The Method
Five moves for putting AI to work in finance
Most AI pilots in finance stall for the same reason. They start with a tool instead of the work, and no one can say what a finished, defensible result is supposed to look like. These five moves are how I teach finance teams to close that gap, and the same sequence runs underneath the enablement work I do with them. It is deliberately boring where it needs to be, because the numbers have to hold.
1. Map it
Frame the work as a journey. Where the team is now, where it is going with a worked example of the destination, and the route between them. A team that cannot describe the finished artifact is not ready to automate the path to it.
2. Split it
Separate the deterministic work from the non-deterministic. AI writes the formula or the code for the parts that have one correct answer, and performs only the judgment that genuinely needs a model. This is where most finance workflows are won or lost.
3. Fuel it
Give the model the minimum necessary context, delivered as a clean folder, and nothing more. Most bad output is a data problem wearing a model costume.
4. Guard it
Run the red-lines check before anything sensitive moves. The right data in the right tool tier, scoped access, approvals on, and an audit trail that survives a review.
5. Review it
Close with the five-point human review. Verify every source, own every conclusion, strip the AI writing patterns, trace every number back to its origin, and reconcile. The person who signs the memo still owns the memo.
The tool is not the hard part
A finance team can buy every model on the market and still produce work no controller would sign. The difficulty was never access to the tools. It is knowing which parts of a close, a forecast, or a technical accounting memo a model should touch, which parts it must not, and how to prove afterward that the number is right. A method answers that the same way every time, so the result does not depend on who happened to run it.
The five moves are platform-agnostic on purpose. They work whether your team lands on Google, OpenAI, or Anthropic, because the sequence is about the work, not the vendor. That matters in a market where the leading tool changes every few months.
Teaching Professor of Finance, Santa Clara University · ex-Google Cloud · CPA, MBA
The five moves come out of the finance teaching and the AI enablement work Devon runs with finance and accounting teams.
See it, learn it, or run it with your team
The full method is taught free, with hands-on labs and worked examples, inside the Applied AI for Finance course. The same method is the backbone of the enablement work I run inside finance and accounting teams. You can start wherever is useful.