Streamlining Lease Abstraction with AI
Aug 28th, 2023
In the fast-paced realm of real estate and commercial property management, staying ahead of the curve is essential. One of the most critical yet time-consuming tasks in this domain is lease abstraction — the process of distilling complex lease agreements into concise and structured data points. Traditional lease abstraction methods involve manual labor, often consuming countless hours and leading to potential errors.
In this tutorial, we’ll delve into how to automate lease abstraction using a custom-trained AI model in conjunction with a Large Language Model (LLM). We’ll unravel the intricacies of transforming lease abstraction from a labor-intensive chore into an efficient, accurate, and streamlined process. Whether you’re a real estate professional, property manager, or tech enthusiast, this tutorial will empower you with insights into harnessing AI’s potential to reshape the landscape of lease management.
Throughout this journey, we will cover the following areas:
- Custom AI Training: Discover the nuances of training a custom AI model tailored specifically for lease abstraction. Learn how to effectively label a training dataset and train a custom AI model.
- Leveraging Large Language Models (LLMs): Delve into the capabilities of LLMs and comprehend how they complement custom AI models, elevating the accuracy and depth of lease abstraction.
- Creating Custom Workflows: Combine our AI models and LLMs into one unified workflow using Kudra. Experience the transformation of arduous manual abstraction into a seamless automated process.
Whether you’re aiming to optimize your property management practices or keen to explore the intersections of technology and real estate, this tutorial serves as a valuable starting point to learn how to automate lease abstraction through advanced AI integration. Let’s get started!
Custom Model Training
To train our model on lease comprehension, we need to provide it with exmaples of labeled leases to learn from. Fortunately, there is a publicy annotated dataset that we can use to train our model. The entities that we are interested to extract from the leases are:
LEASED_SPACE
LESSOR
LESSEE
CLAUSE_NUMBER
CLAUSE_TITLE
DEFINITION
SUB_CLAUSE_NUMBER
START_DATE
END_DATE
REDFLAG
TERM_OF_PAYMENT
SIGNING_DATE
DESIGNATED_USE
EXTENSION_PERIOD
NOTICE_PERIOD
VAT
TYPE_LEASE
EXPIRATION_DATE_OF_LEASE
INDEXATION_RENT
SUB_CLAUSE_TITLE
ANNEX
DEFINITION_NUMBER
DATE
NUMBER
TITLE
SPACE
CLAUSE
OF
PERIOD
USE
To train the model, we upload the 123 annotated documents to UbiAI Text Annotation Tool. The advantage of using UbiAI, is we can modify the labeled dataset as we wish and launch multiple model trainings on demand without any code required.
Annotated lease dataset in UbAI
For this tutorial, we are going to fine-tune Roberta, a state-of-the-art transformer model, on our annotated dataset. To launch the training, head to the Models menu and select roberta-base from the drop down menu and click Run:
UbiAI Model Training Dashboard
The training will typically take couple of hours. Once trained, we can use the model to extract important clauses from leases.
Lease Summarization
One important aspect of the lease abstraction process is to condense the multi-page lease into a concise one-page summary that is easily comprehensible. While summarization aids in making the document more digestible, it can cause us to overlook crucial clauses concealed within the document. By integrating our trained custom AI model with a summarization process, we can achieve a comprehensive overview of the lease without sacrificing any essential details.
To accomplish this, we will utilize kudra.ai to establish a workflow that seamlessly integrates our custom AI model and a summarization engine, all without requiring any coding.
Below is the entire workflow created in few minutes:
Kudra workflow building interface
Here is how it works:
- The first component in the workflow “Import Text” imports our PDF file, it supports both digital PDF andscanned PDF, and feeds it to the Custom Entity Model component.
- Within the custom entity model component, we select our custom trained AI model.
- Summarizer: Here, kudra.ai allows you to call OpenAI’s GPT 3.5 model and create your own prompt for summarization as shown below:
Kudra chatGPT editor
The variable [[input_text]] contains the content of the PDF which is concatenated to the prompt and sent to theGPT model to summarize it.
Automated Lease Abstraction
We are now ready to launch our custom workflow! Simply upload a test lease and press on the Run button in the kudra app. The output of each component will be visible in the review page.
Our custom AI model has identified numerous entities, including one red flag and two term payments:
Moving on to the summarization component, below is the summary of the lease:
The summary provides a brief description of the lease with the identification of some important clauses. It is worth noting that the content of the summary can be tuned by engineering the prompt right from the kudra interface.
Conclusion
Real estate and commercial property management is evolving rapidly, demanding innovative solutions to streamline complex processes. Lease abstraction, a pivotal task within this domain, has traditionally been labor-intensive and error-prone.
By combining a custom AI models and LLMs, we’ve explored how lease abstraction can be automated effectively. Powered by kudra.ai, our seamless workflow showcases the potential of AI in real-world applications, enabling professionals to achieve greater accuracy, efficiency, and insight. The transformation from time-consuming tasks to agile processes signifies a paradigm shift in property management practices.
Whether you’re a seasoned real estate professional, a diligent property manager, or an enthusiastic technologist, this tutorial equips you with the knowledge to leverage AI’s capabilities in lease management.
If you are looking to automate data extraction from lease documents, please schedule a demo today!