Finance and accounting teams spend their days among spreadsheets, formulas, reports and correspondence: month-end close, explaining budget variances, management report commentary, auditor questions. Generative AI can save real time on much of this work.
But finance is where AI's most dangerous mistake costs the most: a wrong number that looks convincing. This article covers the most productive use cases for finance and accounting teams, the key risks, and practical controls to manage them.
7 most productive use cases
1. Excel formulas and macros. Turning a request like "sum each customer's last 90 days based on the dates in column B" into a formula, simplifying nested IFs or finding the cause of an error. Excel does the calculation; AI only writes the formula.
2. First drafts of report commentary. Building the skeleton of variance explanations. The team supplies the numbers and reasons; AI tidies the language.
3. Understanding regulations and standards. Summarising a long circular or standard, or getting a first view on "what does this clause mean for us?". The final check must always be made against the official source.
4. Procedures and policies. First drafts of close checklists, approval workflows and expense policies.
5. Correspondence. Payment reminders, supplier emails, draft replies to auditor questions.
6. Data cleaning ideas. Suggesting step-by-step methods or Power Query steps to tidy a messy export.
7. Learning and training. Explaining an accounting concept to a new team member with different examples, or preparing short quizzes.
4 finance-specific risks
The Institute of Chartered Accountants in England and Wales (ICAEW) highlights similar risk areas in its generative AI guide: bias, inconsistency, hallucination, and data privacy.
1. Wrong numbers and calculations
Language models are not calculators. They generate text based on probabilities, so they can give a total, a percentage or a currency conversion fluently and incorrectly. Asking the same question twice can even return different results.
2. Confidential data and trade secrets
Unpublished financial results, budgets, prices, salaries, customer and supplier details are among an organisation's most sensitive data. ICAEW also stresses that confidential client or organisational information should not be entered into public tools. Where this data includes personal data, data protection obligations apply as well.
3. Out-of-date regulatory knowledge
Models have a knowledge cut-off, while tax rates, exemptions and reporting rules change often. AI can present an old rule as current, in a confident tone.
4. AI-assisted fraud
Generative AI makes it easier to produce fake invoices, fake emails and even messages imitating an executive's voice. Finance teams are the first target of such attempts.
7 controls that manage the risk
- Check every number against its source. No figure from AI should go into a report without being compared to the accounting system or source document.
- Let tools, not AI, do the maths. AI writes the formula; Excel or the ERP does the calculation.
- Use approved enterprise tools only. Financial data should never go into tools used through personal accounts.
- Share only what's needed. A formula or method doesn't require real data; a few rows of sample data are enough.
- Keep the four-eyes principle. AI-assisted reports go through the usual approval workflow.
- Confirm payment instructions through a second channel. Verify urgent payment or bank detail change requests received by email or message using a known phone number.
- Record usage. Knowing which tool was used in which report makes audit questions easier to answer.
An example workflow: budget variance commentary
- The team extracts variances from the accounting system and notes the main reasons.
- AI receives only line names, the direction of each variance and the team's notes; amounts are not shared unless necessary.
- AI writes a draft in management report style.
- The analyst adds figures from the system and checks every sentence.
- The report goes through the normal approval process.
In this workflow AI saves time, but it is never the source of any number.
Conclusion
The value of generative AI for finance teams is real, but it is only safe with the right habits. The best results come from teams that use AI for language and structure, and their systems for numbers.
To see your team's current usage risk, try the free GenAI Usage Risk Assessment. We tailor our Generative AI at Work course with finance and accounting examples, and Using Generative AI Safely quickly builds data privacy and verification habits across the whole team.
Sources