# Worth, demand-side playbook The bottleneck is not the product, it is the first paying buyer. This is the buyer-first pitch, the target list, and the outreach plan to land one signed pilot. Everything in the supplier-facing GTM is downstream of this. The reframe in one line: we do not sell "sovereignty" to buyers. We sell **compliant, verified-expert and regulated-cohort supply, on tap, with provenance the buyer can defend in an audit.** Sovereignty is how we get the supply and the moat. Compliance is what the buyer pays for. --- ## 1. The buyer one-pager ### Headline **Verified-expert human data, on tap, with provenance you can defend.** Nurses, engineers, clinicians, and patient cohorts, matched to your request in days, every contribution consented per-use and cryptographically traceable. ### The buyer's actual problem (lead with theirs, not ours) | Buyer pain | Worth answer | |---|---| | RLHF quality is collapsing: bot-polluted panels, fake credentials, "experts" who are not | Every contributor is verification-gated (license, employer, tenure). You buy *verified* nurse/engineer/clinician input, not a Mechanical-Turk guess. | | Provenance is a black box: you cannot prove where a training example came from | Every datum carries a signed, per-use consent record. Clean chain-of-consent for the EU AI Act, FDA RWE submissions, and your own legal team. | | Sourcing rare experts is slow and manual: recruiters, spreadsheets, weeks | An agent matches your request to qualified, consented supply and returns it in days. You post a bounty, we fill it. | | Regulated data (clinical, patient cohorts) is legally radioactive to source directly | We broker consented patient-cohort data unions. The consent and de-identification are the product. You get usable RWE without the liability. | ### What you can buy (the SKUs) - **Verified-expert RLHF / preference labels**: ranked responses, evaluations, "what would a senior X say", from credential-verified professionals (nurses, MDs, staff engineers, attorneys, native-language experts). - **Expert reasoning traces**: annotated decision/debugging traces from senior practitioners, the highest-demand training signal for reasoning models. - **Consented data licenses**: scoped, time-boxed, revocable access to real longitudinal data (sleep/HRV, CGM/glucose, code repos, clinical outcomes), anonymized to spec. - **Regulated cohort RWE**: patient-data-union licenses (e.g. T1D, oncology) with consent + de-ID + ethics framing handled. ### Why per-use revocable consent is *your* feature, not a risk The instinct is "revocable = I lose the data after I trained on it." It does not work that way and it protects you: - Revocation is forward-only: it ends *future* access. It never claws back a license you have already paid for and used within scope. Your trained model is safe. - What you gain: a defensible, auditable consent trail per example. When a regulator, a journalist, or opposing counsel asks "where did this come from and did the person agree", you have a cryptographic answer. That is cheaper than a legal indemnity and it is the thing pooled-token or scraped-data competitors cannot produce. - Net: you trade a small amount of "perpetual ownership theater" for real legal defensibility. For anyone training on health, code, or person-level data under the EU AI Act, that is the better trade. ### Proof it is real Live, populated demo (4 personas operating a filled marketplace, real bounty/consent/payout flows): **worth-demo.pages.dev**. This is mocked to show mechanics; the supply network is what we are standing up with founding contributors now. ### The ask A paid pilot: one batch of verified-expert supply against a real request of yours. You define the spec and acceptance bar, we deliver consented + traceable supply, you pay on delivery. Small, fast, designed to prove quality and provenance, not to lock you in. --- ## 2. The wedge, and the sequence (do not skip) Lead with the ONE wedge where demand already pays today, then earn the harder, bigger ones. | Wave | Wedge | Why now | Buyer | |---|---|---|---| | **0 (now)** | Verified-expert RLHF / preference labels | Labs pay for this *today* (Mercor/Handshake/Surge prove $50-150/hr). Supply is high-value + motivated. Low regulatory load. | Foundation + coding-model labs | | **1** | Expert reasoning traces | Same buyers, higher price, same rails. Hottest reasoning-model signal. | Coding / reasoning labs | | **2** | Consented longitudinal data licenses | Reuses the consent rail; sports-science + sleep buyers are reachable + lighter than pharma. | Research labs, QS / sports science | | **3 (later)** | Regulated cohort RWE | Real money but an 18-month regulated, ethics-board, legal slog. Do it once you have revenue + a buyer pulling you in. | Pharma CROs | Do not start at Wave 3 even though the money is biggest there. Start at Wave 0, where a cold email can get a paid pilot in weeks. --- ## 3. Target list (who to email, in priority order) Two buckets. Bucket A = data/expert-supply intermediaries and labs who already buy this and feel the quality/provenance pain. Bucket B = regulated-data buyers for later waves. Verify current contacts before sending; titles below are the role to find on LinkedIn. ### Bucket A, verified-expert RLHF / reasoning (Wave 0-1, email first) | Target type | Who to reach | The hook for them | |---|---|---| | Frontier model labs (data/post-training teams) | Head of Data / Post-training, RLHF lead | "Verification-gated experts + per-example provenance for AI-Act defensibility." | | Coding / reasoning-model labs | Data lead, eng-evals lead | "Senior-engineer reasoning traces, credential-verified, traceable." | | Expert-data marketplaces hitting a supply or trust ceiling (potential buyer *or* channel partner) | Founder / Head of Supply | "We bring consented, credential-verified niche supply (clinicians, rare languages) you struggle to source, with provenance." | | AI eval / red-team firms needing verified domain experts | Head of Expert Network | "On-tap verified clinicians/engineers for domain evals." | | Specialist annotation buyers (medical/legal AI startups) | Founder / Head of ML | "Verified-clinician labels instead of generalist annotators." | ### Bucket B, regulated / cohort data (Wave 2-3, warm only) | Target type | Who to reach | The hook | |---|---|---| | Pharma CROs running RWE | RWE / real-world-data lead | "Consented patient-cohort data unions, de-ID + ethics handled." | | Digital-health + sports-science research groups | Principal investigator, data-partnerships | "Consented longitudinal CGM/sleep/training data, licensable per-study." | | Patient-advocacy orgs / disease foundations (supply + credibility partner) | Partnerships lead | "Your members own and earn from their data, you get research leverage." | Action: I cannot pull live named contacts without a research pass. Say the word and I will run a research workflow to produce the actual named-target list (companies, named decision-makers, recent signals that they are buying expert data) for Bucket A. --- ## 4. The cold outreach (buyer-first, short) ### Cold email template (Wave 0, to a labs data lead) > Subject: verified-expert RLHF with provenance you can audit > > Hi [name], your post-training quality lives or dies on whether your "experts" are actually experts, and increasingly on whether you can prove where each example came from. > > We run a network of credential-verified professionals (nurses, staff engineers, clinicians, native-language experts) who supply RLHF labels, evaluations, and reasoning traces. Two things make it different: every contributor is verification-gated, and every contribution carries a signed, per-use consent record, so you get a clean, auditable provenance trail for the EU AI Act and your legal team. > > Worth a 20-minute pilot? You give us one real spec + acceptance bar, we deliver a consented + traceable batch, you pay on delivery. Live demo of the mechanics: worth-demo.pages.dev > > [name] ### The call (what closes it) 1. Open on *their* pain (quality + provenance), not our sovereignty story. 2. Show the demo: post a bounty → matched to a verified expert → per-use consent grant → delivery → signed receipt. The receipt is the money shot: "this is the audit trail your legal team wants." 3. Scope a tiny paid pilot live: their spec, their acceptance bar, fixed small batch, pay on delivery. 4. The only goal of call one is a yes to the pilot. Pricing flexibility > deal size here. The first paid pilot is the asset. --- ## 5. The money model (how we make a fuckton, briefly) - **Take rate** on every match (target 10-20% of GMV; experts keep 80-90%, which is also the supplier-acquisition pitch). - **Verification fees** + premium-tier placement (the badge system). - **Buyer-side compliance SaaS**: the provenance/audit layer as a recurring product, the part incumbents do not have and the part regulated buyers will pay to keep. - GMV → revenue: one lab spending mid-five-figures/month on verified supply at a 15% take is a real first contract. Stack labs, then add the compliance subscription, then unlock the pharma wave where single licenses are five to six figures. --- ## 6. What I need from you to move 1. **Go/no-go on the reframe** (buyer = compliance + verified supply; sovereignty = moat, not pitch). If yes, I align the landing + supplier GTM to it. 2. **Green light to run the named-target research pass** for Bucket A (companies + named decision-makers + buying signals). 3. Your call on whether we approach the expert-data marketplaces as **buyers or channel partners** (I lean: partners for niche supply, buyers for provenance, test both). The single next datapoint that matters: one signed LOI or paid pilot, sourced with this demo. Nothing else de-risks the idea.