OpenAI Trains AI to Handle Entry-Level Banking Tasks

OpenAI is spearheading an ambitious internal initiative, codenamed Mercury, designed to automate repetitive tasks typically handled by junior investment bankers. According to a Bloomberg report, the company has enlisted over 100 former bankers who are leveraging their financial modeling expertise to train AI systems.

The team comprises ex-employees from major firms including Goldman Sachs, JPMorgan Chase, Morgan Stanley, Brookfield, Evercore, and KKR, as well as MBA graduates from Harvard and MIT. These professionals are hired through third-party vendors, work on flexible schedules, and earn approximately $150 per hour. Weekly, they develop financial models simulating transactions such as mergers, restructurings, or initial public offerings (IPOs). They use simple prompts to convert outputs into Microsoft Excel, refine the models based on feedback, and ensure all work adheres to standard industry formats.

The recruitment process for this project is almost entirely automated. Candidates are first interviewed by an AI chatbot for approximately 20 minutes, followed by assessments on knowledge and modeling skills. The financial models they produce are evaluated, and feedback from reviewers is fed directly into OpenAI’s training datasets. This iterative process aims to teach the AI how to independently generate financial models, potentially eliminating the need for junior analysts to perform time-consuming, repetitive tasks.

An OpenAI spokesperson stated to Bloomberg that the company collaborates with “experts across various fields to enhance and evaluate our models’ capabilities,” noting that these specialists are hired and compensated by external vendors.