Your forecasting cycle still runs on spreadsheets, late nights and a scenario or two that you had time to build. The question every group CFO is now asked is not whether AI belongs in FP&A, but where it earns its place — and where the hype quietly falls over. The honest answer is narrower and more useful than the headlines suggest.
What are finance teams actually using AI for in FP&A today?
Three things, mostly, and all of them are analytical rather than autonomous. The first is probability-weighted scenario modelling: instead of building two or three scenarios by hand, the model runs many, weights them by likelihood, and shows you the distribution of outcomes rather than a single line. The second is driver-based forecasting — letting the model learn the relationships between your operational drivers and financial results, so a forecast updates when volume, price or headcount moves rather than when someone rebuilds the sheet. The third is drafting: giving the model prior-period figures, current actuals, the budget and a short context note, and getting a first draft of board-pack commentary back in minutes. None of these replace the FP&A professional. Each removes the mechanical work that sits between them and the analysis.
Does AI really speed up the planning cycle, or is that marketing?
It is real, but the gains come from process, not magic. Teams that embed AI into planning and forecasting commonly report their budget and planning cycle time falling by roughly a third — 30 to 40 per cent is the shape people describe once the model is genuinely inside the workflow rather than bolted on. The most visible change is in scenario work: runs that used to take days of manual rebuilding can drop to minutes, because the model does the recalculation and you spend your time interrogating the results. That matters most when the board asks the question you did not prepare for — a supplier shock, a currency move, a delayed acquisition — and you can answer in the meeting rather than the week after.
Can AI write our board commentary?
It can write the first draft, and that is the right ambition — not the last word. Feed the model prior-period numbers, current actuals, budget and a paragraph of context, and it will produce readable commentary that explains what moved and by how much. That saves the blank-page hour. What it cannot do is know why the numbers moved in the way that matters to your board: that the variance in one region reflects a go-to-market change you made in March, or that a strong month hides a key customer starting to wobble. The draft states the arithmetic; you supply the meaning. Every credible team treats the output as a starting point that a finance professional edits, corrects and signs off — never as copy that ships unread.
Where does AI in FP&A go wrong?
It goes wrong when the model is asked to replace judgement instead of informing it. A forecast is only as good as the process around it, and AI does not fix a broken one — it accelerates it. The commercial context that changes a number lives outside the data: a changed pricing strategy, a customer relationship cooling, a regulatory shift the historical data has never seen. The model cannot know these things, so a team that defers to the output loses exactly the value FP&A exists to add. This is also the most common reason pilots fail to scale, a pattern we cover in our piece on why finance-AI pilots stall: the technology works in the demo, but no one has decided how it fits the planning discipline, who owns the assumptions, or where the human check sits. The winners keep the professional firmly in the loop and use the model to do more analysis, not less thinking.
Where should a finance function start with AI in forecasting?
Start with one process, one clear pain, and a decision the output will actually inform. Scenario modelling is usually the best first move: the value is obvious, the data you need is data you already have, and the model works alongside your existing plan rather than replacing it. Give it clean drivers, a defined question, and a person who owns judging the answer. Prove it there, build the habit of editing and challenging what the model produces, and only then widen to driver-based forecasting and commentary drafting. The direction of travel is towards what we call AI-embedded finance — level five, the top of the maturity ladder, where intelligent automation runs quietly across planning, close and reporting and the team spends its time on the commercial questions. You do not arrive there in one jump. You get there one well-run process at a time.
How do we know if this is worth the effort for our group?
Size the gap before you buy the tool. The value of modernising FP&A is not the licence cost avoided; it is faster cycles, better decisions and finance time redirected from assembly to analysis — and that value is measurable for your specific group. That is the principle behind everything we publish: the score is the headline; the pounds are the point. A structured read of where your finance function sits today, across close, consolidation, reporting, budgeting, forecasting, statutory work and the AI and ESG frontier, tells you where AI pays off first and what the gap to level five is worth. Once you can see the pounds, the case for starting stops being a matter of faith.
Common questions
What is AI used for in FP&A right now?
The three uses working reliably today are probability-weighted scenario modelling, driver-based forecasting, and drafting first-cut board-pack commentary from the numbers. All three are analytical aids that speed up mechanical work. None of them remove the finance professional's judgement — they free up time for it.
Does AI make financial forecasting faster?
Yes, chiefly by collapsing scenario work that once took days into minutes and by cutting overall cycle time. Teams that embed AI in the planning process commonly report budget and planning cycles falling by around a third. The gain comes from redesigning the process around the model, not from the tool alone.
Can AI write board-pack commentary?
It can produce a solid first draft if you give it prior-period figures, current actuals, the budget and a short context note. What it cannot do is know why the numbers moved in ways that matter commercially — a changed strategy or a key customer at risk. A finance professional must edit, correct and sign off the draft before it goes near the board.
Will AI replace FP&A analysts?
No. AI handles recalculation, pattern-finding and drafting; the analyst supplies the commercial context the model cannot see and the judgement the board relies on. Teams that defer to the output instead of using it lose the value FP&A exists to add. The professionals who thrive use AI to do more analysis, not less thinking.
Where should we start with AI in FP&A?
Begin with one process where the pain is clear and the data already exists — scenario modelling is usually the best first move. Prove the value there, build the habit of challenging the model's output, then widen to driver-based forecasting and commentary. This is the first step towards AI-embedded finance, the top of the maturity ladder.