# WORTH: Go-To-Market Plan

## TL;DR

- **What WORTH is:** A two-sided marketplace plus personal agent. SUPPLY = individuals who own their data and verified expertise in a sovereign vault, with an agent that matches them to DEMAND = buyers (AI labs, pharma/CROs, research, brands) posting structured requests for data, expert judgment (RLHF), and services. WORTH brokers the match, enforces per-use revocable consent with cryptographic provenance, and takes a 10-20% cut.
- **Why it wins:** Per-use revocable, GDPR/clinical-grade consent (not pooled token-governance), real fiat payouts (not pennies in a volatile token), a verified-expert premium, mainstream non-crypto UX, and an agent that tells suppliers "here is how to earn quickly."
- **The wedge:** Verified-expert RLHF for AI labs FIRST (fast-moving buyers, demand on fire, "real nurse/engineer/lawyer" is the pitch, lightly regulated). Clinical/health data unions (pharma, deeper money, slower/regulated) SECOND.
- **Cold-start solution:** DEMAND-FIRST. Land one buyer with one budgeted request, concierge-recruit ~100 verified suppliers to fill it, fulfill, publish a case study, then automate matching and open the feed.
- **Money:** $100M GMV at a 20% take = $20M revenue, plus buyer-side SaaS/compliance fees and supplier verification fees.

---

## 1. Core Insight and Why Now

**Insight:** The people who generate the most valuable training signal (expert humans and data owners) capture almost none of the value. Scale AI and Surge sit in the middle, own the relationship, and keep the margin. WORTH flips ownership to the supplier and pays real money, which is both a better deal for suppliers and a cleaner provenance story for buyers.

**Two tailwinds, both peaking now:**

| Tailwind | Evidence | What it forces |
|---|---|---|
| The data wall | Epoch: high-quality public text exhausted 2026-2032. AI-training-data market ~$4.8B (2025) to ~$22.6B (2034). | Labs must buy fresh, expert, consented data. Scraping is tapped out. |
| The provenance/legal wall | NYT v OpenAI and the wave of copyright suits. | Labs need CONSENTED, provenance-clean data to de-risk training. Pooled or scraped data is now a liability. |

**Why WORTH and not Vana:** Vana proves demand-side appetite (1.2M users) but fails on payout and consent.

| | Vana | WORTH |
|---|---|---|
| Payout | Pennies in a token down ~96% | Real fiat, set per request |
| Consent | Pooled, token-governance | Per-use, revocable, GDPR/clinical-grade |
| Agent | None | Personal agent advises "how to earn quickly" |
| UX | Crypto-clunky | Mainstream, non-crypto |
| Regulated data | No | Yes (clinical phase) |
| Verified expertise | No | Yes, the premium tier |

---

## 2. The Two-Sided Cold-Start Solution: Demand-First + Concierge Supply

Marketplaces die when both sides wait for the other. WORTH does not open a feed and hope. It manually closes the first loop, then automates.

**The exact playbook (one loop, repeated):**

1. **Land ONE buyer** with ONE budgeted request. A signed PO for a specific dataset/RLHF batch with a dollar figure attached. No buyer, no recruiting.
2. **Concierge-recruit ~100 verified suppliers** to fill exactly that request. Hand-sourced, hand-verified, hand-onboarded. Founder-led, not self-serve.
3. **Fulfill** the request with white-glove QA. WORTH staff (or a thin tool layer) sit in the middle and guarantee quality.
4. **Pay suppliers real money, fast.** The payout receipt is the marketing asset.
5. **Case-study it:** publish supplier earnings and buyer outcome (with consent). This is the recruiting magnet and the next buyer's proof.
6. **Then automate** matching and **open the feed.** Replace concierge steps with the agent and the request console once the loop is proven.

**Principle:** Be the human in the middle until the loop is profitable and repeatable. Software replaces the founders, not the demand.

---

## 3. First Wedge: Verified-Expert RLHF

**Why this first:** Buyers move fast and are budget-loaded right now. "A real, credential-verified nurse/engineer/lawyer rated this answer" is the entire pitch and it is barely regulated compared to clinical data. This is the path of least resistance into a market on fire.

### The first buyer (profile + how to reach)

| Attribute | Target |
|---|---|
| Type | Mid-size frontier-adjacent AI lab or a well-funded vertical-AI startup (legal AI, medical AI, coding AI) |
| Pain | Their RLHF/eval quality is capped by generic crowd labelers. They need domain experts and a provenance trail. |
| Budget owner | Head of Data / Head of Post-Training / RLHF lead, not procurement |
| Why they say yes | Better signal than Scale's crowd, plus a consent/provenance receipt that de-risks their training set |

**How to reach them:**
- Warm intros through the founder/advisor network into post-training and data teams.
- Direct outbound to RLHF/data-quality leads (these roles are public on LinkedIn and in job posts).
- Show up where they hire labelers and complain about quality: ML eval communities, post-training Discords/Slacks.
- Lead with the demo (see section 8) and a one-line offer: "100 credential-verified [domain] experts, consented and provenance-stamped, batch delivered in 2 weeks."

### The first 100 suppliers (which community + how)

- **Target one tight expert community first.** Best opening pick: **licensed nurses / clinicians** (huge population, credential-verifiable via license number, underpaid, already moonlight, and it warms up the later clinical wedge). Strong alternates: **software engineers** (for coding-model RLHF) or **lawyers** (for legal-AI RLHF). Match the community to the first buyer's domain.
- **How to recruit:**
  - Founder-led outreach in the profession's existing watering holes (nursing subreddits and Facebook groups, dev forums, bar-association adjacent channels).
  - The hook: "Get paid for your expertise. Verify your license, rate AI answers, earn real money. We tell you exactly how to earn the fastest."
  - Verification at intake: license/credential check is the moat and the premium justification.
- **Why 100:** Enough to fill one batch with quality redundancy, small enough to onboard by hand and keep QA tight.

### The first request shape

- Structured RLHF/eval batch: a fixed set of prompt-response pairs in the expert's domain.
- Task: rank/rate/correct responses, write gold reference answers, or flag domain errors.
- Each submission carries: verified-credential tag, per-use revocable consent record, cryptographic provenance stamp.
- Deliverable to buyer: a clean labeled dataset plus a compliance/provenance manifest.

---

## 4. Supply Acquisition: Gamified Hook + Build-in-Public + Viral Mechanics

The recruiting engine that scales past the first 100.

**The hook: "Find out what you're worth."**
- Free **Worth Score** at signup: estimates earning potential from credentials, data, and scarcity of expertise. It is the lead magnet.
- **Bounties:** live, dollar-denominated requests visible as "earn $X for Y." Concrete, not abstract.
- The **agent** advises: "Verify your nursing license to unlock $Z bounties" and "complete this batch first, it pays fastest."

**Build-in-public:**
- Publicly post real supplier payouts, total GMV, and "WORTH paid out $X this week."
- Founder narrative on the data-ownership and fair-pay thesis. Honest contrast with Vana's broken token.

**Viral mechanics:**

| Mechanic | How it spreads |
|---|---|
| Worth-Score sharing | Shareable score card ("My expertise is worth $X/hr on WORTH") seeds curiosity |
| Referral | Pay both sides for verified-expert referrals (experts know other experts) |
| Leaderboards | Top earners by domain, opt-in, drives competitive participation |
| Community embedding | Become the default "get paid for your expertise" channel inside one profession before expanding |

**Strategy:** Saturate ONE expert community to density before opening the next. Density makes recruiting word-of-mouth and makes the next buyer's batch easy to fill.

---

## 5. Demand Acquisition: Selling Labs and Pharma on Consented + Verified + Provenance Data

**The wedge pitch is lawsuit-safety.** Post-NYT-v-OpenAI, every training set is a potential liability. WORTH sells the receipt.

**Sales narrative (RLHF/labs):**
- "Higher-signal data: credential-verified experts beat generic crowd labelers."
- "Lower legal risk: every record is consented, revocable, and provenance-stamped. You can prove where it came from."
- "Disintermediate the markup: you pay experts more, and still pay less than Scale's margin."

**Sales narrative (pharma/CRO, phase two):**
- "GDPR/clinical-grade per-use consent that Vana's pooled model structurally cannot offer."
- "Auditable provenance for regulated submissions."

**The buyer-side product:**

| Component | What it does |
|---|---|
| Request Console | Post structured requests: spec, budget, required credentials, volume, deadline |
| Matching | Agent-driven supplier matching against credential and data criteria |
| Compliance Dashboard | Per-record consent status, revocation log, provenance manifest, exportable audit trail |
| Delivery | Clean dataset plus the compliance manifest as a single deliverable |

The Compliance Dashboard is the actual product labs cannot get from Scale or Vana. It is what they show their legal team.

---

## 6. Money Model

**Revenue lines:**

| Line | Mechanic | Notes |
|---|---|---|
| Take rate | 10-20% of each transaction | Core line. 20% on high-value verified-expert work, lower on commodity. |
| Buyer SaaS/compliance fee | Subscription for Request Console + Compliance Dashboard + audit exports | Recurring, sticky, sold to legal/data orgs. |
| Verification fee | Charged to suppliers (or buyer-subsidized) to verify credentials | Funds the moat, signals quality. |

**Illustrative GMV to revenue:**

| GMV | Take @ 20% | Plus SaaS/compliance (illustrative) | Approx. revenue |
|---|---|---|---|
| $10M | $2.0M | ~$0.5M | ~$2.5M |
| $100M | $20.0M | ~$3-5M | ~$23-25M |
| $500M | $100.0M | ~$15M+ | ~$115M+ |

**Unit logic:** Verified-expert RLHF carries high per-task value, so even modest task volume produces meaningful GMV. The compliance fee converts a transactional marketplace into a recurring-revenue platform.

---

## 7. Sequencing, Phases, and Risks

### Phases (rough timing)

| Phase | Timing | Goal | Done when |
|---|---|---|---|
| 0: Demo + first buyer | Months 0-2 | Land one budgeted buyer with the mocked app | Signed PO for one request |
| 1: Concierge loop | Months 2-4 | Recruit ~100 verified suppliers, fulfill, pay, case-study | One loop closed, paid, documented |
| 2: Densify supply + repeat buyers | Months 4-8 | Saturate one expert community, land 3-5 more buyers | Repeatable concierge loop, viral recruiting live |
| 3: Automate + open feed | Months 8-12 | Replace concierge with agent + Request Console | Self-serve matching live, take rate flowing |
| 4: Clinical wedge | Months 12-18+ | Health data unions, pharma/CRO buyers, GDPR/clinical consent | First regulated data union transacting |

### Top 3 risks + mitigations

| Risk | Mitigation |
|---|---|
| Cold-start (neither side shows up) | Demand-first. Never recruit supply without a budgeted buyer. Concierge the first loop end-to-end. |
| Incumbent inertia / Scale defends | Do not fight on commodity labeling. Win on verified-expert quality + the provenance receipt Scale cannot issue without re-architecting consent. Disintermediate by paying suppliers more. |
| QA at scale | Keep humans-in-the-loop through phase 2, redundant ratings, credential gating, and reputation scoring before automating matching. |
| (Bonus) Regulatory, clinical phase | Sequence clinical SECOND, after per-use revocable consent + provenance are proven on lighter RLHF data. Build legal review into the data-union structure. |

---

## 8. The Demo's Role

- The mocked WORTH app is **the sales tool, not just a prototype.** It does two jobs:
  1. **Land the first buyer:** walk a lab's data lead through the Request Console + Compliance Dashboard so they can see the consented, provenance-stamped deliverable before any data exists.
  2. **Recruit founding supply:** show experts the Worth Score, the bounties, and the agent saying "here is how to earn quickly," so they verify and join.
- The demo lets WORTH sell both sides of an empty marketplace by making the closed loop tangible. It is what converts the demand-first playbook from a pitch into a signed PO and 100 onboarded suppliers.