Project Mercury recruits ex-bankers for data and model training

OpenAI has recruited over 100 former investment bankers from Goldman Sachs, Morgan Stanley, and JPMorgan Chase to help train its artificial intelligence on how to build financial models, according to Bloomberg.

The goal is to replace the long, painful hours of manual modeling work done by junior analysts across Wall Street, and these bankers are part of a hush-hush internal project called ‘Mercury,’ which is paying them $150 per hour to feed the system real data, structure, and logic used in dealmaking; everything from IPOs to restructurings to leveraged buyouts.

The project gives these contractors early access to OpenAI’s finance-focused models designed to replicate the daily tasks analysts perform in Excel and PowerPoint. It’s reportedly one of the startup’s biggest pushes yet into corporate workflows, part of a wider effort to make its tools practical for industries that depend on deep quantitative work.

Even with a $500 billion valuation, OpenAI has not yet turned a profit, so every experiment like this is meant to prove its commercial muscle. Weird?

Project Mercury recruits ex-bankers for data and model training

Inside Mercury, former dealmakers write prompts and build complete financial models, simulating how real analysts perform their jobs. They’re asked to produce one model every week.

Each submission goes through a reviewer who provides feedback before it’s accepted into the system. The contractors operate independently, and all payments and management are handled through third-party suppliers, OpenAI said in a statement.

The job doesn’t even start with a person. Applicants go through a 20-minute interview with an AI chatbot that pulls questions directly from their résumés. If they pass, they face a test on financial statements, and finally a live modeling challenge to prove they can build clean, professional spreadsheets. Once accepted, they follow detailed instructions: keep formatting consistent, italicize percentages, and maintain exact margin sizes like in actual bank models.

A person involved said participants come from firms such as Brookfield, Mubadala Investment, Evercore, and KKR, with a few current Harvard and MIT MBA candidates joining the mix. Every model they create follows Wall Street standards so that OpenAI’s systems can learn not only the math but also the professional style bankers are expected to use.

The company’s spokesperson said OpenAI collaborates with experts “to improve and evaluate the capability of our models across different domains,” emphasizing that all contributors are vetted professionals. The idea is to use real experience to teach the AI how to manage deal workflows without endless trial and error.

Typical analysts at investment banks work 80 hours a week building Excel models and PowerPoint decks while handling last-minute change requests from their superiors—the same culture that inspired the “pls fix” meme on Wall Street. With this project, those repetitive tasks could soon be handled by algorithms.

Join a premium crypto trading community free for 30 days - normally $100/mo.