A finance analyst in 2026 should learn four things about AI, in this order: how to get a model to do the first draft of the narrative around the numbers, how to use it to write the SQL and the spreadsheet formulas you currently ask someone else for, how to check a model's arithmetic and sources before anything with a number in it leaves your desk, and where the line is on confidential data. Python is optional. Judgement is not. The analysts who are being promoted are not the ones who learned to code; they are the ones whose month-end pack now takes two days instead of five and whose commentary the CFO trusts.
This is written for FP&A, management-accounting and business-analyst roles. For the quant and data-science end of finance, the Data Scientist path is the right one.
Is AI actually useful in finance work, or is it a hallucination risk?
Both, and the split is precise: it is very good at language about numbers and unreliable at the numbers themselves. A frontier model will turn a variance table into a clear paragraph of commentary, draft the board narrative from your bullet points, summarise a hundred pages of a contract or an accounting standard, and explain a formula you inherited. It will also, with total confidence, add a column wrong, invent a figure that is not in the source, or apply last year's rate. Every useful workflow below is built on that split: the model writes, you compute; the model drafts, you check.
The tools inside the products you already use — Excel's Copilot, Google Sheets' Gemini, the AI features in Xero, NetSuite and the planning platforms — have the same strengths and the same failure modes, just closer to the data.
What should a finance analyst learn first?
The narrative. Month-end commentary, board papers, budget notes and the "what happened and why" email are the largest block of writing in the job, and a model does the first draft of all of them well if the brief is right. The brief is: the audience, the decision they need to make, the table or figures pasted in, the two or three drivers you already know matter, and the length. "Write commentary on this variance table" produces boilerplate. "Write 150 words for the CFO explaining the A$340k adverse variance in logistics, leading with the fuel surcharge, noting the volume effect is timing, in the tone of the attached last-month note" produces something you can send after one edit.
Practise on last month's pack, where you already know the answer, and compare. Keep the prompts that worked. That document becomes the team's standard.
Should a finance analyst learn to code?
SQL, yes — a month of evenings, and it pays back immediately. Python, only if you find you want to go further. The wall every analyst hits is the same: the data is in a system and the person who can get it out has a backlog. With SELECT, WHERE, GROUP BY, a join and a date function — and a model that will write the query when you describe the question in English — you stop waiting. The SQL and Data for AI course is built for this person and assumes no programming.
The same applies to spreadsheet formulas and Power Query. Describe what you want in plain language, let the model write the formula, then test it on a row where you know the answer. Analysts who do this stop being limited by which functions they remember.
How do you check AI output before it goes to the CFO?
Assume one number is wrong and find it. Three habits catch nearly everything:
- Every figure in the draft traces to a cell you gave it. If it does not, delete it. Models fill gaps with plausible numbers.
- The model does not do the arithmetic. Compute totals, variances and percentages in the spreadsheet; paste the results in; ask for the words. A model asked to "calculate" will sometimes get it right and you will not know which times.
- Ask it what it is unsure about. It will tell you, and the list is usually where the errors are.
This is the skill that makes you valuable rather than replaceable, because it is the one the tool cannot do for itself. It is also increasingly what interviews test — how to tell if a candidate can actually use AI is the hiring manager's version.
What must never go into an AI tool?
Anything price-sensitive, anything under a confidentiality clause, and anything about a person. For a listed company that includes unreleased results, forecasts and deal information — putting them into a consumer AI tool is a disclosure to a third party, and the continuous-disclosure and insider-trading rules do not have an exemption for chatbots. For everyone, it includes payroll, individual performance data, and customer or supplier terms.
The rule that keeps you safe: the free consumer tier of a tool may train on what you type; the paid business tier your company has approved generally does not; know which one you are in before you paste. If there is no policy, how to write an AI usage policy for your workplace has a one-page template. Your finance team should probably be the one that writes it.
What does this do to the finance career path?
It compresses the production work and raises the value of the judgement work — the same shift every analytical role is going through. The junior analyst whose job was building the pack is exposed; the analyst whose job is explaining the pack to someone who has to decide is not, and now has three more days a month to do it. The Finance Professional role page lists what the role now expects; AI and entry-level jobs is the honest treatment of the junior end.
For an analyst who wants to move toward the data side — forecasting models, anomaly detection, the FP&A platforms' modelling layers — SQL is the bridge and Python is the next step, and AI in finance teams: a practical guide is the longer walkthrough.
Where to start this month
Take last month's commentary. Write the brief. Get the draft. Compare it to what you sent. Do that for four weeks and you will have learned more than any course — and you will have a standard prompt your team can use. Then do the SQL. The free AI for Finance skill check takes three minutes with a free student account and tells you which of the four skills above you already have; the AI for Finance course is the graded version, built around your own month-end.