AI in FP&A: How Finance Teams Are Using It Today

A forecast can be out of date before the month is over. Sales move, hiring plans change, costs come in differently than expected, and FP&A has to work out what changed before deciding what happens to the plan.

‍

AI can take on more of that work, from investigating variances to refreshing forecasts and preparing reporting. The bigger question is how much of it finance can trust AI to handle.

‍

This article looks at where AI fits across the FP&A cycle, where it falls short, and where people still need to step in.

‍

[tldr]

What does AI in financial planning and analysis mean?

AI in financial planning and analysis (FP&A) means using AI to help collect, analyze, forecast, explain, and act on the financial and operational information behind planning and decision-making.

‍

That can range from relatively simple work, such as helping with an Excel formula or drafting variance commentary, to more connected workflows where AI can update forecasts, investigate changes in actuals, prepare reporting, or carry work across several steps.

‍

The difference matters because FP&A is not just a reporting function. The team has to understand what happened, decide whether the assumptions about what happens next still make sense, and help the business act on that information.

‍

AI can take on more of the work around those decisions. It does not automatically own the decisions themselves.

How are FP&A teams actually using AI today?

AI adoption in FP&A looks very different depending on what counts as "using AI."

‍

A 2025 survey by FP&A software vendor Drivetrain of 258 FP&A professionals, run in July 2025, found that 81% were using AI in their FP&A work, but only 26% were using it regularly; the other 55% used it occasionally. Most of that use was in tasks that are easy to implement: 57% used AI for report writing and commentary and 48% for Excel automation, while only 28% applied it to forecasting and planning. [1]

‍

Research from FP&A Trends Group, sponsored by EY, paints a similar picture from another angle. In a poll of 322 senior finance leaders run between January and March 2025, 53% said they were not using AI in FP&A at all, while 27% were using it for reporting or workflow automation, 10% for pattern and trend analysis, and 10% for forecasting.  [2]

‍

The two findings are less contradictory than they look. Drivetrain counted individual FP&A professionals who use AI at all, including occasionally, while the FP&A Trends poll asked senior finance leaders how AI is used in their FP&A function.

‍

Taken together, they suggest AI is already common for individual tasks, but much less embedded in core FP&A workflows such as forecasting and planning. Gartner's 2026 AI report card for finance, based on a March 2026 survey of 204 finance leaders, points the same way: financial forecasting and insight generation were among the lowest-rated AI use cases [3].

Where AI fits across the FP&A cycle

AI is most useful where the task is repeatable, the underlying data is available, and finance can check the result. That creates very different levels of usefulness across the FP&A cycle.

‍

Pulling and validating actuals

AI can help bring actuals together, map them into the right structure, and flag missing or unusual values before they reach a forecast or report. That can reduce some of the manual work between the month-end close and FP&A analysis.

‍

But if the books are not reconciled or different systems use inconsistent dimensions, finance still has to fix the underlying data.

Variance analysis and commentary

AI can compare actuals with budget, forecast, or prior periods, identify the accounts driving the movement, and draft a first explanation.

‍

Causation still needs more context. A system may see that cloud spend increased, but not know whether that came from customer growth, a migration, or a deliberate change in infrastructure. FP&A still needs to confirm why the number moved.

Forecast refresh and driver-based planning

AI can pull in the latest actuals, recalculate the model, and show where performance has moved away from plan. It can also update driver-based forecasts as inputs such as headcount, volume, pricing, or unit costs change.

‍

Finance still has to decide what those changes mean for the assumptions ahead. Updating the math is easier than deciding whether the business will recover, slow down, or follow a different path.

Scenario planning

Once the main drivers are defined, AI can run alternative scenarios quickly and show how changes affect revenue, margin, cash, or runway.

‍

The value comes from choosing the right scenarios to test. Finance still has to decide which possibilities are realistic enough to influence a decision.

Headcount planning

AI can consolidate employee data, open roles, compensation, start dates, and hiring plans, then show the financial effect of changing those assumptions.

‍

The decisions themselves sit outside the model. Whether a team needs another five hires, when those roles will start, or whether priorities have changed still depends on FP&A and the business.

Cash-flow forecasting

AI can combine current balances with expected collections, payments, payroll, and other cash drivers to keep the forecast updated as new information arrives.

‍

Cash timing is still uncertain. A large customer may pay late or a planned payment may move, so finance still needs to judge where the assumptions are most likely to break.

Board and management reporting

AI can turn existing analysis into a first draft of management commentary, summarize major movements, and help prepare recurring reporting.

‍

It cannot decide which numbers deserve the most attention or what leadership needs to hear. FP&A still owns the message, context, and final narrative.

Ad hoc analysis and business partnering

AI can help answer one-off questions by gathering information, building an initial analysis, testing alternatives, and preparing questions for the business.

‍

That can make FP&A better prepared, but it cannot replace the conversation itself. Business partnering still depends on challenging assumptions, understanding stakeholder context, and helping people decide what to do next.

What determines how far AI can go in FP&A

AI is only as useful as the data it has to work with. If actuals are late, dimensions are inconsistent, or teams are working from different versions of the forecast, it starts from a weak foundation. It can make analysis faster, but it cannot make unreliable inputs accurate. In a July 2024 Gartner survey of 1,203 data management leaders, 63% of organizations either did not have, or were unsure whether they had, the right data management practices for AI [4].

‍

How much work AI can take on also depends on the level of autonomy it is given. An assistant might explain a formula or summarize results, while a copilot helps an analyst prepare the work. An AI agent can go further by gathering information, using tools, and carrying a workflow through permitted steps.

‍

As AI moves from analyzing information to updating forecasts, changing records, or triggering other workflows, the need for permissions, controls, and human oversight increases. Gartner advises CFOs to start AI agents on a low-risk, contained workflow where errors are visible and reversible [5].

Where different AI tools fit in FP&A

There is no single type of AI tool that makes sense for every FP&A job. The main difference is how close the AI sits to the financial data and the workflow it is helping with.

‍

‍

General-purpose AI is often the easiest place to start because it can remove friction without changing the finance stack. For repeatable work, instructions can also be packaged into reusable workflows rather than rebuilding the prompt every month. 

‍

DualEntry's free Claude finance skills, for example, include workflows for flux analysis, a 13-week cash-flow forecast built from AR and AP aging, and a check of the SaaS metrics in a board deck, all of which run on exports from any ERP or GL, or live through the DualEntry MCP server.

‍

Dedicated FP&A platforms become more useful when teams need structured modeling, scenarios, and collaborative planning, while ERP-native AI can work closer to the accounting data those plans depend on.

What still needs a finance person?

The value of AI in FP&A drops quickly when the task depends on information the system cannot see or a judgment it cannot make from the data alone.

‍

That tends to happen in a few places.

‍

Checking the numbers. AI can produce an answer that looks convincing even when an input, formula, mapping, or assumption is wrong. Material analysis still needs a way to trace the output back to its source.

‍

Choosing assumptions. A forecast is not simply an extrapolation of historical data. FP&A has to decide what has changed and which assumptions still describe the business.

‍

Explaining causation. Finding a variance is easier than proving what caused it. The reason may live with sales, operations, product, or another part of the business rather than in the ledger.

‍

Handling confidential data. Financial plans can contain payroll, customer, pricing, M&A, and other sensitive information. Teams need clear rules around which tools can access it and how that information is handled.

‍

Owning the narrative. A technically correct analysis can still emphasize the wrong thing. FP&A decides what leadership needs to understand and what decision the analysis is supposed to support.

‍

This is why overall AI accuracy scores are not enough. DualEntry's Accounting AI Benchmark shows that performance varies significantly by accounting task. In Accounts Receivable, for example, scores across the models tested as of October 2026 range from 41.7% to 91.7%.

‍

FP&A teams should evaluate AI on the specific work they plan to give it rather than assuming one overall capability applies everywhere.

Put AI to work in FP&A with DualEntry

AI becomes more useful in FP&A when it does not have to wait for finance to export, clean, and reconnect the data every time the team wants to analyze something.

‍

That is where DualEntry's agentic ERP changes the workflow. Planning, actuals, reporting, and the underlying accounting data can live in the same system, giving AI access to the financial context behind the numbers rather than another static copy of them.

‍

DualEntry's budgeting and planning software keeps budgets in the same system as the ledger. Teams can budget by account and by department, location, or project, keep a board-approved plan and a reforecast side by side as separate budgets, and compare either one with actuals at the same level of detail, without rebuilding the connection between planning and accounting each month.

‍

Its financial analytics software carries that into analysis and reporting. Flux reports flag the variances that exceed your thresholds and let you drill into the transactions behind them. For each flagged line, AI can draft an explanation based on the account's historical data and the variance, and Accounting Intelligence can explain in plain language what drove an account's movement, using the underlying GL transactions. Each AI explanation stays a draft until someone on the team accepts or discards it.

‍

That gives FP&A a shorter path from what happened to what changed and what should we do about it, while keeping the source numbers and supporting detail close to the analysis.

‍

The goal is not to automate the judgment out of FP&A. It is to spend less of the team's time getting the analysis ready and more of it deciding what the business should do next.

AI in FP&A FAQs