AI Services Market › Use cases › Work & Operations
Work & OperationsDone with Claude Code, Claude 2026-11-02
Score off-market real estate deals from emails
A pipeline has a Python worker read deal emails, Claude extract fields into fixed JSON, and code score them; the poster is stuck on incomplete data and login-walled sources.
- What They Did
- The poster wired a dedicated Gmail to collect saved-search alerts and wholesaler emails, then had a Python worker split multi-deal emails and send each deal to a low-cost Claude model to extract fixed JSON fields, leaving unknowns null. Plain Python, not the LLM, handled the single-family and multifamily scoring, with paid lookups only after a cheap price check; output went to Supabase, a Streamlit dashboard, email alerts, and a shared Google Doc for coordination.
- How It Went
- It worked well enough to centralize incoming deals and score what had data, but many leads stalled because about half arrived without a street address and others lacked unit counts or rents. The poster says those missing fields, plus login-walled sources that block automation, kept deals stuck at a needs-info stage.
- Worth Knowing
- Their main catch is cost control and gating: expensive data lookups only run after a cheap initial price check.
Try it yourself with Claude Code
Help me build a Python script that reads real estate deal emails from [mailbox or folder of exported .eml files], uses Claude to extract fields into a fixed JSON schema (address, asking price, units, rents, NOI, seller contact), and uses plain code to score each deal against [my criteria]. When data is missing, mark the field as null and list it in a 'needs follow-up' output rather than guessing. Done means I can run it on a sample of 10 emails and get a ranked CSV of deals with scores and missing-data flags.
Do this with a skill.
These AI Services Market playbooks do this exact job.