US Housecleaning Leads at Scale: A 30,000-Record Case Study

Project scope and data schema

We collected a set of must-have fields including the business name, publicly available email, phone number, city, and state. In addition, we enriched the data with secondary fields such as website URL, zip code (when available), and the source of information (e.g., Google Maps, Yelp, Craigslist, Facebook, or Google Search). We also inferred categories like “House Cleaning Service” or “Maid Service,” and captured ratings and review counts from platforms such as Google or Yelp, when present. Other fields included hourly rates in USD (if posted on Craigslist or Facebook), publicly advertised offers or coupons, the last-seen timestamp in ISO date format, and free-text notes (for example, “move-out cleaning” or “pet friendly”). Importantly, we only collected public business contact information intended for customer inquiries, used official APIs where appropriate, respected rate limits, and followed all platform rules.We collected only public business contact information intended for customer inquiries. We used official APIs where appropriate, respected rate limits, and followed platform rules.

Keeping it strictly residential

The most important rule was relevance. We needed home cleaning and maid services only. To enforce that rule, we used layered checks:

Allow List Signals

• Categories like “House Cleaning Service,” “Maid Service,” “Residential Cleaning,” “Home Cleaning.”
• Text signals like “apartment,” “home,” “residential,” “deep clean,” “move-in,” “move-out,” “weekly,” “bi-weekly.”

Deny List Signals

• “Janitorial,” “Commercial Cleaning,” “Industrial,” “Facility Management,” “Post-construction,” “Pressure Washing,” “Window Cleaning,” “Carpet Cleaning Only.”

Model and Rules

• A lightweight text classifier scored titles, categories, descriptions, and reviews.
• If a listing claimed both commercial and residential, it needed a clear residential signal to stay.
• Edge cases were queued for human review.

This method cut out noise and kept precision high for residential maid services.

Technical setup

• Extraction: Python async workers, source-specific parsers, official APIs where appropriate.
• Scheduling and queues: Airflow plus a message queue for rate-safe crawling.
• Storage: PostgreSQL for structured records and S3 for optional raw snapshots.
• NLP and rules: spaCy style tokenization, keyword scoring, and a simple classifier.
• Record linkage: blocking keys with phonetic and fuzzy matching for final merges.
• QA console: internal review panel to approve or reject edge cases.
• Delivery: CSV and Parquet. Optional direct load to HubSpot, Salesforce, or Close.

JSON template

Offer and pricing examples from public listings

• “Back-to-school special. Save $25 on move-out cleans.”
• “$35 per hour with supplies. $30 if you provide supplies.”
• “First-time clients get 15% off deep cleaning.”

When posts listed a rate or promotion, we captured it. If not, those fields were left blank.

Delivery and integration

File formats

• CSV for easy imports into most CRMs.
• Parquet for analytics.
• Optional PostgreSQL dump when teams want a direct restore.

CRM mapping

• Pre-mapped templates for HubSpot, Salesforce, and Close.
• Phone and email fields kept consistent with CRM validators.
• Optional “do not contact” flags if you want to respect suppression lists.

Documentation

• A short data dictionary with field definitions.
• A one-pager on refresh cadence and how to request updates.
• A changelog that describes known caveats and open questions.

Insights the client gained

Where the market is dense

• The dataset showed strong clusters in California, Texas, Florida, New York, and Illinois.
• Growing pockets were found in North Carolina, Colorado, Tennessee, and Arizona.
• This helped with territory planning and headcount allocation.

What offers move the market

• First-clean discounts and move-in or move-out specials were common.
• This informed ad creatives and email subject lines.

Who to call first

• Yelp and Google ratings highlighted hyperlocal leaders.
• The client set a minimum rating filter for premium outreach waves.

Pricing cues

• Craigslist and Facebook often listed hourly rates.
• Even partial coverage was useful to build a price band for each metro.

What we included and what we excluded

Included

• Residential housecleaning and maid services that publicly advertise to homeowners, renters, and property managers.
• Publicly available contact details.
• Ratings, review counts, rates, and offers when publicly posted.

Excluded

• Janitorial services, commercial-only providers, pressure washing only, window cleaning only, and unrelated services.
• Hidden or private contact information.
• Anything that required bypassing a login or ignoring platform rules.

Add-ons the client considered

• Deliverability boost with warmed sending domains and bounce tracking.
• Additional enrichment such as business age, business registry lookups, and service radius estimation.
• Monthly monitoring for rating changes, new reviews, and fresh offers.
• Lead scoring that combines rating, review count, and presence of a website or posted rates.

Appendix: field dictionary

• business_name — Name as shown on the listing or website.
• email — Public business email from listings or the company website. MX-checked.
• phone — Normalized to E.164 when possible.
• city, state, zip — Parsed from listing text or Google Maps.
• website_url — Canonicalized with protocol and path cleanup.
• source — google_maps, yelp, craigslist, facebook, or google_search.
• category_inferred — “House Cleaning Service,” “Maid Service,” etc.
• rating, review_count — From Google or Yelp when available.
• hourly_rate_usd — From Craigslist or Facebook posts when present.
• offer_or_coupon — Any public promotion captured verbatim.
• last_seen — ISO date the record was captured.
• notes — Short free text with service specifics.