Generative AI for Financial Term Sheet Generation

In drafting loan agreements at financial institutions, the key task is to accurately extract necessary information from call records, company documents, and financial statements and organize it into a Term Sheet. AI21 presented a two-stage flow that combines Contextual Answers and Jamba-Instruct to automate this process.

• Information Extraction: Contextual Answers scans call records and company documents to identify and extract relevant clauses, figures, and conditions.
• Draft Generation: Based on the extracted information, Jamba-Instruct creates a draft Term Sheet with the necessary sections and clauses.
• Review and Supplementation: Jamba-Instruct re-examines the draft to find missing items and revises it as needed.
• Final Draft Creation: The feedback and original draft are combined to produce a final Term Sheet that preserves key figures such as loan amounts while newly supplementing missing key conditions—such as disclaimer clauses—in a context-appropriate way.

This approach is focused on reducing repetitive work for reviewers and lowering legal and financial risks caused by omissions or errors. It also emphasizes that in domains like finance where accuracy and contextual understanding matter, it's important to use purpose-fit tools together. A code notebook is provided alongside to show the step-by-step implementation, and Jamba-Instruct is currently available in public preview.