How we're using public data and machine learning to rate apartments and landlords
Finding a new apartment can be one of the most stressful tasks to face–especially in New York City and San Francisco. Crazy prices, pushy brokers, cancelled viewings and trying to figure out if every "bargain" on Craigslist is a scam make looking for a new rental a nightmare. What's worse is that unlike most other decisions, there's almost no accurate, unbiased information on the internet to help you–until now.
For restaurants, there's Yelp. For cars, there's CARFAX and Consumer Reports. For apartments, there's Augrented. Read on to find out how our unique AI-driven apartment ratings work, and how they can save you time, money and stress.
Where does the data come from?
Unlike sites that let you review your landlord or apartment, Augrented only uses official data sources. Every rental building in New York City is regulated by the Housing and Preservation Department, which requires landlords to register every year. The Department of Buildings, Department of Health and Mental Hygiene and many other authorities also inspect and regulate multifamily properties in NYC.
In San Francisco, while there's no list of landlords and their buildings, the Rent Board and other city departments like the Department of Building Inspection are responsible for ensuring apartments are safe and properly maintained.
All of these organizations are required to make most of their records public–either in response to state freedom of information laws, or local laws aimed at making data on housing freely available to tenants and others.
What does the data cover?
The data in Augrented building and landlord rating reports covers a huge range of topics.
For every New York City apartment, we show you:
- Housing Code violations
- Environmental Control Board violations
- Complaints to 311 and the Department of Buildings about building conditions
- Complaints to the Department of Health and Mental Hygiene about hazardous materials
- Office of Administrative Trials and Hearings cases involving the building
- Evictions by marshalls
- Rat inspections by the Department of Health
- "Housing Part" actions brought by tenants or the HPD against negligent or harassing landlords
- Open Market and Handyman Work Orders issued when landlords fail to maintain their properties
- Vacate Orders
- Tax, water and other charge liens against the building that have been sold
- Construction and electrical permits
- Sales recorded by the Department of Finance
Every San Francisco apartment building report includes
- Department of Building Inspection violations and complaints
- Complaints to 311
- Department of Health inspections of restaurants at the location
- Soft Story earthquake safety retrofit requirements
- Deck safety certification (604 Affidavit)
- Planning Department records
- Tenant buyouts
- Fire incidents, complaints, violations and inspections
- Construction, electrical and plumbing permits
- Department of Health apartment building complaints (coming soon)
We also provide basic building data from the HPD/PLUTO tax data (NYC) or Assessor-Recorder rolls (SF).
How do you know which buildings a landlord owns?
Finding out who's your landlord can be surprisingly difficult. Particularly in New York City, many landlords hide behind anonymous LLCs registered at the same address as hundreds of other shell companies. Augrented shows you the registration data provided to the HPD, and we use the same "fuzzy matching" technology used by fraud detection systems to identify buildings that have been registered (or entered) with slightly different details, such as misspellings and reversed names. The owner (if it's an individual), managing director or head officer is treated as the "landlord."
In San Francisco, we use the names of the person or entity responsible for paying property taxes.
How do you predict what's going to happen?
We use state-of-the-science machine learning techniques to create a predictive model, which is "trained" to predict which buildings will have code violations in the coming year. It's tested using historic data–predicting what would happen in 2019 and then checking the predictions against real data–and then generates predictions for the coming year. These are the same technologies used by investors to predict asset prices, credit default risks, and similar events.
Our model includes basic building features, neighborhood identifiers (although not the address or anything which specifically identifies the building to the algorithm) and, most importantly, the building and landlord's history.
How accurate are Augrented risk ratings?
We only publish ratings which are accurate for the overwhelming majority of properties in testing. Both our New York City and San Francisco renter risk models accurately predict outcomes for over 90% of buildings–that's far higher than credit risk models (including those used by landlords to "rate" tenants!).
Of course, no model is perfect, and it's possible for buildings which are predicted to be "low risk" to experience unexpected issues–or for the landlord of a "high risk" property to suddenly improve their business practices! Our model predicts violations, and maintenance issues may exist which tenants do not report for a variety of reasons, including not knowing how to access city services, landlord intimidation or lack of awareness of their rights to safe housing.
I'm a landlord, can I make you remove my building's report (or pay you to change it)?
You can't pay to change your Augrented report, any more than you can pay Equifax to raise your credit score–that's how ratings work!
If a report is showing incorrect landlord data, such as not recording a change of ownership, that will be updated when the HPD or Assessor-Recorder records change.
Records are located using the identifiers entered by city agencies, including address, parcel number (block and lot), Building Identification Number, HPD identifier codes, geospatial data, and other features. Please note that you may not have been notified of all violations and complaints by city agencies at the time they occurred. However, if you believe data is incorrectly linked to a property you own, please let us know. You'll need to provide proof of identity, ownership, and a transcript of the building's record certified by the agency responsible for the dataset in question.