The future of commercial real estate isn\u2019t about following trends;<\/strong><\/h3>\n\n\nit’s about setting them. In a market where every decision counts, it\u2019s the VTS mission to arm CRE professionals\u2014owners, brokers, and operators\u2014with the kind of real-time insights that drive smarter, faster, and more profitable decisions. This is why AI is a cornerstone of our product strategy. The success of our customers demands nothing less.<\/strong><\/p>\nIn this article, we\u2019ll dive into how we\u2019re evolving beyond traditional real estate tech by integrating the latest advancements in AI and machine learning. Using cutting-edge tools and techniques like natural language processing, network science, statistics, and transformers, we\u2019re combining data from thousands of sources to give you a clear, real-time, and actionable view of the CRE market<\/strong>. This is more than just a new way to look at data\u2014it\u2019s a new way to lead in commercial real estate.<\/p>\n\n\n\u201cThe revolution hasn\u2019t happened yet\u201d<\/strong><\/h3>\n\n\nAlmost all large industries from travel to finance have been completely turned upside down by data and machine learning. But real estate always seems one step away from fully being \u201cdisrupted\u201d by technology.<\/p>\n\n\n
In fact, it\u2019s still quite a few years behind. Whereas most financial markets have centralized, digitized reporting and analysis, most CRE transactions and the details within them have been scattered across various digital applications, news articles, or buried in an excel spreadsheet at a brokerage firm. If you wanted to learn why a particular office building stood empty for 17 months while the office building adjacent was 100% occupied, you might call a broker, tour the space to compare interior fit-outs, or query third-party data to understand leasing activity in the particular submarket.<\/p>\n\n\n
This is because the industry has remained very fragmented, and there\u2019s been real professional inertia to maintain opacity. Matt Levine<\/a> of Bloomberg states that buildings are like bonds: \u201cthere are a lot of them, they\u2019re all different, they each have idiosyncratic weird issues, the trade size tends to be in the eight figures, no individual building trades all that often\u201d, and that the facts about them are subjective, often not \u201cthat computer legible\u201d.<\/p>\nFor these reasons and more, at VTS, we believe AI\u2019s biggest impact within CRE will not come from models that draft emails or provide rote answers to empirical questions<\/strong>. It\u2019ll come from far greater challenges associated with analyzing data streams with higher levels of abstraction, and empowering professionals to understand \u201cfacts\u201d about the real world.<\/strong><\/p>\n\n\n![\"\"](\"https:\/\/www.vts.com\/wp-content\/uploads\/2024\/09\/TNS-Fast-2.gif\")
<\/figure>\n\n\nAmong several capabilities, our models are trained to make real-world connections between <\/em>![\"\"](\"https:\/\/www.vts.com\/wp-content\/uploads\/2024\/09\/Tenant-Clusters-via-Word-Embeddings.gif\")
<\/figure>\n\n\nWhat we’ve done with artificial intelligence <\/h3>\n\n\n
Technical advances in AI are quickly catching up with CRE industry needs, with various solutions being built to track, aggregate, and link information across a tenant\u2019s and a property\u2019s lifecycle. From de-anonymization (the practice of linking anonymous data to publicly available information), to entity disambiguation (the task of assigning unique identities to disparate records), and word embeddings (the representation of words in high-dimensional vector space) – these technologies power the domain-specific AI solutions that VTS has built internally to connect the dots across billions of data points and thousands of unique client integrations.<\/strong><\/p>\n\n\nOur data platform orchestrates flows of first-party data between our Lease Management<\/a>, Market<\/a>, Tenant Experience<\/a>, and Data<\/a> products so that we can derive insights not just on a single building, but also the relationships between them and understand the interaction between physical space and tenant behavior in the real world<\/strong>. For example, we can track for a unique tenant company:<\/p>\n\n\n\n- The number of repeat views they had for a vacant space online<\/li>\n\n\n
- The physical tours they took across buildings<\/li>\n\n\n
- The amenity reservation behavior of their employees<\/li>\n<\/ul>\n\n\n
By focusing on training domain-specific ML models to do single tasks (like de-anonymization and disambiguation) really well, we\u2019ve also unlocked the ability to produce unique insights about market supply and demand.<\/p>
Take our tenant requirement disambiguation model, for example, which finds similar-looking deal logs that represent the same unique tenant touring in a specific market. The model leverages similar techniques to other AI applications, namely high dimensional vector embeddings, to find similar records across spatio-temporal attributes (in other words – <\/em>deal opportunities logged in similar locations and at similar times with similar broker or company attributes). The model currently performs at 92%+ accuracy to human-labeled data, and is on average, 3% off in demand prediction numbers<\/strong>.<\/p>\n\n\n![\"\"](\"https:\/\/www.vts.com\/wp-content\/uploads\/2024\/09\/LexicalFast.gif\")
<\/figure>\n\n\nDiagrammatic representation of how our model finds relationships across deal logs, using semantic, lexical, and spatio-temporal distances\u2014distances that are impervious to errors in data entry<\/em><\/p>\n\n\nThis model is one element that powers our monthly VODI index<\/a> and is part of the broader VTS Demand Model<\/a>. We have employed similar techniques on company and property disambiguation tasks, whereby we match entities across thousands of systems to the same real world record. We do this by leveraging pre-trained word embeddings, and calculating a custom loss function to perform matching operations that overcome bad data entry and missing attributes <\/strong>(such as \u201cspace\u201d, \u201ccondition\u201d, \u201ctenant name\u201d, \u201ctenant industry\u201d).<\/strong><\/p>\n\n\n![\"\"](\"https:\/\/www.vts.com\/wp-content\/uploads\/2024\/09\/vodi-national-change-2024-07-v1.avif\")
<\/figure>\n\n\nSnapshot of July 2024 VODI Index<\/a><\/em><\/p>\n\n\nWhat\u2019s next<\/strong><\/h3>\n\n\n
In this article, we\u2019ll dive into how we\u2019re evolving beyond traditional real estate tech by integrating the latest advancements in AI and machine learning. Using cutting-edge tools and techniques like natural language processing, network science, statistics, and transformers, we\u2019re combining data from thousands of sources to give you a clear, real-time, and actionable view of the CRE market<\/strong>. This is more than just a new way to look at data\u2014it\u2019s a new way to lead in commercial real estate.<\/p>\n\n\n Almost all large industries from travel to finance have been completely turned upside down by data and machine learning. But real estate always seems one step away from fully being \u201cdisrupted\u201d by technology.<\/p>\n\n\n In fact, it\u2019s still quite a few years behind. Whereas most financial markets have centralized, digitized reporting and analysis, most CRE transactions and the details within them have been scattered across various digital applications, news articles, or buried in an excel spreadsheet at a brokerage firm. If you wanted to learn why a particular office building stood empty for 17 months while the office building adjacent was 100% occupied, you might call a broker, tour the space to compare interior fit-outs, or query third-party data to understand leasing activity in the particular submarket.<\/p>\n\n\n This is because the industry has remained very fragmented, and there\u2019s been real professional inertia to maintain opacity. Matt Levine<\/a> of Bloomberg states that buildings are like bonds: \u201cthere are a lot of them, they\u2019re all different, they each have idiosyncratic weird issues, the trade size tends to be in the eight figures, no individual building trades all that often\u201d, and that the facts about them are subjective, often not \u201cthat computer legible\u201d.<\/p>\n For these reasons and more, at VTS, we believe AI\u2019s biggest impact within CRE will not come from models that draft emails or provide rote answers to empirical questions<\/strong>. It\u2019ll come from far greater challenges associated with analyzing data streams with higher levels of abstraction, and empowering professionals to understand \u201cfacts\u201d about the real world.<\/strong><\/p>\n\n\n Among several capabilities, our models are trained to make real-world connections between <\/em> Technical advances in AI are quickly catching up with CRE industry needs, with various solutions being built to track, aggregate, and link information across a tenant\u2019s and a property\u2019s lifecycle. From de-anonymization (the practice of linking anonymous data to publicly available information), to entity disambiguation (the task of assigning unique identities to disparate records), and word embeddings (the representation of words in high-dimensional vector space) – these technologies power the domain-specific AI solutions that VTS has built internally to connect the dots across billions of data points and thousands of unique client integrations.<\/strong><\/p>\n\n\n Our data platform orchestrates flows of first-party data between our Lease Management<\/a>, Market<\/a>, Tenant Experience<\/a>, and Data<\/a> products so that we can derive insights not just on a single building, but also the relationships between them and understand the interaction between physical space and tenant behavior in the real world<\/strong>. For example, we can track for a unique tenant company:<\/p>\n\n\n By focusing on training domain-specific ML models to do single tasks (like de-anonymization and disambiguation) really well, we\u2019ve also unlocked the ability to produce unique insights about market supply and demand.<\/p> Take our tenant requirement disambiguation model, for example, which finds similar-looking deal logs that represent the same unique tenant touring in a specific market. The model leverages similar techniques to other AI applications, namely high dimensional vector embeddings, to find similar records across spatio-temporal attributes (in other words – <\/em>deal opportunities logged in similar locations and at similar times with similar broker or company attributes). The model currently performs at 92%+ accuracy to human-labeled data, and is on average, 3% off in demand prediction numbers<\/strong>.<\/p>\n\n\n Diagrammatic representation of how our model finds relationships across deal logs, using semantic, lexical, and spatio-temporal distances\u2014distances that are impervious to errors in data entry<\/em><\/p>\n\n\n This model is one element that powers our monthly VODI index<\/a> and is part of the broader VTS Demand Model<\/a>. We have employed similar techniques on company and property disambiguation tasks, whereby we match entities across thousands of systems to the same real world record. We do this by leveraging pre-trained word embeddings, and calculating a custom loss function to perform matching operations that overcome bad data entry and missing attributes <\/strong>(such as \u201cspace\u201d, \u201ccondition\u201d, \u201ctenant name\u201d, \u201ctenant industry\u201d).<\/strong><\/p>\n\n\n Snapshot of July 2024 VODI Index<\/a><\/em><\/p>\n\n\n\u201cThe revolution hasn\u2019t happened yet\u201d<\/strong><\/h3>\n\n\n
<\/figure>\n\n\n<\/figure>\n\n\nWhat we’ve done with artificial intelligence <\/h3>\n\n\n
\n
<\/figure>\n\n\n<\/figure>\n\n\nWhat\u2019s next<\/strong><\/h3>\n\n\n
![\"\"](\"https:\/\/www.vts.com\/wp-content\/uploads\/2024\/09\/Screenshot-2024-09-23-at-3.14.58\u202fPM.png.webp)
<\/source><\/picture>\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t