The LEVERTON Guide to Legal AI

A Close-Up View of How AI Works for Data Extraction

A Guest Post by Legal AI + Prop Tech Company, Leverton

AI: The What and The How

Top companies all around the world now utilize AI more than ever before in building effective and efficient business tools.

There is no doubt that AI can do amazing things: for example, in the legal space, many leading firms are now utilizing AI-powered natural language processing and machine learning systems to extract useful data for lease extraction, due diligence, and compliance checking work.

AI systems can provide companies with more potent actionable insights, improving their decision-making processes and potentially, as well as their overall success. Though many want to give AI a try, few understand the process of how AI really works in business. And, how to get started with AI remains a mystery for many.

Common questions include how does AI actually work, how to train AI to be ‘intelligent’, how much time and data it takes to do so, how to increase accuracy, and if it is truly 100% machine or half machine-half people? People want real examples of how it works at a company in a step by step way.

A Helping Hand in AI

A prime example of harnessing AI to benefit businesses can be seen in Leverton, a legal AI and proptech company. At every company, decision making is a necessary process that relies on clear reasoning and often, uncovering, and analyzing data.

Data for any company therefore must be accurate, concrete, and clearly understandable. Relevant data is often hidden in the complicated language of corporate documents. Leverton helps its customers across a variety of industries to tackle this challenge.

They do this by providing an AI-powered data extraction platform, that applies deep learning/machine learning technology to extract and structure data from real estate leases, financial, insurance, and more corporate documents in 30 languages.

Diving in Deeper: A Step-by-Step Look at Leverton’s AI Platform

Leverton’s CEO Abhinav ‘Abe’ Somani provides a deeper look into how the AI at Leverton really works.

According to Abe, first, a document such as a lease agreement, a NDA, or other corporate or legal document, must be uploaded into the platform. Next up is processing: the document is then essentially turned into an image file, flattened, and then an optical character recognition (OCR) process begins.

The purpose of the OCR is to turn all of the uploaded ‘images’ into machine readable text. This allows for turning the visual text into a text file that can be read by the machine. Once the OCR is completed, the text is “read through” like a human using natural language processing.

Then, data extracting and organization begins: the machine then looks to find information that the ‘data model’ tells it to look for, such as a specific name, rent amount, due date, or term or condition. As the machine finds information, it returns the information to the corresponding data model fields. These could be pick lists, date fields, numbers, yes/no, or generic text fields.

The machine is calibrated to only return fields that it has X % confidence in returning (a threshold that is set by the company). This can be fine-tuned as needed. After the machine sends information back, human beings must validate its answers, correct wrong answers, or fill in missing information. This validation process is what continuously helps the machine get smarter, better, and more efficient.

The more Leverton emphasizes that a correct answer is correct, the more likely the machine is to return similar correct answers. The more we correct the incorrect answers, the more likely the machine is to avoid those incorrect answers in the future. A good pilot with Leverton includes somewhere between 500 and 5,000 documents, depending on the document type.

The amalgamation of doing this process thousands to millions of times creates a robust machine and deep learning library from which the algorithm can draw upon. After validation, this data can then be integrated with 3rd party ERP and BI systems, allowing for deeper data analysis and decisioning.

As AI evolves to become more robust and effective, it now appears more important than ever for companies to become ‘first movers’ in applying AI to their decision-making processes, as those who do so have the powerful potential to become market leaders. After all, AI is here to stay — and now is the time to get started.