AI in finance: what it is, and what it is not

Most finance teams now have access to an AI assistant. Using it well starts with an accurate picture of what it is.

What these tools do

A large language model predicts likely text from patterns in its training data. That makes it strong at drafting, summarising, restructuring, explaining and translating. It is weak at exact arithmetic, current facts and anything it has not been given.

Where they help finance

• Summarising long contracts, board papers and regulatory texts.
• Drafting variance commentary from numbers you provide.
• Explaining accounting treatment options for review.
• Writing formulas, macros and queries, which you then test.

Where they fail

• They can state wrong facts with confidence, including invented citations and rates.
• They do not know your ledger unless you connect or paste it.
• Laws and rates change; training data ages.

Picking a first use case

Choose a task that is frequent, text-heavy, low in risk, and easy to check. Variance commentary and meeting summaries qualify. Tax filings and payments do not, at least not without strong review.

Exercise

List five recurring tasks in your team. Score each for frequency, risk and ease of checking from 1 to 5, and pick the one with the highest total of frequency and ease minus risk.