does the clone's response follow the same thinking path as the real expert? Not word-for-word — same journey, same landing point, same approach. "If we gave our friend tequila, would the clone react the same way?"
4. Score across rubrics — Logic match, voice match, framework adherence, model collapse resistance. Did it stay in character? Did thinking patterns hold? Did it drift to generic ChatGPT responses?
5. Expert review — Put the expert in front of it. "Does this think like you?" Not perfect — 80% is the target. If it handles 68% of questions at a high level and 80% at a confident level, the remaining 20% is what makes the real expert valuable.

Phase 5

Platform Deployment

Multi-tenant architecture. Think apartments, not houses. Each expert gets a unit in the building. Costs stay low during alpha. If someone goes dormant, it doesn't eat capital.

Multi-Tenant Model

Previously, each expert was a standalone deployment (single-tenant) — their own codebase, hosting, database. Expensive to maintain when idle. Now all experts share infrastructure with isolated data. First tenant hits the most bugs. Second tenant finds the multi-tenant bugs. By tenant three, it should be smooth drops.

What Gets Deployed

Alpha Testing Protocol

Before wide launch, 5 users from the expert's existing audience. Not strangers — people who already know the expert and can evaluate whether "this thinks like [Expert]." Key question for them: not "is this useful?" but "is this them?"

Phase 6

Launch & Iteration

Launch is not the finish line. It's the beginning of the feedback loop. We ship fast, learn what breaks, and iterate. 48-hour shipping rule applies.

The Real-Time Proposal System

The killer demo that closes JVs. Here's what it looks like in practice:

1. Derek is on a discovery call with a potential expert. Our transcription system is running.
2. Halfway through the call, Derek hits a button. Selects a template.
3. The system imports the transcript to that point, applies it to the template, and generates an interactive proposal page — customized to everything they just discussed.
4. By the end of the call, Derek sends a live URL: athio.com/[expert-name]
5. The page includes a light version of their chatbot clone. The expert can try talking to "themselves."
6. Their mind is blown. The only obvious answer is yes.

Iteration Cadence

• 48-hour shipping — Every cycle must produce a deliverable. Not "worked on it" — shipped it. Live. This prevents scope creep, forces prioritization, and builds momentum.
• Expert feedback loops — After alpha users engage, collect: "Does this sound like me?" "Where did it drift?" "What frameworks are missing?" Feed back into Track B/C extractions.
• Feature roadmap — Ideas captured during boarding go on the roadmap, not into the build. "Needed it 10 times" rule: don't build features until the demand is proven. Gamified onboarding, video clones, advanced voice agents — all roadmap items, not launch blockers.
• Tenant scaling — After expert #3, deployment should be near-frictionless. Goal: eventually automate the full pipeline from content folder → enriched RAG → live clone via agentic workflows.

Reference

Key Dependencies Map

For every step, ask: what's the gate? What blocks this from proceeding? What's the second-order effect? Do we have it ready, have a template, or need to create it?

Operating System

Principles That Run This

1. 48-Hour Ship Rule — Pick the one thing. Deliver it live within 48 hours. Document the 2 things you wanted to add but couldn't as roadmap items. The constraint forces clarity by removal.
2. Dependencies Before Building — Map every step's gates. What blocks this? What's the second-order effect? Do we have it or need to create it? If you skip this, someone will have to do those little things later, and that's where friction lives.
3. 0-to-1 Over Perfection — Get the dirty version live. 80% done and operational beats 100% planned and theoretical. Everything costs twice as much and takes three times as long. But if you keep on track, you get there.
4. Don't Do It One-Off — Whatever you do, document the flow so it can be recreated. Process → checklist → habit → throw the checklist away. That's how we systemize and scale without energy drain.
5. ICP Discipline — Build for who pays today, not who might pay tomorrow. Changes to the expert's target audience happen after validation, not during build. Lock the ICP early.
6. Anti-Sell Posture — We're selective. We reject most applicants. If the clone scored well against the expert and they still don't want it, we show them the door and point them to Delphi. No hard feelings. The right people recognize value through action, not persuasion.