Street View OCR — the automated sign-writer v2 experiment

The person who painted the phone number onto every tradesman's van holds a de-facto contact list of our exact target. The automated equivalent: Google Street View, shopfront signage and Maps Business entries frequently show the painted number on the storefront or van — and Vision OCR can read it. This is the lowest-priority, sharpest-cornered channel: it works, but it must be built carefully. Official APIs and manual capture only — never bots.

Maps APIStreet View Static + Places
Vision OCRreads the painted number
Postal-onlyoutput — QR-to-claim
🟠v2, lowest priority

The idea — Google Maps Street View API

What we discussed

Use the official Google Maps Platform APIs — the Street View Static API and Places API — to pull imagery of shopfronts and vans in a target patch. Many local trades paint their mobile number straight onto the signage. Run each image through Vision OCR to lift business name + number, then feed it into the same pipeline as every other channel.

This is the machine version of the sign-writer channel — instead of a human referral partner, the storefront itself becomes the data source.

Why it's attractive

No worker to recruit, no board to physically visit — a desk-based sweep of a postcode. Once built, it scales across towns for API cost only.

It's the automated twin of the Noticeboard mechanical-turk model — same output (name + number → postal), different capture.

The pipeline

1 Define patch

Pick a postcode / high-street run. Query Places API for businesses; get lat,lng for each frontage.

2 Pull imagery

Street View Static API — official, paid, ToS-clean. Grab the storefront view at each point.

3 OCR

Google Vision reads painted signage → candidate name + number strings.

4 Parse + validate

LLM extracts a plausible business name + UK/DE mobile; low-confidence → manual review, never auto-accept.

5 Dedupe → CRM

Hash against existing CRM; store only name + number + source + town (minimisation).

6 Postal + QR

Each new lead → one Stannp A-pile postcard with QR-to-claim. The prospect self-initiates. 🟢

The sharp corner — compliance (get this right)

PointRuleVerdict
Data sourceUse official Maps Platform APIs (Street View Static, Places) — paid, licensed. No logged-in scraping of Google.🟠 API only
Nature of the numberA sole trader's mobile is personal data under UK/EU GDPR — even painted on a van.🟠
Lawful basisLegitimate interest + a written LIA, data minimisation (name + number + source + town only), privacy notice, easy erasure.🟠
Use of the numberPostal touch → QR-to-claim only. Prospect self-initiates. No cold call, no cold SMS.🟢 postal-only
Cold call / SMS to the numberPECR (UK) / UWG §7 (DE) — a painted number is not consent.🔴 avoid
Bots / scraping GoogleToS breach + GDPR.🔴 avoid

Verdict from the strategy: 🟠 Street View OCR → postal — treat as a v2 experiment, lower priority than the human channels. The referral partnership with a real sign-writer stays 🟢 cleanest of all; this is its automated, more-constrained cousin.

Where it sits in the priority order

On the cheapest-to-test ranking it's #6 of 6 — behind inbound FB posts, classifieds, the noticeboard turk, sign-writer referral, and ads. Build cost + GDPR care = lowest priority. Prove the free claim-flow first; automate with OCR at scale only once everything upstream is working.

Klaudius · Street View OCR · source outreach/lead-sourcing-strategy.md §4 · v2 experiment · official APIs, no bots, postal-only