When I was working at my last real estate tech startup, one of the problems we tried to solve for our multifamily investor clients was how to determine the best comps for any property.
It doesn’t matter as much for investors who operate in a single market – they know their comps well. But for anyone looking to invest in a new market, it’s difficult to know which properties are truly most comparable. Also, for companies with a national footprint, a strong algorithm for identifying rental comps enables them to create rent surveys for every property in their portfolio as often as they want. Very valuable for portfolio benchmarking.
We focused primarily on rent comps (which, to be honest, was because it was the easiest property type to get data for), but our customers frequently asked for expense comps and sales comps too. And we were using rent comps to inform the income side of an automated valuation model for multifamily properties, so it was important to get the most accurate comps possible.
Initially, we didn’t want to use asking rent as an input. Can’t predict rent using rent as a variable, right? But that meant that we were delivering a few comps that were very similar in terms of beds/baths, square footage, amenities and location, but sometimes very different in terms of rent.
This makes sense. You can have two units that look very similar on paper, but the quality of the buildings may be very different. This difference is easy to tell from looking at listing photos for the two properties, but at the time, it wasn’t feasible to do this at scale.
We thought of some approaches to help us derive quality, one of which involved letting users choose their own comps and learning from their choices – but there wasn’t quite enough volume on the platform to make it work in every market.
So we separated our models and used price in determining comps. Combining the FULL data set we had from rental listings with data on local amenities, demographics, crime and school ratings, we were able to generate some pretty good results for rent comps… but our outputs always seemed to miss a comp or two that a seasoned real estate investor knew were comparable. We were close, but not perfect.
This time around, we decided to focus on the problem of determining quality from property photos. After painstakingly labeling thousands of listing and property photos, we built QualityScore.ai to accomplish this.
With the Quality Score API, you can determine the room type, amenities/finishes present, and interior & exterior quality from any property photo. Developers can just send a batch of image URLs to the API and get this data returned via JSON, and we have integrations with Excel and Google Sheets available for non-developers.
At this point, we haven’t used Quality Score for comparable property detection, but we have combined it with the RealType API. To generate listing descriptions and alt-text for pictures automatically. So far, we’ve seen the most traction in these areas:
Some of our clients use it to analyze large portfolios (or even whole cities, the API is pretty cost-effective) to source opportunities where properties are in a great market but in poor condition. It’s the tried and true “buy the worst house on the best block” on steroids – they can survey buildings in every submarket simultaneously and focus on the best ones.
Listing sites see the value in extracting amenities and quality from photos and letting people filter/search for homes and apartments in a more robust way. Plus, we can generate both listing descriptions using ChatGPT and alt-text for photos, which are both critical for SEO. The rental listing market is very competitive, so it makes sense these companies invest heavily to rank ahead of the competition.
Every year, real estate servicers have to do physical property inspections to determine the condition of properties for insurance and lender reporting. Both inspection companies and the companies that provide their software have shown interest in the Quality Score API as a way to automate some parts of these inspections. It may not replace the need for annual inspections altogether, but it can help these companies do a “preliminary screening” and focus their efforts more effectively.
Earlier this year, we built a rent comps tool using Quality Score, we see a ton of potential in using it to identify rental comps like an experienced real estate broker would, but at scale.
The part that was missing before was that subjective human feedback on how comparable any two properties were in terms of quality. It’s pretty time consuming for someone to put a rent survey together for just one property, let alone at scale.
By combining data on rent, beds/baths, square footage, market characteristics, AND an objective measure of interior apartment quality, though, it’s possible. And quality score provides the missing piece.
Property managers, investors, brokers and appraisers all use HelloData to analyze multifamily comps, optimize rents, and increase deal flow.
