The richest no-site data is not on the internet at all: newsagent windows, Post Office and supermarket community boards, laundrette noticeboards, "white van man" cards under wiper blades, postcards in shop windows. Every card is a no-site sole trader with a phone number. Nobody has digitised this. So we pay local people to photograph these boards; AI turns pixels into business name + number → CRM → a postal A-pile with QR-to-claim.
A local person is paid per photo / per verified lead to photograph community boards and card racks around town. Photos land in a simple upload flow → OCR + vision extract fields → dedupe → CRM → each becomes a postal A-pile target with a QR-to-claim.
Unique, un-competed data that lives only on physical boards — the scalable moat.
Local Facebook groups, "earn on your bike round" framing. Adults for v1 (students 18+, retirees, dog-walkers, delivery riders already out) — sidesteps child-employment law. Treat as self-employed micro-taskers.
Give each worker a patch (postcode area) to avoid overlap.
| Model | Pro | Con | Use when |
|---|---|---|---|
| Per usable photo (£0.30–0.50) | Simple, motivating, fast to start | Fraud risk: re-shooting same board, junk photos | v1, with dedup + quality gate |
| Per verified NEW lead (£0.75–1.00) | Pays only for outcomes; self-cleaning | Slower feedback; needs dedupe pipeline live | v2 once OCR trusted |
| Hybrid ⭐ £0.25 base/photo + £0.50/new lead | Aligns effort (shoot lots) with outcome (find new); blunts fraud | — | v1 recommended |
| Step | Assumption | Cost |
|---|---|---|
| Bike round | 40 photos in ~2 hrs | 40 × £0.25 = £10.00 |
| Extraction | ~5 usable leads/photo → 200 raw → 120 unique new | Vision £0.05 + LLM £0.15 ≈ £1 |
| Lead bonus | 120 new verified leads | 120 × £0.50 = £60.00 |
| Postal touch | 1 Stannp A-pile postcard + QR each | 120 × £0.50 = £60.00 |
| Total | 120 targeted no-site, QR-to-claim prospects | ≈ £131 |
| Cost / verified lead | £71 sourcing ÷ 120 | ≈ £0.59 |
| Convert @ 3% claim | ~3.6 paying customers | CAC ≈ £36 |
Against the £300–500/mo rank-and-rent anchor — or even a modest £30–40/mo subscription (£360–480/yr LTV) — acquiring a customer for ~£36 is a ~10× first-year return. We can pay bike-round workers generously (~£35/hr-equivalent) and still print money. The economics work comfortably; claim-rate is the constraint, not sourcing cost.
EXIF on each photo; reject anything outside the assigned patch or re-used.
Detect the same board re-shot; dedupe numbers against CRM so re-submits earn no bonus.
OCR/LLM must return a plausible name + number; low-confidence → 2-second manual review, no auto-pay.
Spot-check 5% of a worker's boards; three junk batches → offboard.
Worker sends photos to a number → n8n / webhook pulls media → Google Vision OCR + GPT-4o parse → dedupe → CRM row → weekly pay tally. Zero app to build — workers already have WhatsApp.
"Take photo → auto-upload with GPS → see your running total." Build only once volume justifies it.
| Point | Rule | Verdict |
|---|---|---|
| Paying minors | Don't at v1. Recruit 18+ only, self-employed micro-taskers. Revisit minors only with council permits + safeguarding. | 🔴 v1: 18+ |
| Capturing phone numbers | Sole trader's mobile = personal data. Basis = legitimate interest + LIA + minimisation (name + number + source + town) + privacy notice + easy erasure. | 🟠 |
| Use of the number | Postal + QR-to-claim only. Prospect self-initiates → keeps UK PECR + DE UWG §7 🟢. | 🟢 postal |
| Worker contract | Simple self-employment micro-task agreement; they warrant photos are of public boards; we are controller, they are our eyes. | 🟢 |
Pick one town. Recruit 2 adult bike-round workers. Pay £0.25/accepted photo + £0.50/new verified lead, weekly via bank transfer. Intake via WhatsApp Business. Pipeline: WhatsApp → n8n → Google Vision OCR + GPT-4o parse → dedupe → CRM. Each lead → one Stannp A-pile postcard with QR-to-claim. Track claim-rate as the north-star metric. Budget for a 120-lead pilot ≈ £131.
outreach/lead-sourcing-strategy.md §3 · the un-competed moat · postal + QR-to-claim