Horizontal and vertical AI both have their places in our professional lives. Below, we make the case for vertical AI adoption in CRE and highlight some of the competitive and performance boosting benefits of AI.
AI has made itself known in recent years, popping up with a variety of use cases. Whether we use ChatGPT to automate a task, digitize a physical copy of a document, or order our groceries while standing in the kitchen, AI exists today in a variety of professional and personal capacities.
Many of these tools like ChatGPT and Dall-E 2, companies like IBM, or products like Alexa fulfill vastly different needs with their AI. They all have something in common, though, they are examples of horizontal AI. These machine learning models have much broader use case applications compared to tools characterized as vertical AI, like Prophia.
Both horizontal and vertical AI have their strengths and weaknesses. But for CRE, an industry with a frontier’s worth of unexplored, proprietary data, Prophia, and other vertical AI PropTech companies, provide the appropriate niche and focus required for processing data at scale.
In the piece below, we explore vertical and horizontal AI and the ways in which Prophia’s generative models stand to change the way CRE pros process their portfolio insight, track trends, and interact with data.
Jump To A Section in This Article
Horizontal Vs Vertical AI
Domain Expertise Differences Between Vertical and Horizontal AI
Development & Deployment Differences
Strengths of Vertical AI
Weaknesses of Vertical AI
Final Analysis: Finding the Right AI Fit
Just like vertical and horizontal SaaS, vertical and horizontal AI describe two different applications of AI. Both of these AI types are designed to solve problems. However, where they differ is within their scope and focus. For instance, where horizontal AI might supply general-purpose intelligence across a number of domains, vertical AI performs more specialized functions.
Dall-E (Dall-E 2), for example, is a horizontal AI, it has broad image generation capabilities that can be used in a variety of industries such as graphic design, 3D design, brand marketing, product design, etc. The technology is extremely deft in image creation but it does not serve one specific industry, thus it is considered horizontal AI.
On the other hand, Prophia’s AI is vertical, it performs functions for a specific domain: CRE. While its capabilities are very broad, like what one might discover in a horizontal AI solution, its AI would not be able to perform tasks required by an industry like design.
Since vertical AI is tailored for specific tasks or domains, it often requires deep expertise and domain knowledge. For Prophia, this meant accessing approximately 100,000 proprietary CRE documents to build a set of training data for the first iteration of our technology. The success of this exercise, which began approximately five years ago, has resulted in Prophia’s vertical AI capable of recognizing over 200 commercial real estate terms found in standard leases and the ability to automate CRE tasks.
Horizontal AI, on the other hand, aims to achieve a broad understanding of intelligence and learn generalized patterns that can be applied across various domains. While it may still require expertise in AI and machine learning, the emphasis is more on developing general-purpose algorithms and models that can be applied to multiple problem areas.
A great example of this AI scope is IBM, a multinational technology corporation with very powerful, but broad AI capabilities with a myriad of business applications. While IBM’s AI products can conceptualize terms found in large documents, their generative AI, watsonx.ai, has not trained exclusively in one business vertical. Therefore, it cannot craft a lease abstract with the same contextual knowledge as Prophia, which was trained exclusively on CRE documentation.
In terms of development, both AI approaches require an understanding of data science at a very high level. However, developing vertical AI solutions often involves a more targeted approach and the data used for training are tailored to the particular use cases. Deployment of vertical AI systems is usually limited to the specific domain for which they were designed.
We’ve been training Prophia on CRE data since 2019. Our training data model has reached approximately 100,000 documents and today the AI recognizes over 200+ CRE-specific terms found in proprietary lease documents. This makes our technology incredibly powerful when processing CRE documents and capable of quickly abstracting CRE data into lease abstracts or generating a dynamic stacking plan. But this level of development and automation capabilities are challenging, if not impossible, for horizontal AI.
Building horizontal AI systems often involves a more comprehensive and generalized approach. The development process focuses on building models and algorithms that can be applied across multiple domains. These systems are designed to be adaptable and can be deployed in various contexts, allowing for greater flexibility in their application.
For an industry like CRE, an accurate and comprehensive data set is required for boosting the performance of a portfolio. This accuracy is one of the greatest advantages of vertical or “narrow” AI. Unlike horizontal systems, vertical generative AI produces outputs with greater accuracy because the model’s training data is limited to a singular, focused subject.
Imagine vertical AI like a lighthouse and horizontal AI like stadium lights. A horizontal solution can illuminate large themes in your data, but overwhelmingly, reporting and output will lack the nuance of an AI model trained in a specific field or domain. This lack of industry nuance, in turn, can make an output from a horizontal AI less accurate than a system trained on industry-specific or proprietary data.
Due to its narrow scope, vertical AI is able to build a precise and accurate understanding of its data ecosystem. This level of accuracy is key in an industry like CRE where confidence in portfolio data ultimately leads to the ability to secure capital, satisfy tenants, and improve existing contracts.
Vertical AI is designed to excel in a specific domain or task—i.e. Prophia in CRE. This orientation to the market allows us to hyper-focus on customer type and design around use cases accordingly, a level of specificity that is not possible with horizontal AI.
Prophia takes this specialization even a step further with its AI by threading in a crucial human element. With every building upload, Prophia’s human CRE experts perform manual due diligence to ensure every term and concept captured in our data synthesis is accurate. This allows Prophia to solve for the accuracy challenges in digitization that Philip Russo recently raised on the Commercial Observer.
Since vertical AI is optimized for specific tasks or domains, it can achieve exceptional performance within its specialized area. The models and algorithms used in vertical AI systems are finely tuned and trained on relevant data, allowing them to deliver precise and accurate results. This high performance makes vertical AI suitable for critical applications where precision is crucial, such as portfolio management in CRE.
Narrow AI systems focus their resources and computational power on a specific task or domain. This targeted approach allows for efficient resource allocation, as the algorithms and models can be optimized for the specific problem at hand. By utilizing resources effectively, vertical AI can deliver faster processing times and reduced computational requirements compared to more generalized approaches.
On the topic of the market advantages of this domain specificity, tech executive, Gokul Rajaram, elegantly states, “...vertical AI apps don’t face exceptional incumbents, need more domain expertise leading to fewer startup competitors, have a unique data advantage and focused GTM, leading to higher chance of winner take most. Even seemingly narrow niches of a vertical are absolutely massive opportunities. Easier to justify an early stage investment.”
Unlike horizontal AI, vertical AI often requires deep understanding and integration of domain-specific knowledge. This integration enables the AI system to leverage existing expertise and knowledge in the domain, leading to more accurate and contextually relevant outcomes.
In more standard examples of AI development, narrow or vertical AI is faster to train, develop, and deploy than more generalized approaches. This is largely due to the ability to program with the focus on a specific problem or task, allowing engineers and data scientists to streamline the deployment process. In the case of Prophia, the private nature of CRE data made the training and development process slower than standard AI development, but the calculated approach really paid off.
Slowly growing Prophia’s training data from 10,000 to 100,000 proprietary CRE samples allowed the AI to gradually mature and adequately prepare the models to face innumerable variables found in CRE. In the years, deployments have naturally become more streamlined and today Prophia’s vertical AI is perfectly optimized to handle specific business and admin tasks related to CRE, from lease abstraction to critical date reports.
Additionally, vertical AI solutions can be customized to fit the specific needs of a domain or organization. Developers have the flexibility to fine-tune the models and algorithms to address the unique requirements and challenges of the targeted application. This customizability enables vertical AI systems to adapt to specific use cases and deliver tailored solutions that align with the specific goals and constraints of the domain.
Vertical AI is designed to excel in a specific domain or task, which means it may struggle to generalize its knowledge and skills to new or unfamiliar situations. The specialized nature of vertical AI systems limits their ability to adapt and apply knowledge beyond their intended scope. This lack of generalization can make it challenging to use vertical AI in scenarios where flexibility and broad applicability are required.
In the case of Prophia, the AI has been designed to effectively handle a variety of the strategic and administrative tasks CRE professionals encounter everyday. And this is exactly the role vertical AI systems are meant to play in a professional capacity.
This specificity however, can result in limited versatility, as the AI may not be easily adaptable to tasks outside of its scope and domain. Modifying or repurposing a vertical AI system for a different application may require significant retraining or redevelopment, making it less agile compared to more generalized approaches.
Vertical AI heavily relies on domain-specific knowledge and training data. To achieve high performance, the AI system needs access to large and representative datasets that accurately reflect the target domain. Acquiring and curating such data can be challenging, especially in specialized domains where data availability may be limited. Additionally, the need for domain expertise to build and maintain vertical AI systems adds complexity and resource requirements.
For Prophia’s purposes, this data dependency is actually the source of Prophia’s strengths. The AI’s exposure only to CRE documents has made it incredibly versed in the nuances of certain legalese, contractual terms, tenant rights, and more. Ultimately, this specificity has made Prophia an incredible asset for CRE professionals.
Vertical AI systems trained on domain-specific data may be susceptible to biases present in the training data. Biased data can result in biased predictions or decisions, potentially leading to unfair or discriminatory outcomes. Moreover, the narrow focus of vertical AI increases the risk of overfitting, where the AI system becomes overly tuned to the specific training data, compromising its ability to handle variations or edge cases that differ from the training distribution.
Prophia has specific guardrails in place to protect our AI from bias and overfitting. In our training process, we separate data into a training set and a validation set. This validation set does not participate in training, and it can be used to validate that training does not slip into overfitting.
Additionally, since compiling our training data set in 2019, we have established a selection process that allows us to eliminate data that is too similar from our ongoing data training. This allows our model to avoid certain bias and promote greater accuracy.
The current state of the CRE market and the industry-agnostic trend of AI adoption puts CRE professionals in an interesting spot. Many firms still use manual and antiquated practices to manage data, but unfortunately, this approach just isn’t cutting it anymore. The level of human error and precious time spent on lease abstraction or aggregating data from disparate sources for a report will not stand up to the type of speed and precision offered by AI.
What’s more, we believe AI should enhance the capabilities of CRE professionals, not replace them. In order to do that, the individuals and teams working everyday with portfolio data need a tech solution capable of detecting the same level of nuance and contextual knowledge they’ve gained from years of on-the-job experience.
Thus, the domain specificity of a vertical AI solution, like Prophia, can offer CRE professionals the level of data control and accuracy they need to make ROI-generating decisions and stay on top of a market influx. After nearly five years developing our AI using proprietary CRE data, the current iteration of Prophia is leading CRE firms in lease abstraction automation, rent roll reconciliation, customizable portfolio reports, as well as stacking plan digitization. Access to this clear, standardized, and distilled data will give firms who chose to optimize tasks with vertical AI a competitive advantage, greater business intelligence, and relief from economic pressure.
If any of these data struggles ring true for you and your team, it might be time to integrate a vertical AI solution into your techstack. Get started by exploring Prophia’s AI solution in under a minute in the video above, or reach out to a member of our team to schedule a personalized demo.