Tenant Outreach Targeting

PEU's Outreach Data Specialist asked what buildings call the Tenant Support Helpline for help most often, who owns them, and whether that landlord owns other buildings that might need outreach too. PEU's helpline fields 50,000+ calls a year, but outreach teams currently only canvass buildings that already generate high 311/HPD complaint volume — missing tenants who never knew how to complain in the first place. The desired outcome was to match high-call buildings to their responsible landlord and proactively reach tenants in that landlord's other buildings.

Sponsor: Emily Grigg, Outreach Data Specialist · NYC PEU

Tenant Outreach Targeting turns one distressed building into a portfolio-aware canvass list. It scores buildings for distress using public complaint and violation data, links each building to its landlord's other properties through a self-hosted ownership-clustering graph, and ranks sibling buildings by distress, unit exposure, and rent-stabilization status into a prioritized, exportable outreach queue. Because PEU's own helpline call records aren't public, the product substitutes HPD complaints and 311 data as an intake proxy today, with a documented, single-join swap path ready for when PEU connects its real call data.

• Intake · /tenant-outreach
• Validation · /tenant-outreach/validation
• Methodology · /tenant-outreach/methodology
• Helpline Drop-In · /tenant-outreach/helpline-drop-in

• HPD Multiple Dwelling Registrations — 193,567 rows
• HPD Complaints and Problems — 499,993 rows; HPD Violations — 412,502 rows
• 311 Service Requests — 1,500,000 rows
• HPD Housing Litigations — 239,434 rows; Marshal-Executed Evictions — 119,276 rows
• Rent-stabilized unit counts (2022 taxbill artifact) — 430,103 rows across 46,164 BBLs
• Public Advocate Worst Landlord Watchlist (2025) — 99 landlords, used for validation

Socrata ingests land into Postgres (`nyc.tenant_outreach_*`), and a transparent, weighted scoring pass — plus a Union-Find ownership-clustering graph, not a machine-learning model — derives building distress scores and landlord portfolios, which serve the live queue and building pages directly.

Live verification rated this product NEEDS WORK: the underlying data and scoring were cross-checked row-for-row against NYC Open Data and hold up, but two documentation/UI issues were flagged — a mismatched helpline-CSV contract and a methodology page that wrongly showed reconciliation data as "not yet computed."

• PEU's own helpline call records (200,000+ calls) — not public; complaints/311 data is used as a proxy today
• Outreach-outcome data that would be needed to train a learned prioritization model
• Individual tenant contact information — never surfaced; output stays at the building level