AI for Finance and FP&A: From Month-End Close to Forecasts Leaders Trust

How to use AI in FP&A in an auditable way: variance commentary, rolling forecasts, scenarios and anomaly checks at close, keeping calculation separate from writing.

It is the 5th of the month and the FP&A team at a building materials distributor is still waiting for numbers from three warehouses. Leadership wants to know why gross margin fell, but the team has not started analysing, because almost all their time goes to pulling data, reconciling and formatting slides. Finance is one of the functions where AI helps most clearly, but it is also where small mistakes are the most expensive.

What is AI in FP&A?

FP&A (financial planning and analysis) is the work of planning, budgeting, forecasting and explaining numbers so leaders can decide. AI helps on three layers: reading and organising data, detecting anomalies, and writing explanations in plain language. A Gartner survey published in September 2024, covering 121 finance leaders, found that 58% of finance functions already use AI, with the most common uses being intelligent process automation (44%) and anomaly and error detection (39%).

First principle: separate calculation from writing

Language models are good at reading, summarising and writing, but they should not be your calculator. Every number must come from formulas in Excel, SQL or Python code that can be traced, and only then should AI read the results and write the explanation. If you use an AI that can run code, have it show you the formula or code, not just the number.

Four FP&A tasks where AI really helps

1. Variance commentary: explaining why numbers differ from plan

Feed it an already-calculated actual versus budget table with context, such as rising steel prices or a major customer delaying delivery. Have AI draft explanations split into price, volume and mix, then the analyst reviews and adds what the data does not show.

2. Driver-based rolling forecasts

Use AI to help find drivers that explain sales, such as the number of construction projects in the area or seasonality, then build a spreadsheet model with adjustable assumptions. Each month, update the actuals and have AI summarise which driver moved the forecast.

3. Scenario planning

Ask questions like "if transport costs rise 10% and contractor customers pay 15 days later, what does next quarter's cash flow look like?" AI helps structure the scenarios and write the summary, while the numbers are still calculated in your model.

4. Close and anomaly checks

Before closing the books, have AI scan for unusual entries such as expenses posted to the wrong cost centre, duplicates, or amounts that jumped sharply from last month, and produce a list for accountants to review. AI does not say what is wrong; it says where to look first.

Example: a building materials distributor

A two-person FP&A team sets up a new workflow: sales and cost data from the ERP flows into a standard spreadsheet on the 2nd of each month, formulas calculate variances automatically, AI drafts explanations by warehouse and product group, and produces an anomaly list for accounting to check. By the 4th the team has a draft report, and uses the remaining time to talk with sales about late-paying customers, work that actually improves next month's numbers.

Let formulas calculate, let AI explain, and let you decide.

Cautions

• Numbers AI makes up: AI may produce figures or percentages that look reasonable but do not exist in the data. Every number in a report must trace back to its source.
• Confidential data: unpublished financials, salaries and customer data must only be used with company-approved tools that do not train on your data.
• Auditability: auditors and executives must see where numbers came from. Keep the formula files and the prompts you used.
• Tax and accounting standards: AI can give general information, but tax interpretation and accounting treatment should be discussed with your auditor or a tax advisor.

How to get started

1. Standardise one monthly report template first.
2. Start with variance commentary, because it is low risk and clearly saves time.
3. Write prompts that forbid creating new numbers and only cite figures in the table.
4. Add anomaly checks to the close, with accountants making the final call.
5. Expand to rolling forecasts once the data is clean and the team trusts the workflow.

To start systematically, try the course Copilot in Excel for summarising and analysing data and read Business data analysis with Excel and Power BI to build your data foundations first.