property-manager-intel

name	property-manager-intel
description	Build a ranked rental property manager list with scale, geography, and contact info — separating third-party PMs, brokerages, institutional landlords, and listing platforms. Use for PM targeting or "who manages rentals in [market]".
user-invocable	true
argument-hint	[market name or 'national']
allowed-tools	Read, Grep, Glob, Bash, Write, Edit, WebFetch

Property Manager Intel

Target: $ARGUMENTS. Pass a market name (MSA, city, or ZIP) for a single-market pull, or national for a nationwide list. If no argument is given, ask the user which scope they want before proceeding.

Trigger phrases

Invoke this skill on any of: "property manager list", "PM national list", "PM coverage analysis", "property manager intelligence", "property manager ranking", "rental property managers in [market]", "build PM profiles", "PM targeting", "who manages rentals in [market]", or any variant of "build/make/give me a PM list".

Produces a four-tab xlsx segmenting rental listings by the type of entity doing the leasing. Third-party PMs, brokerages, institutional landlords, and software / listing platforms are different businesses and are ranked on separate tabs.

• PM Profiles — third-party property managers, including corporate- housing operators (Blueground, Nomad Homes). The main ranked list.
• Brokerages — Keller Williams, RE/MAX, Coldwell Banker, and similar real-estate brokerages. Their agents list rentals, but the brokerage itself does not manage rentals the way a PM does.
• Institutional Landlords — Invitation Homes, American Homes 4 Rent, Progress Residential, FirstKey, Tricon, and similar SFR operators that manage their own portfolios.
• Rental Software & Platforms — Zumper, AppFolio, ShowMojo, and similar listing platforms and PM software. Visible as a separate leasing-activity surface rather than ranked alongside PMs.

Workflow

1. Confirm scope

Ask the user: "Do you want a national PM list, or a specific market (MSA, city, or ZIP)?"

• Market — resolve via mcp__...__search_locations with the user's query and location_type set if known (metro / city / zip / county). Use the returned parcl_id in the next step.
• National — do NOT pass parcl_ids. The endpoint treats an omitted parcl_ids array as "national search across all rentals".

2. Pull rental listings

Call the motivated_renter_properties MCP tool.

IMPORTANT: filters is a nested object parameter. Pass it as a JSON object — do NOT JSON-encode it as a string. The tool will reject a stringified filters value with a type error. Same for parcl_ids inside filters: pass as an array of integers, not a string.

Market scope — arguments to pass:

• filters: {"parcl_ids": [<resolved_parcl_id>], "limit": 50000, "offset": 0} (object, not string)
• preview: false (boolean, not string)

National scope — omit parcl_ids entirely:

• filters: {"limit": 50000, "offset": 0} (object, not string)
• preview: false

Two-step flow per batch:

1. Call once with preview set to true to see estimated_records and credit cost. For national, estimated_records is capped at limit, so it doesn't give you a true total — you have to paginate until exhausted.
2. Call with preview set to false to get a presigned download_url. Download the CSV (curl / requests — URL expires in 1 hour).

Paginate by incrementing filters.offset by limit until a batch returns fewer than limit records. Save each batch CSV and concatenate.

Write the combined CSV to a working directory (e.g. ./pm_workspace/listings.csv).

3. Run the pipeline

All four scripts declare dependencies inline via PEP 723, so uv run resolves them automatically into an isolated environment — no venv activation required. Prerequisite: uv installed (brew install uv if not).

cd scripts/

uv run resolve_pm.py \
    --input ../pm_workspace/listings.csv \
    --output ../pm_workspace/resolved.csv

uv run dedup_pm.py \
    --input ../pm_workspace/resolved.csv \
    --output ../pm_workspace/deduped.csv

uv run build_profiles.py \
    --input ../pm_workspace/deduped.csv \
    --output ../pm_workspace/profiles.csv

uv run build_spreadsheet.py \
    --input ../pm_workspace/profiles.csv \
    --output ../pm_workspace/pm_list.xlsx

Each script prints a one-line summary — read it to verify counts look reasonable before moving on.

If uv is unavailable, the scripts also work with a plain python3 <script>.py invocation, provided pandas, numpy, and openpyxl are importable from the active Python. No behavior difference — the PEP 723 metadata is inert to the Python interpreter.

4. Deliver

Give the user:

• The xlsx file path.
• A short summary: unique PMs + unique listings per tab, top 5 third-party PMs by active rentals, top 3 institutional landlords by active rentals, and top 3 entities on the Rental Software tab.
• Any points from "How to Read This Output" below that apply.

How Routing Works

The company_type column is the authoritative classification. Three curated reference lists in scripts/lib/blocklists.py provide overlay logic for cases the company_type column does not cover:

• RENTAL_SOFTWARE — routes listing platforms, PM software, and generic email-domain names to the Rental Software & Platforms tab.
• BROKERAGES — routes brand-name real-estate brokerages to the Brokerages tab.
• INSTITUTIONAL_LANDLORDS — routes named SFR operators to the Institutional Landlords tab and collapses spelling variants (e.g. "FirstKey Homes" / "FirstKeyHomes" / "Firstkey Homes, LLC") to a single canonical display name.

PM Resolution Waterfall (7 tiers, first match wins)

Per listing row, these candidate fields are checked in priority order (matches scripts/resolve_pm.py resolve_row):

1. company_name when company_type is property_management or institutional_landlord.
2. property_manager direct (when not a known rental-software override).
3. company_name fallback (any company_type, not software).
4. agent_business when is_property_management_company == 1 (not software).
5. agent_business when property_manager is a known software / listing platform — the agent_business is the actual manager.
6. agent_business expanded (no PM flag required, not software) — the coverage-expansion tier that catches PMs missing from earlier tiers.
7. Last-resort: the first populated candidate even if it's a software name, so categorize() can still route the listing to the Rental Software tab. Software listings are segmented, not dropped silently.

Categorization (which tab a listing lands on)

company_type is the primary routing signal. The overlay lists run as a fallback when company_type is NaN or other.

company_type → tab:

company_type	Tab
property_management	PM Profiles
corporate_housing	PM Profiles (Blueground, Nomad Homes, etc.)
institutional_landlord	Institutional Landlords
brokerage	Brokerages
rental_software_platform	Rental Software & Platforms
listing_platform	Rental Software & Platforms
homebuilder	filtered out (no PM signal)
other / NaN	fall through to override fallback

Fallback (runs only when the primary table above does not match):

1. entity_name / parent_company matches an institutional alias → Institutional Landlords tab.
2. Resolved PM name matches an institutional alias → Institutional tab.
3. Resolved PM name matches RENTAL_SOFTWARE → Rental Software & Platforms tab.
4. Resolved PM name matches BROKERAGES → Brokerages tab.
5. Default → PM Profiles tab.

Whenever a listing routes to Institutional Landlords, its display name is resolved to the canonical brand (e.g. "First Key Homes" and "FirstKey" both display as "FirstKey Homes").

How to Read This Output

A few things worth keeping in mind when working with the delivered xlsx:

1. Institutional active_rentals = listings, not doors managed. The numbers on the Institutional Landlords tab reflect active rental listings at the time of the pull. A large institutional operator may own tens of thousands of homes nationally with only a fraction actively listed at any given moment. The tab is a leasing-activity view, not a portfolio-size view.

2. The Brokerages tab tracks leasing activity, not PM relationships. Brokerages (Keller Williams, RE/MAX, Coldwell Banker, and similar) surface here because their agents list rentals. That's real market activity worth seeing, but it usually reflects an agent-of-record relationship rather than a third-party PM contract. Treat it as a distinct signal from the PM Profiles tab.

3. The Rental Software & Platforms tab is a leasing-activity surface, not a PM directory. Zumper, AppFolio, and similar services appear here because their users list rentals through them. These are platforms facilitating listings, not property managers — the tab is here so the activity is visible without being ranked alongside PMs.

4. Per-PM geography is based on active listings. num_states / num_msas / num_cities / num_zips count where a PM currently has rentals listed, not where they operate. A PM with 5 cities this pull may operate in 20 — the active-listing snapshot is a lower bound.

5. Contact-fill rates vary by source. representative_email and representative_phone are populated where the listing carries agent contact info; fill rate is roughly 40-60% on third-party PMs and near-zero on institutional operators (who don't expose listing-agent contacts at the property level).

6. The primary_company_type column is the authoritative classification. Values include property_management, brokerage, institutional_landlord, corporate_housing, rental_software_platform, listing_platform, and other. Use this column to filter the PM Profiles tab to only rows explicitly classified as property_management.

When the user asks to iterate

Common iterations after the first run:

• "Too many brokerages in my third-party tab" — identify the names, add their normalized form to BROKERAGES, re-run resolve_pm + downstream.
• "Institutional X is missing" — add to INSTITUTIONAL_LANDLORDS dict with alias set, re-run.
• "I want just [one MSA] from an existing national pull" — filter the listings CSV by msa_name or msa_parcl_id before running resolve_pm.

Schema Notes

The pipeline expects these columns in the input CSV (from motivated_renter_properties):

parcl_property_id, latest_event_date, state, city, zip5, msa_name, property_type, sq_ft, latest_listing_rent, days_on_market, motivated_renter_index_value, motivated_renter_index_label, algorithmic_pricing_detected, property_manager, company_name, company_type, parent_company, is_property_management_company, agent_name, agent_business, agent_email, agent_phone, entity_name, investor_id.

company_type values: property_management, rental_software_platform, brokerage, homebuilder, institutional_landlord, listing_platform, corporate_housing, other.

Each parcl_property_id appears at most once per pull.