FP&A's Real Bottleneck Is the Data Gathering. That's Exactly What AI Should Run.

Most of the time you spend on a forecast goes to assembling and cleaning the data before you can even think, and that's the exact slice of FP&A that AI runs best - pulling actuals out of the ERP, stitching together the export from three systems that don't talk to each other, flagging the line that moved 40% for a reason nobody documented. So the honest read on AI in FP&A is that it comes for the data-janitor hours first and the judgment last, which, if you're an analyst drowning in workbook prep, is roughly the best possible order for it to arrive in.

And it is arriving, just slower than the headlines suggest. Gartner's 2025 survey found 59% of finance leaders reported using AI in the finance function, barely up from 58% the year before, even as the same researchers project nine in ten finance teams will run at least one AI-enabled tool within two years. So the gap between who's experimenting and who's actually wired it into the close-and-forecast cycle is wide open right now, and that gap is the opportunity. Let me walk through where AI genuinely earns its keep in FP&A and where you still own the call.

What does AI actually do in FP&A today?

It takes the assembly line, not the strategy. In practice that means gathering and reconciling the actuals, drafting a first-pass variance explanation you then sanity-check, generating a few scenario versions off your assumptions so you're editing instead of building from scratch, and turning a messy variance table into a readable first draft of the board narrative. None of that replaces the analyst - it gets you to the part of the job that needed you in the first place faster, with more hours left to actually think about what the numbers mean.

This is the same pattern we've been writing about across the function. We covered the reconciliation version of it in our piece on rolling reconciliations, and the variance-analysis version in the controller's playbook on automating flux and variance - and FP&A forecasting is the next obvious place the same logic lands, because the forecast is only as fast as the data feeding it.

Why the data gathering is the real bottleneck

Ask any analyst where the forecast actually stalls and it's rarely the modeling - it's getting the inputs clean and trustworthy enough to model on. The benchmarking data agrees: finance teams consistently name data reliability and accessibility as their top technology obstacles, well ahead of any shortage of fancy tools. You can have the slickest planning platform on the market and still lose two days a month to chasing a number that three systems each report slightly differently.

That's why AI lands so well here, because the bottleneck was always the plumbing and AI is good at plumbing. When a tool can pull the actuals, normalize the formats, and reconcile the obvious differences before you sit down, you walk into the forecast with the boring 70% already staged. Get the data layer right and everything downstream - the model, the scenarios, the story - gets faster and less error-prone almost for free.

Where AI earns its keep

Start with the unglamorous, high-frequency stuff. Data consolidation across systems is the obvious first win, because it's repetitive, rules-based, and soul-crushing by hand. First-pass variance commentary is the second - let the tool draft "revenue came in 6% under plan, driven mostly by the enterprise segment" and spend your energy validating and explaining the why. Scenario generation is the third, because spinning up best, base, and downside cases off a shared assumption set is exactly the kind of structured, repeatable work a machine should handle while you argue about whether the assumptions are right.

If you're choosing tools for any of this, the thing that actually matters is whether the software can be driven by an agent end to end, not just whether it has an AI logo on the marketing page. We made that whole argument in our software directory framing - evaluate the workflow, not the feature list - because a planning tool that an agent can operate is worth far more to a lean FP&A team than one with a chatbot bolted onto the corner.

Where you still own the call

The assumptions are yours, full stop. AI can generate a downside scenario, but deciding that the downside is the one leadership should plan around is judgment, and judgment is the job. Same with the narrative that actually moves a decision - a model can tell you revenue missed, but a CFO doesn't act on "revenue missed," she acts on "revenue missed because the deal we were counting on slipped to Q3, and here's whether that's a timing problem or a demand problem." That sentence is the whole value of FP&A, and damn few tools get anywhere near it.

So the realistic division of labor is AI on the inputs and the first drafts, you on the assumptions, the judgment, and the story. The analysts who get this end up doing more of the interesting work, not less work - which, not coincidentally, is exactly what the best-paid FP&A seats have always been about, as we laid out in our breakdown of FP&A analyst roles.

What I'd do to start

Pick one recurring forecast input and hand just that to AI. The monthly actuals pull, the headcount reconciliation, the one consolidation step that always eats your Friday - wire up a single piece of the data layer and watch how far it gets before it needs your brain. You'll learn more from that one build than from a quarter of reading vendor decks, and you'll come away a little pissed at how many hours you've handed to copy-paste over the years. Then do the next input. That's the whole play - steadily move the data-gathering grind to the machine so your time goes to the assumptions and the story, which is the part that was always actually yours.

Frequently asked questions

Can AI do financial forecasting on its own?

Not the part that matters. AI is strong at gathering and cleaning the data, drafting variance commentary, and generating scenarios off your assumptions, but setting those assumptions and judging which scenario to plan around is human work. Treat it as the analyst's prep engine, not the decision-maker.

Where should an FP&A team start with AI?

With the data layer. Data consolidation and reconciliation across systems is the highest-frequency, lowest-judgment work in the forecast cycle, which makes it the safest and most valuable first place to automate. Get the inputs staged automatically and everything downstream speeds up.

How many finance teams actually use AI today?

Gartner's 2025 survey put it at 59% of finance leaders reporting AI use in the function, roughly flat from 58% the year before, with most expecting to run at least one AI-enabled tool within two years. Adoption is real but uneven, so wiring it in now is still a genuine edge.

Will AI replace FP&A analysts?

It replaces the data-gathering hours, not the analyst. The judgment, the assumptions, and the narrative that drives decisions are exactly what AI is worst at and what FP&A is most valued for, so the work shifts toward those rather than disappearing.

What should I look for in an AI-ready planning tool?

Whether an agent can actually operate the workflow end to end, not whether there's an AI badge on the homepage. A tool an agent can drive is worth far more to a small FP&A team than one with a chatbot bolted on - evaluate the workflow, not the feature list.

The forecast isn't getting any less data-hungry, so the analysts pulling ahead are the ones handing the plumbing to AI this quarter and spending the reclaimed hours on the calls only they can make. Wire up one input this week, even a rough one, and you'll already be ahead of most of the function.