In commercial real estate, it’s essential to perform a detailed real estate market analysis when setting rents and analyzing acquisition opportunities. But these days, opportunities are scarce, rent growth is flat or declining, and the multifamily industry in particular is more competitive than ever. Most real estate professionals are running at 110% capacity, and unfortunately, this means they don’t always have time to comprehensively analyze rent comps.
The data science team at HelloData.ai recently launched a new rent comp analysis algorithm that combines the power of computer vision and AI to generate an incredibly robust multifamily market analysis in seconds. In this post, we’ll delve into some of the techniques we're using to identify and analyze comparable properties with greater precision than ever before.
The look and feel of a property can significantly influence a prospective resident’s decision on where to rent. Everyone understands that renters look at location, price, and listing photos to determine where they want to live, but until recently, the only way to analyze the listing photos was to actually look at the listings.
Using computer vision, we built an algorithm that analyzes both listing photos and Google Street Views to objectively assess the quality and condition of real estate – without a human ever setting foot on the property. This visual analysis (which represents about 20% of the weight in our rent comp detection model) helps us account for the unique attributes of both the subject property and its comps – facilitating true apples-to-apples comparisons.
The look and feel of the neighborhood is something every real estate broker and investor understands is important for potential residents, but we’ve never seen other real estate data providers take this into account – so we developed a way to capture this information.
Beyond the property itself, our computer vision algorithms examine the character of the surrounding neighborhood by analyzing Google Street View images up and down the block on which the property is located. This approach helps account for each neighborhood’s overall quality and unique attributes, which we are using as an input for our comps model.
The layout and room sizes in a floorplan can have a massive impact on desirability. Things like narrow hallways and tiny bedrooms aren’t exactly appealing in today’s market, but each particular style and layout could perform very differently from one market to the next.
In addition to analyzing the listing photos and google street views for every property in our pipeline, we use computer vision to study floorplans – capturing room sizes, dimensions, and layout types. Coupled with granular rent and availability data, this approach helps assess the appeal of different unit layouts with a level of granularity that’s never been possible before when analyzing rent comps.
Many rent comp identification models take physical characteristics like year built, number of units and distance from the subject property into account. We take this analysis to another level, though.
Our algorithms extract and standardize data on over 300 unique amenities and attributes at both the building and unit levels. We pull this data from listing descriptions and bulleted lists of amenities using a combination of natural language processing and conditional logic. This highly detailed approach provides a comprehensive picture of the amenity offerings for every property in our database, which we use to score the similarity of amenity packages.
Differing crime rates, demographics, and school district ratings can result in vast differences in real estate investment potential. These variables can even change from one block to the next in some markets, and failing to take them into account can make or break an investment. Additionally, some markets may not have ANY comps nearby, forcing analysts and appraisers to perform time-consuming analyses of adjacent markets.
Our system aggregates, normalizes, and joins market level data to the subject property and its comps. Markets are defined with respect to each property – not an arbitrary geography like a radius or neighborhood boundary. Because of this, we can detect nearly every property that competes for the same renters… even if they’re located several blocks away or in an adjacent suburb with similar demographics.
Most real estate data providers are only collecting rent and availability data on a monthly or weekly basis, so they miss units that are only on the market for a very short time. These are often the most correctly priced units though – the ones that stay on the market for weeks and weeks tend to be overpriced, making them poor comps for analysis.
By collecting data every single day from several major listing sites, we capture nearly every listing that hits the market. Even units that were listed on a Monday and leased on a Thursday will show up in our dataset. This frequency of data capture helps eliminate survivor bias from overpriced units, so our users know that when they are comparing rents with their comps, they are not looking at that vacant penthouse unit that's been on the market for months.
Collecting advertised rents every day is great, but if properties offer concessions, the actual rents are likely lower than advertised. This was becoming less common with the increased adoption of revenue management solutions, which dissuade users from offering concessions... but with real estate market growth slowing in 2023, specials are becoming more frequent again.
To help get as close as possible to actual rents, we developed an algorithm that uses a fine-tuned large language model to extract the dollar value of any specials posted in real estate listings. Even with the complexity of specials offered (and trust us, some of them are so complex we wonder how renters even understand them!), we can pick up the real dollar value of concessions with over 94% accuracy. By combining this data with the last listed rent for each unit, we can develop a very close approximation of the actual rents. No other real estate market analysis solution we’ve seen can do this.
No one in real estate enjoys spending hours scouring listings and calling around for pricing, and there's always subjectivity and debate around which rental comps are the most similar for a given analysis. Manual data aggregation wastes time that very few people these days have to spare, and subjective analysis introduces uncertainty that can put deals at risk. Until recent advancements in AI hit the market, however, there hasn't been a better solution for multifamily real estate analysis.
We believe AI can make real estate apples-to-apples, which means accurately identifying rent comps should no longer require the same level of human input. Using AI for real estate analysis saves people time, reduces risk, and most importantly, levels the playing field - so mom and pop operators can enjoy the same level of analysis as national real estate players. We look forward to a future where AI makes real estate analysis more objective, accessible and enjoyable for everyone!
--
If you're interested in trying out our comps model or learning more about HelloData.ai, reach out at contact@hellodata.ai and let's set up a demo.
Learn more about Hello Data, Inc. from the resources below:
Property managers, investors, brokers and appraisers all use HelloData to analyze multifamily comps, optimize rents, and increase deal flow.
