TLH Salesforce + MLS AI Agents

Overview

This spec outlines a phased approach to integrating MLS data scrubbing and AI agents into TLH's Salesforce instance. The goal: automate deal discovery, property analysis, and reporting while maintaining human oversight.

Current State

Metric	Value
Leads	13,042
Opportunities	5,767
Active Flows	183
Integrations	LeftMain CRM, DocuSign, RingCentral, smrtPhone, CallRail

Note: LSTMJ already built a prototype that syncs MLS data into Salesforce.

Core Components

1. MLS Data Scrubbing Agent

Monitors MLS feeds for properties matching TLH acquisition criteria.

Function	Description
Feed Monitoring	Watch MLS for new listings in target areas (SD, OC, LA)
Criteria Matching	Filter by: estate/probate mentions, senior seller signals, price range, property type
Data Enrichment	Pull property history, tax records, owner info
SF Sync	Create/update Lead records with MLS data
Alerting	Notify team of high-priority matches

2. Deal Analysis Agent

Runs financial analysis on properties.

Function	Description
Pro Forma Generation	Auto-calculate acquisition costs, rehab estimates, ARV
Comp Analysis	Pull recent sales, adjust for condition
Risk Scoring	Flag title issues, permit problems, market risk
Recommendation	Buy/pass/investigate-further rating

3. Reporting Agent

Automated dashboards and alerts.

Function	Description
Daily Pipeline	New leads, stage changes, stuck deals
Market Trends	Price movements, inventory levels, days-on-market
Performance	Conversion rates, CPA, acquisition velocity
Anomaly Alerts	Unusual patterns, data quality issues

Phased Implementation

Phase 1: MLS Integration In Progress

Weeks 1-3

Build on LSTMJ's MLS→SF prototype
Define matching criteria with Paul/Jeff
Set up Lead creation flow
Basic alerting (email/SMS on matches)

Phase 2: Deal Analysis Planned

Weeks 4-6

Connect to comp data sources (Zillow API, Redfin, county records)
Auto-populate Acquisition Pro Forma object
Build risk scoring model
Human review workflow (approve/reject/revise)

Phase 3: AI Agent Layer Planned

Weeks 7-10

Deploy AI agent to interpret listings (natural language)
Auto-draft outreach messages
Integrate with TLH Orchestrator
Connect to Marketing Team agents for lead nurture

Phase 4: Full Automation Planned

Weeks 11-12

End-to-end: MLS → Analysis → Outreach → Pipeline
Self-improving criteria (learn from closed deals)
Expansion to additional MLS feeds (LA, Bay Area)

Technical Architecture

Component	Approach	Notes
MLS Data Source	RETS/API feed or scraping	Need to confirm MLS access
Processing	Serverless (AWS Lambda / Cloudflare Workers)	Per Tran: serverless + databaseless
SF Integration	Salesforce API (existing sf CLI)	Already have API access
AI Layer	Claude API / OpenAI	For natural language analysis
Orchestration	OpenClaw agents	Ties into Marketing Team architecture

Open Questions

🤔 Need input from Paul/Jeff/Tran:

MLS Access: Which MLS feeds do we have access to? CRMLS? Sandicor? Others?
Matching Criteria: What signals indicate a good estate/probate opportunity? (keywords, owner age, days on market, price drops?)
Pro Forma Inputs: What's the standard formula for TLH acquisition analysis? Rehab cost assumptions?
Priority Markets: Start with SD only, or include OC from day 1?
LSTMJ Prototype: What's the current state? Can we demo it?
Integration Depth: Just leads, or full opportunity creation with stages?

Success Metrics

Metric	Target
Leads from MLS scrubbing	50+/week
Time to analysis	<1 hour from listing
Analysis accuracy	90%+ alignment with human review
Deals sourced via automation	20% of pipeline within 6 months