Transforming Loan Underwriting with AI-Driven Contract Analysis

Summary

Our client is a leading loan origination company. They faced challenges in determining cancelability in their underwriting contract analysis process.

Their manual workflow for analyzing borrower contracts was time-consuming and prone to errors. Underwriters often lacked legal expertise, resulting in frequent escalations to the legal team.

They partnered with us to automate this process. We developed an AI solution using OCR and LLMs to determine the cancelability of contracts.

Now, the company uses our AI agent with 100% workflow automation, reducing the time from hours to just 3 minutes per contract. They were able to significantly enhance efficiency and accuracy in their underwriting process. They plan to use this solution more broadly in their business.

‍
Labor-Intensive Contract Analysis Hinders Underwriting Efficiency

Like most loan origination companies, our client requires prospective borrowers to submit current and future job contracts as proof of future income. These contracts are used as collateral and are crucial for making adequate underwriting decisions. 

Our client, however, struggled with analyzing them for cancelability. The process usually involved multiple people, was time-intensive, prone to errors, and—above all—notoriously difficult to automate.

• Manual contract analysis is time-consuming and error-prone

Underwriters spent hours reviewing each contract manually. This task could take up to several hours for a 50-page document or about an hour for an average 8 to 10-page contract. The manual process was tedious and prone to human error, potentially causing critical contract details to be overlooked.

• Frequent escalations to the legal team causing further workflow delays

Underwriters often needed more expertise to determine the cancelability of complex and legally nuanced contracts. This resulted in frequent escalations to the legal team, creating bottlenecks and further slowing down the overall underwriting process.

• Contract analysis can't be automated without AI solutions due to a lack of semantic understanding of the text

The client couldn’t automate their workflow because traditional non-AI systems could not understand the context and semantics of the text. 

Systems without semantic understanding can’t differentiate the context in which certain words are used. For example, the word “canceled“ can have different meanings in different contexts and doesn’t always indicate that a contract is easily cancellable. However, non-AI systems can’t distinguish the difference and would flag every contract with that word.   

These challenges highlighted their need for automation, so the client partnered with us to develop an advanced AI-driven solution. They wanted to streamline the contract analysis process, reduce errors, and improve efficiency in their underwriting workflow.

Custom AI Solutions Streamline Contract Cancelability Analysis

We automated contract cancelability analysis using Decision AI, our AI agent that excels in error detection. It uses algorithms to make accurate decisions and ensures compliance by cross-referencing data with internal or official databases.

In this case, Decision AI evaluates whether a contract is cancelable based on semantic understanding and provides the reasons for that determination. If a contract is not cancelable, it holds substantially more weight in the underwriting decision as it guarantees future income.

The client provided 10 example contracts, which we used to train the AI agent. The AI agent also had to learn what constitutes cancellable language through in-context learning to improve its accuracy and decision-making capabilities.

Here are the exact steps we took:

1. Perform Optical Character Recognition (OCR) on the document:
   

We used OCR to accurately extract text from contract documents, ensuring the system could read and process each contract efficiently.

2. Perform an LLM call to extract sections related to the contract’s cancelability:
   

We employed an LLM because it understands the text semantically. Semantic understanding allows LLMs to comprehend the meaning and context of words, phrases, sentences, and larger bodies of text. 

Unlike non-AI systems, they can understand the underlying concepts, relationships, and nuances of the language being used. For example, they can understand if "cancel" and "contract" are related, or infer when a contract can be canceled without penalties based on the overall context.

3. Perform a second LLM call to analyze the extracted sections:
   

The AI agent analyzed the relevant sections to determine the contract's overall cancelability and provided a reason for its decision.

4. Return a response in JSON format:
   

We chose JSON, an open standard format with human-readable text, to store and transmit data. It allowed us to organize extracted data efficiently.

The final output is a structured JSON response that includes:

• Decision
• Reason for decision
• Relevant sections of the contract

Client’s workflow post-implementation:

1. The client’s prospective borrower submits current/future contracts as proof of income.
   

2. Our AI agent extracts and organizes relevant information using OCR and an LLM.
   

3. Information is structured into a JSON response, making it easy to integrate into the client’s system.

Automated Contract Analysis Revolutionizes Underwriting Efficiency

Our AI solution completely transformed the client's contract analysis and simplified decision-making processes. We achieved 100% automation in contract cancelability determination. This change reduced the analysis time from several hours to just 3 minutes per contract.

Manual underwriting is labor-intensive and error-prone. Decision AI eliminated human errors and provided a transparent view of how it made its decisions. These significant improvements accelerated the client’s entire business operation.

By streamlining these processes, the client now enjoys greater accuracy and increased efficiency in their underwriting workflow. As the client reviews the next steps, they plan to use this solution more broadly in their business.

The successful implementation of this AI solution sets a new standard in the loan origination industry. It demonstrates the potential to revolutionize contract analysis, offering a robust, automated model that other companies facing similar challenges can adopt. 

This project exemplifies how advanced AI technologies can transform workflows, enhance productivity, and increase customer satisfaction even across highly regulated industries.