OpenAI has embarked on an ambitious mission to overhaul one of the most gruelling aspects of investment banking life — the relentless grind of data modelling and pitch-book polishing. Its new experimental venture, Project Mercury, is designed to automate the spreadsheet marathons and late-night revisions that have long defined the early years of a banker’s career.

To drive this initiative, the American AI firm has assembled an elite corps of over 100 former investment bankers from institutions such as JPMorgan, Goldman Sachs and Morgan Stanley. These professionals are reportedly earning around £120 an hour as they build advanced financial models and, crucially, train OpenAI’s systems to replicate the analytical intuition and precision underpinning complex corporate transactions — from IPOs to leveraged buyouts.

Far from a simple data-entry task, the recruits are also tasked with providing rigorous critique of the AI’s outputs, teaching it to interpret ambiguity and nuance — the hallmarks of human financial reasoning.

In a typically futuristic twist, recruitment for the project is itself AI-driven. Prospective contractors sit through a 20-minute automated interview before taking a series of modelling and valuation tests. Each week’s submissions are reviewed by human experts, transforming traditional banking work into a high-stakes feedback loop for machine learning.

Yet not everyone is applauding. Some senior bankers warn that outsourcing these ‘baptism by fire’ tasks could deprive junior staff of essential training grounds that historically shape tomorrow’s managing directors.

For OpenAI, Project Mercury is more than an efficiency play — it marks a strategic move into professional services such as finance, consulting and law, signalling a deeper push toward commercial adoption of generative AI.

Whether this initiative remains a tool for elite institutions or becomes the catalyst for a wider industry transformation may well determine the next chapter in the Square Mile’s evolution.