Scraping Bars & Restaurants Paris + 92/93/94/95 (Google Maps & Uber Eats)
Simon RochwergLarge-scale extraction and enrichment of bars & restaurants in Paris + 92/93/94/95 from Google Maps and Uber Eats. Focus on extracting and verifying emails, capturing all phones (incl. 06/07), cross-source deduplication, and delivery of clean XLSX/CSV files.
🎯 Context & Goals
Build a Bars & Restaurants prospecting dataset for Paris + 92/93/94/95 that maximizes:
- verified, high-quality emails, 2) all phone numbers including 06/07, and 3) coverage by combining Google Maps + Uber Eats.
🤖 What the AI does in the pipeline
-
Website discovery & selection (AI + SERP API)
- Connect Google Maps entities to SERP API for targeted Google queries.
- An AI ranker automatically selects the best official website (disambiguates homonyms, ignores marketplaces/directories).
-
Structured extraction (Parser + AI normalization)
- Our parser scrapes the selected website and extracts emails, phones (fixed + 06/07), addresses, social links (IG/FB/TikTok), and legal mentions.
- An AI normalizer standardizes formats (email/phone/address), fixes common anomalies, and completes fields from public hints.
-
Smart email prioritization (AI scoring)
- An AI classifier scores each email (e.g.,
contact@,info@,firstname.lastname@domain, etc.). - It prioritizes direct/personalized addresses (outreach-friendly) and de-prioritizes generic, no-reply, or platform addresses.
- An AI classifier scores each email (e.g.,
-
Anti-bounce verification (technical validation)
- Syntax (RFC) and MX/DNS checks.
- Optional SMTP routing check (no real send) to reduce bounce rate.
- Deduplication within and across sources (GMaps ↔ Uber Eats) using email/phone/URL keys.
-
Consolidation & quality controls
- Merge Google Maps (address, category, hours, reviews, site) with Uber Eats (often more 06/07 and additional emails).
- Paris by arrondissement + communes in 92/93/94/95 for maximum coverage.
- Filter out non-prospect domains (marketplaces/platforms).
-
Production-ready deliverables
- XLSX + CSV (stable schema), recap sheet, field dictionary, source IDs, and extraction timestamps.
🔎 Anonymized sample (domains visible)
(5 rows from the exports; names/coordinates partially masked — domains kept visible)
| source | name | phone | city | |
|---|---|---|---|---|
| Google Maps | So p. | a***@gmail.com | ********90 | Valenton |
| Google Maps | C***r … Cr | i***@orange.fr | ********50 | Chennevières-sur-Marne |
| Google Maps | C***n St *… | c***@yahoo.fr | ********50 | Puteaux |
| Google Maps | T**k * P****t … | i***@laposte.net | ********40 | Ivry-sur-Seine |
| Uber Eats | Bo R***t | c***@restaurant-paris.fr | ********21 | Paris 11ᵉ |
Final deliveries contain full (unmasked) emails and phone numbers. Masking here is only for the example.
📈 Key Figures (highlights)
- Paris (Google Maps): Bars 1,205 emails / 2,947 rows (40.9%); Restaurants 1,422 / 3,711 (38.3%).
- 92/93/94/95 (Google Maps): 3,708 unique emails delivered.
- Uber Eats (Paris): ~3,000 restaurants in Paris + ~1,800 nearby; strong 06/07 coverage and additional emails.
Why Uber Eats? It’s more “direct” on the merchant side: you frequently find more mobile numbers (06/07) and emails that don’t appear on Google Maps — a powerful complement for outreach.
🧰 Stack & Practices
- Python, SERP API, parallel workers, robust retry logic.
- AI for ranking, normalization, and email scoring (prompting + business rules).
- Email validation (syntax, MX, optional SMTP check), cross-source deduplication.
- XLSX/CSV exports + field dictionary; execution logs for auditability.
✅ Outcomes
- A clean, verified, consolidated dataset ready for multi-channel outreach.
- Significant increase in mobile (06/07) coverage thanks to Uber Eats.
- Immediate import into CRM / n8n; lower bounce rate and less manual qualification.
💰 Budget & Milestones
- Google Maps (Paris + 92/93/94/95): €600 incl. VAT.
- Uber Eats (Paris): €250 incl. VAT (discount applied).
- Total: €850 incl. VAT — deliveries over ~3 weeks.