Google Maps · Places
Operating businesses with websites, globally. Overture Maps Foundation Places (Meta + Microsoft + AWS + TomTom joint dataset, CDLA Permissive 2.0). 75.5M POI worldwide; 46.4M with at least one website. Foundation for outbound: scrape contact pages for emails, pull Google reviews, AI-distill pain points → personalized cold outreach.
Stratified 50K sample (80 categories × 700) of US POI with websites, served from SSR JSON for instant browsing. Full 46.4M filtered parquet on NAS + S3 — see the live search below to query the entire dataset.
DuckDB queries the 4.5GB filtered parquet on S3 directly via /api/places. Cold start ~15s, warm ~2s. Search every Overture place with a website worldwide — no SSR cap.
Per-business: 15 most-recent Google reviews via Outscraper API, then Claude Haiku distills the top 3 customer pain points. Powers personalized cold-email generation. Budget ~$200 for 3K leads.
Contact-page scrape + MillionVerifier SMTP verification. Augments Overture's sparse `emails[]` field (most rows have 0). Cost ~$0.001/verification.
Why Overture, not Google Places API
- Bulk-enumerable — Google Places API requires a search query (you can't list-all). Overture is a single parquet you scan.
- Storable — Google Places ToS caps caching at 30 days. Overture is CDLA Permissive 2.0 (Apache-style): permanent storage, commercial OK, redistributable.
- Free — Google would cost $20/1K Place Details. Overture costs $0; data is hosted on requester-anonymous AWS (
s3://overturemaps-us-west-2/). - Global by default — 200+ countries in one drop. Filter to
country = 'US'when you want a US slice. - Updated monthly — sponsored by Meta, Microsoft, AWS, TomTom; release pinned in
ingest/paths.py:OVERTURE_RELEASE.
What Overture does not give you: customer reviews, verified emails, social handles in structured form. Those are downstream stages — see the planned datasets above.