OpenAI is working on an internal effort that points directly at one of finance’s most repetitive knowledge-work pipelines: the financial models built by junior investment bankers. According to a Bloomberg report, the project is called "Mercury" and uses banking expertise to train AI systems on the mechanics of deal modeling.
The work is not described as a public product launch. It is an internal training project built around examples, feedback, and standard industry formatting, with the goal of helping AI produce financial models on its own.
What OpenAI Is Training The AI To Do
The Bloomberg report says OpenAI has hired more than 100 former bankers for the project. Their role is to use their financial modeling experience to teach AI systems how to handle work that junior investment bankers often spend hours completing.
The project focuses on repetitive tasks tied to financial models. Each week, participants build a model that simulates transactions such as mergers, restructurings, or IPOs. Those are the kinds of exercises that require structure, formatting discipline, and careful translation of assumptions into Microsoft Excel.
The process described in the source is practical rather than abstract. Participants write simple prompts, move the output into Microsoft Excel, and then refine the models based on feedback. The models follow standard industry formatting, which matters because banking work often depends not only on the calculation, but also on whether the work is organized in a familiar way.
Who Is Working On Mercury
The project draws on people with direct finance experience. Bloomberg reports that the team includes former employees from major firms like Goldman Sachs, JPMorgan, Morgan Stanley, Brookfield, Evercore, and KKR. It also includes MBA students from Harvard and MIT.
The participants are recruited through third-party vendors. They work flexible schedules and earn about $150 an hour. That setup suggests a training operation built around specialized expertise without requiring all contributors to become full-time OpenAI employees.
The work is also highly specific. OpenAI is not just asking finance professionals for general opinions about banking. It is asking them to produce concrete financial models, respond to feedback, and help shape examples that can be used as training data.
How The Recruiting Process Works
Recruitment for Mercury is described as almost entirely automated. According to Bloomberg’s source, candidates begin with a 20-minute interview conducted by an AI chatbot. After that, they complete knowledge and modeling tests.
The finished models are then reviewed. The feedback from reviewers goes directly into OpenAI’s training data, according to the report. That makes the evaluation process part of the learning loop: the AI is not only exposed to finished work, but also to corrections and judgments about that work.
This detail is important because financial modeling is not simply a matter of filling cells. The source describes a workflow where prompts, Excel translation, review, and refinement all matter. In that setup, the system is being trained on both the output and the process used to improve it.
Why Junior Banking Work Is A Target
The source frames Mercury around the repetitive groundwork handled by junior analysts. These tasks can be time-consuming, structured, and format-heavy. That makes them a natural area for AI training, at least as described in Bloomberg’s report.
The goal is to teach AI how to produce financial models independently. If successful, that could reduce some of the tedious work junior analysts currently perform. The source does not say that all junior banking work would disappear, and it does not describe a timeline for any commercial rollout.
What it does show is a clear direction: OpenAI is using domain experts to train models on specialized professional workflows. In this case, the domain is financial modeling, and the experts are former bankers and MBA students with relevant experience.
What OpenAI Says About Expert Training
An OpenAI spokesperson told Bloomberg that the company works with experts "across various domains to improve and evaluate our models’ capabilities." The spokesperson also noted that those experts are recruited and paid by outside vendors.
That statement places Mercury within a broader pattern of expert-assisted model development. The finance work described by Bloomberg appears to be one example of how OpenAI uses people with domain knowledge to improve and evaluate AI capabilities.
For investment banking, the significance is straightforward. A system trained on prompts, Excel-based models, transaction simulations, and reviewer feedback could become better at producing the structured work that analysts regularly prepare. The report does not claim that the AI already replaces bankers. It says OpenAI is training systems with that capability in mind.