OpenAI paid more than 100 former bankers $150 an hour to teach a model how to build financial models in Excel — IPOs, restructurings, M&A — using Wall Street formatting conventions. One model per week. Structured like a training program. The explicit goal: reduce the grunt work of junior analysts by encoding it into software. I have been thinking about this through a different lens, because I am spending this summer as a Demand Planning Intern. Supply chain went through this exact transition. Not as a warning. As a preview.
THE SIGNAL
OpenAI's reported finance training program deployed more than 100 former bankers at $150 per hour to encode financial modeling workflows into AI systems built around Wall Street formatting conventions. A parallel effort is reportedly underway for consulting grunt work. This is tacit knowledge extraction: the deliberate encoding of professional judgment into replicable software. The distinction matters. Automation compresses time. Knowledge extraction shifts who owns the judgment.
WHY IT MATTERS TO YOU
Supply chain saw this first. ERP systems encoded the forecast-building expertise of experienced planners into automated workflows. Planner roles shifted from building the model to governing it — validating outputs, managing exceptions, overriding when the system was wrong. That transition happened over 20 years in the supply chain. OpenAI appears to be compressing the same arc into 3 to 5 years in banking and consulting. The roles that survived were the ones that understood the system well enough to run it.
THE SKILL OR TOOL
The exception manager is the role emerging in finance and consulting right now, without being named yet. It is the person who validates the AI output, catches the model's errors, and decides what it cannot touch. In the supply chain, that person is the demand planning analyst. In banking, it will be called something else, but the function is identical. Building toward it deliberately means two things: deep workflow knowledge in your target function, and AI fluency specific to that function's outputs. The Anthropic AI Fluency certification and the Goldman Operations Forage simulation are the two-credential starting stack. Both free. anthropic.com/academy and theforage.com.
THE QUESTION
If the model learns everything a junior analyst knows — the formatting conventions, the modeling structure, the workflow logic — what is the one thing you are building right now that the model cannot learn from a training set?
— Marco Meneses
Finance & Business Analytics @ Wilkes University | Greenwood Project FinTech Scholar
theanalystedgehq.com
Reading this on LinkedIn? Get it in your inbox every Thursday at theanalystedgehq.com