AI FP&A use cases: what AI does well in financial planning and analysis, and what it needs underneath
Manoj Tutika · Published 16 November 2026 · Last checked 16 November 2026
Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.
Which AI FP&A use cases actually work today?
AI in FP&A is useful where the work is repetitive, numerical and checkable: explaining variances, rolling a forecast forward, testing a driver model, running a scenario and drafting the board pack. It is less useful where the work is judgement: choosing assumptions, deciding what to cut, telling the board what it does not want to hear.
The hard part is rarely the model. The 2026 FP&A Trends survey of 475 respondents found that 47% of FP&A effort goes on collecting and validating data, and only 32% on generating insight and acting on it. Only 19% rated their data quality as advanced, and 44% said they can see less than three months ahead with confidence (published 9 July 2026, checked 16 November 2026). AI helps with the second half of that split only when the first half is solved.
What does each use case look like in practice?
| Use case | What AI does | What it needs underneath | What stays with the team |
|---|---|---|---|
| Variance commentary | Compares actuals to budget, ranks material movements, links each to its transactions, drafts the explanation | Closed actuals coded to the right account and cost centre | Checking the reason is real, deciding what to escalate |
| Rolling forecast | Rolls the horizon forward each month, replaces forecast with actuals, re-projects run-rate lines | Current actuals and open receivables and payables | Overriding where the business has changed |
| Driver-based model | Builds the link from drivers such as headcount, orders or price to revenue and cost, tests which drivers explain history | Clean history by month and the operational drivers | Agreeing which drivers the business manages by |
| Scenario analysis | Runs best, base and worst cases, shows the cash and covenant effect | A model with explicit assumptions and a cash view | Choosing which scenarios matter and acting on them |
| Board pack | Drafts the narrative, tables and charts from the closed month | Final numbers, last month's pack, the board's standing questions | The message and the recommendation |
| Forecast accuracy review | Compares each forecast to actuals by line and horizon, flags persistent bias | Saved forecast versions and actuals | Changing the method where bias persists |
Variance commentary
This is the most mature use case. A good AI draft looks like this: "Software costs are £18,400 against a budget of £12,000. £4,900 is an annual licence booked in full in October that should be spread over twelve months; £1,500 is two new seats." The first sentence is arithmetic. The second is the value, and it only exists if the AI can see the invoice behind the line. Microsoft's variance analysis in Excel, in preview, and Claude for Excel both work on the workbook in front of them; the quality depends on what is in it.
Rolling forecasts and driver models
AI is good at the mechanics: dropping the past month, adding a new one, replacing forecast with actuals and re-projecting lines that follow a run rate. It is also good at testing a driver model against history, for example whether revenue tracks active customers or orders better. What it cannot know is that the sales director expects a large contract to slip a quarter. That input still comes from people.
Scenarios
ICAEW's guidance on cash and forecasting asks for worst, most likely and best case scenarios, warns about confirmation bias, and asks finance teams to compare forecasts with actual results to improve them (4 November 2022). AI makes running three or ten scenarios cheap. The discipline of choosing assumptions that are honest, and of checking last quarter's forecast against what happened, is still yours.
Board packs
AI drafts the pack well when it has last month's pack, the closed numbers and the board's standing questions. Give it those, and the first draft takes minutes. Keep the recommendation and the message human.
What needs live books underneath?
Every use case in the table depends on actuals that are complete, current and correctly coded. In most UK SMEs and mid-market companies that is not the case until well into the following month. Purchase invoices arrive late, bank lines sit unreconciled, accruals are booked on day eight of the close, and a cost lands on the wrong cost centre.
AI on top of those books produces a confident explanation of numbers that are not final. Ask an assistant on the 3rd why margin fell, and it will tell you, using a month in which half the supplier invoices are missing.
The fix is not a better prompt. It is a ledger that is right during the month, not after it: bank lines matched daily, supplier invoices booked on arrival, accruals and depreciation posted on schedule. Then FP&A can start on the 1st with numbers it trusts. See management accounts with AI and budget vs actuals for small companies.
What are the limits of AI in FP&A?
- It explains, it does not decide. AI can show that a scenario breaks a covenant in month seven. Deciding what to do about it is the CFO's job.
- It inherits bias. If last year's forecasts were always optimistic, an AI that copies your process will be too unless you measure bias and correct it.
- It needs review. Anthropic's own Excel page says its add-in is not recommended for audit-critical calculations without verification. The same applies to any model output that reaches the board.
- It is only as current as the data. A forecast built on a two-week-old export is a two-week-old forecast.
What do Thred's agents do for FP&A?
Thred is a team of AI agents that handle daily work for finance teams and accounting firms, on your existing accounting software. For FP&A, the point is simple: the agents keep the actuals right, so the analysis starts from numbers that are final.
The Bookkeeper keeps the record up to date, booking supplier invoices and matching bank lines as they arrive. The Analyst monitors performance and variance, ranks the movements that matter and links each one to the transactions behind it. The Treasurer manages your cash and keeps the short-term cash view current. The CFO is your finance lead when you have none, turning the month into a board-ready story. The Collector and The Payer manage receivables and payables, so the open items in your forecast are real. The Connector finds context in your connected tools, such as the email that explains a one-off cost, and The Accountant handles filings and optimises for taxes.
On Exact Online the agents read and write directly, with every write shown as an approval card until you set a rule, and The Accountant runs the period close per administration. Agents for Xero and Moneybird work alongside. Read AI cash flow forecasting for the cash side, or see Thred Layer.
Frequently asked questions
What is AI FP&A?
AI FP&A is the use of AI models in financial planning and analysis: drafting variance commentary, updating forecasts, building and testing driver models, running scenarios and preparing board packs. The analyst still owns the judgement; the AI does the gathering, calculating and first draft.
What are the best AI FP&A tools?
There are three groups. Assistants inside spreadsheets, such as Copilot in Excel and Claude for Excel. Planning platforms that add AI to budgeting and forecasting. And agents that work on the ledger itself so the actuals are current. Most teams need a mix, and the ledger layer decides how good the rest can be.
How is AI used in FP&A today?
Mostly for variance commentary, forecast updates and reporting. The 2026 FP&A Trends survey of 475 respondents found that 47% of FP&A effort still goes on collecting and validating data, against 32% on insight and action, and 15% expect to use AI agents within six months (checked 16 November 2026).
Can AI do variance analysis?
Yes, and it is the most mature use case. AI compares actuals to budget or forecast, ranks the movements that matter, links each one to the transactions behind it and drafts the explanation. A person checks that the explanation is the real reason, not just the arithmetic.
Can AI do financial planning and analysis on its own?
No. AI can prepare forecasts and scenarios, but choosing assumptions, judging risk and deciding what the business should do remain human work. The useful split is that AI does the preparation and the first draft, and the FP&A team spends its time on the questions behind the numbers.
What data does AI need for FP&A?
Closed, correctly coded actuals by month, cost centre and entity; a budget or forecast in the same structure; operational drivers such as headcount, orders or customers; and open items such as receivables and payables. Without current actuals, every AI output inherits the error.
Sources
- FP&A Trends, Closing the AI readiness gap in 2026 (published 9 July 2026, checked 16 November 2026)
- ICAEW, Nine principles for finance professionals (4 November 2022, checked 16 November 2026)
- Microsoft Learn, Variance analysis in Excel overview (preview) (checked 16 November 2026)
- Anthropic, Use Claude for Excel (checked 16 November 2026)
Read next
- AI cash flow forecasting: what AI adds to a 13-week forecast, with a worked £ example
- Budget vs actuals for small companies: a weekly view instead of a quarterly surprise
- Management accounts that start with the decision, not the P&L
- Controller vs FP&A vs CFO: who does what in a finance team, from bookkeeper to CFO
- AI management accounts: what goes in the pack, AI commentary on real figures, and what the reviewer checks
Where Thred comes in
Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.
