AI-Powerd Contract Analysis in Shareflex

AI-Powered Contract Analysis in Shareflex

Shareflex Contract is an excellent system for managing both incoming and outgoing contracts. It includes convenient lists that give you quick insight into contracts that are about to expire or will be automatically renewed.

In this post, we show you how our AI-driven contract solution can answer complex questions about your contracts.

Examples of questions the system should be able to answer

  • Identify contracts with penalty clauses and explain the costs in the worst-case scenario.
  • Highlight contracts that could be harmful to the company and explain the risks.
  • Identify existing NDA agreements and provide a summary of their content.

Manually analyzing such questions can be time-consuming for a contract manager. This study explores whether AI can perform this task faster and more effectively.

Demo

Search for contracts with penalty clauses and analyze what the possible worst-case costs could be.

Analysis of the result

The system can instantly scan hundreds of contracts to find those with penalty clauses. It then analyzes the contract text (from the PDF) and provides a clear answer to the worst-case question. Quite impressive.

How it is built

Tools

Shareflex Contract is based on SharePoint and runs entirely in the cloud. The AI analysis tools also operate in the cloud, making it relatively easy to implement this system. There are different ways and tools to achieve the same result. In this setup, we use the following cloud tools:

  • Make.com
  • OpenAI
  • CloudConvert
  • Pinecone
  • Retool

Pinecone vectors

To quickly find contracts based on a user’s question, we first summarized the contracts with OpenAI. These summaries are then converted into vectors and stored in a vector database (Pinecone). The Make scenario for this process is shown below. Each time a contract is saved in Shareflex Contract, a Pinecone record is automatically created in the background.

Analyzing and generating answers

The process works as follows: a user’s question is first cleaned and converted into a vector, which is then used to retrieve the most relevant contracts from the database. These contracts are evaluated for relevance, and only the suitable results remain. The retrieved contracts are then summarized and analyzed, so the user receives a clear and well-founded answer to their question.