AI can make finance work faster, but mostly around the model rather than instead of the model. The useful gains come from structuring assumptions, preparing scenarios, explaining changes, drafting commentary, and reducing repetitive reporting work. The dangerous version is pretending a model is still robust when no one can explain how the numbers were produced.
Use AI around the model before you trust it inside the model
Most businesses should start by using AI around financial workflows rather than allowing it to silently drive the numbers. That means helping with scenario framing, assumption capture, variance commentary, narrative summaries, and first-pass reporting packs before it takes on anything that needs finance-grade traceability.
that means spreadsheet-heavy work still lives in Excel or Sheets, while ChatGPT or Claude help accelerate the explanatory layer and the repetitive prep around it.
The best-fit finance workflows are repetitive and explainable
AI works best where the team can still explain every important assumption and quickly sense-check the output. If a workflow becomes hard to audit, it is not a good first candidate.
- Drafting scenario narratives and assumptions before the numbers are finalised
- Preparing board and leadership commentary from model outputs and actuals
- Turning recurring reporting inputs into first-pass packs and update summaries
- Explaining formula logic, structure, and model changes for internal clarity
A good-fit versus bad-fit test
The right use case improves speed without obscuring accountability. The wrong one makes the workflow faster and less trustworthy at the same time.
- Good fit: scenario comparison, commentary drafting, assumptions logs, recurring report preparation
- Good fit: internal explanation of model structure, formula logic, and options under review
- Bad fit: unreviewed pricing logic, contractual numbers, or financial outputs no one can trace
- Bad fit: any workflow where the team cannot explain the source data or validate the assumptions
Keep the finance owner and the audit trail visible
AI can make reporting and modelling work more efficient, but only if the workflow still has an owner, a review point, and a clear record of what changed. If the business cannot show where the inputs came from, which assumptions were accepted, and who signed off the output, the efficiency gain is not worth much.
- Keep the source workbook, assumptions, and outputs clearly separated
- Use AI to prepare commentary, not to hide weak logic
- Retain named human sign-off on anything that leaves the business
- Treat explainability as part of the workflow, not an optional extra