AI forecasting in FP&A software: what planning tools actually do in September 2026
AI forecasting in FP&A software comes in three forms: machine learning that produces a baseline forecast from actuals, analyst agents that explain variances and drivers in plain language, and assistants that build or change planning models. In the AFP 2025 FP&A benchmarking survey only 23 percent of practitioners used AI, while 92 percent worked in spreadsheets every day. Most planning vendors quote prices on request, so trials, hosting and forecast testing matter more than list prices.
- AI in planning tools means three things: machine learning baselines, analyst agents that explain numbers and assistants that build models.
- Only 23 percent of FP&A practitioners used AI in the AFP 2025 survey; 40 percent were testing it.
- 92 percent of practitioners used spreadsheets daily, and several leading tools keep Excel as their front end.
- Published prices are rare; LivePlan is an exception at $15 or $30 a month billed annually.
- Test an AI forecast against past actuals before it replaces a manual one.
AI forecasting in FP&A software is any feature that uses machine learning or a language model to produce, explain or build a financial plan. It ranges from a statistical baseline for next quarter's revenue to an agent that answers why operating costs rose last month. Planning vendors added these features quickly in 2025 and 2026, and the labels now sound alike even where the features differ.
This guide sorts the features into clear groups, shows how far FP&A teams have adopted them, compares how the planning tools in our directory price, host and sell their AI, and describes a simple way to test a forecast before you trust it. It draws on vendor pages, a benchmarking survey by the Association for Financial Professionals and our dataset as of September 26, 2026. ai-toolfinder is independent and paid by no vendor.
Three kinds of AI in planning tools: baseline, analyst, modeler
The oldest kind is the machine learning baseline. Planful calls its version Projections, IBM Planning Analytics includes AI based forecasting in all its plans, Jedox sells Planning Wizards for revenue forecasting, demand planning and churn prediction, and Corporate Planner produces daily trend forecasts that pick up seasonality without a manual forecast cycle. These models extend history. They do not know about a price increase you decided last week unless someone enters it as a driver.
The second kind is the analyst agent. Anaplan's Finance Analyst monitors performance and spots deviations, Planful's Analyst writes variance explanations at account level, and OneStream's Forecast Agent, generally available since May 2026, explains forecast accuracy and drivers in conversation. These agents save the time spent writing commentary, and their quality depends on how cleanly the model maps accounts and drivers.
The third kind builds the model itself. Anaplan's CoModeler creates model logic from natural language, and Valsight's Simulation Agent turns a management question into scenarios that run on a driver tree. This is where the biggest time savings are claimed and where review matters most, because a wrong formula keeps producing wrong numbers every month.
Quick check
How well do you know AI in FP&A?
Question 1 of 3
What share of FP&A practitioners used AI in the AFP 2025 benchmarking survey?
Source: AFP 2025 FP&A Benchmarking Survey via CFO.com, January 2025
How many FP&A teams use AI forecasting: a minority, most in pilots
Adoption is lower than vendor marketing suggests. The Association for Financial Professionals surveyed 362 finance and FP&A professionals for its 2025 FP&A benchmarking survey. 23 percent used AI, 40 percent were testing it with plans to implement within 12 months, and 36 percent were neither experimenting nor planning to.
The daily tool is still the spreadsheet. 92 percent of respondents used spreadsheets every day and 58 percent used spreadsheet add ins daily, against 12 percent for commercial AI tools and 8 percent for generative AI. Where teams did automate, the leading areas were creating a single version of the truth (25 percent), standard planning calculations (21 percent) and baseline budgets and baseline forecasts (18 percent each).
The reasons for waiting are practical. On a scale from 1 to 5, respondents rated lack of expertise highest at 3.8, followed by unclear return on investment at 3.7, and data safety, fear of incorrect answers and a missing corporate AI strategy at 3.6 each. Company policies banning AI scored lowest at 2.7, so outright bans are rarely the obstacle.
AI use among 362 finance and FP&A professionals, AFP 2025 FP&A Benchmarking Survey via CFO.com.
Most automated FP&A areas, AFP 2025 FP&A Benchmarking Survey via CFO.com.
Sources: [8]
Planning tools in our directory: more European than the close market
Our directory lists 57 tools for budgeting and forecasting, including products whose main focus is reporting, treasury or the close. 28 of them are planning tools first.
The planning market is less dominated by the United States than the close software market. 23 tools have their headquarters in the United States and 9 in Germany, and 14 can keep data in the EU according to our data. For comparison, only 7 close tools offer EU hosting, which our guide to AI in the financial close covers in detail.
Trials are more common here too: 16 planning tools offer a free trial. Most still quote prices on request. You can browse all of them on our budgeting and forecasting page or filter the list below.
Our data · interactive
Filter the planning tools in our directory
Pick a headquarters region or focus and see which share offers EU hosting, certificates, public prices or a trial.
33 in this selection
Line: value across all (33)
Source: ai-toolfinder.com dataset, as of 26 September 2026
Sources: [9]
Excel stays: spreadsheet-first tools versus planning platforms
Given that 92 percent of FP&A practitioners work in spreadsheets daily, many vendors put their AI into Excel instead of replacing it. Vena keeps Excel as its main interface and includes Vena Copilot in both its Professional and Complete plans. Cube works inside Excel and Google Sheets and includes its AI agents for data management, analysis, planning and board material in every plan.
Jedox, headquartered in Freiburg, Germany, offers an Excel experience next to its browser interface, and its Jedox MCP gives outside AI agents governed access to planning data and business logic. Finance teams keep their formulas and layouts this way, while the data comes from a central model with version control.
Platform-first tools take the opposite route. Anaplan runs finance, sales, supply chain and workforce planning on connected models, with separate analyst agents for each area. This suits companies that plan headcount, sales and costs together, but it asks more of the team, because models are built in the platform's own logic rather than in cells.
AI inside Excel or Google Sheets; formulas and layouts stay; a central model adds version control. Examples: Vena, Cube, Jedox.
Models built in the platform; finance, sales, workforce and supply chain can share one model; agents per area. Example: Anaplan.
Two ways planning vendors add AI, based on vendor pages as of September 2026.
Pricing and trials: what planning vendors publish
In our dataset 75 percent of the planning tools quote only on request. What vendors do publish is the structure. Jedox charges a monthly fee per named user, billed annually, in Business, Professional and Professional Plus packages, and its AIssisted planning and forecasting model sits in Professional. IBM Planning Analytics sells Essentials with 16 GB and 5 users, Standard with 32 GB and 10 users and Premium with 64 GB and 20 users, all with AI forecasting.
Whether AI costs extra differs by vendor. Cube and Vena include their AI in every plan. Board lists BEAM, its machine learning component, next to the Office add in and the SAP connector as additional components on its pricing page, so ask whether it is in your quote.
Two vendors make testing easy. Pigment offers a 14 day trial from the first login without a credit card, with limits on storage, calculation capacity and AI credits, after which the workspace turns read only. LivePlan, built for small business owners rather than FP&A departments, publishes its prices: $15 a month for Standard and $30 for Premium billed annually, or $20 and $40 monthly, with a 35 day money back guarantee. Its AI Forecast Builder is part of Premium only.
Self-check
Which kind of planning tool fits your team?
A rough orientation based on how you plan today. It does not replace a demo with your own data.
Is the plan mainly built by a small business owner or an outside adviser, without an FP&A team?
Bringing ChatGPT, Claude or Copilot to planning data through MCP
A newer pattern lets outside AI assistants work on planning data instead of building every agent into the planning tool. The Model Context Protocol (MCP) is the common route. Jedox MCP gives AI agents and enterprise AI tools governed access to data, models, planning structures and business logic inside Jedox. Cube lists an MCP server as part of its agentic finance layer and includes MCP in its Gold plan.
Planful goes the same way: it names ChatGPT, Copilot and Gemini as coming integrations through its MCP server, while its built-in features run on OpenAI. Drivetrain uses frontier models from Anthropic and OpenAI under zero data retention terms. For a finance team this means the assistant employees already use could answer questions on the approved plan, with the permissions of the planning tool, instead of on a spreadsheet someone exported.
The risk sits in the permissions. An MCP connection is only as safe as the roles behind it, so salary lines, bonus pools and board scenarios need the same access rules in the assistant as in the planning tool. Ask the vendor whether the connection respects row and cell level permissions and whether every query is logged, and involve IT security before the first connection goes live.
How to test an AI forecast before you rely on it
A forecast feature should earn trust on your own history. The simplest test is a backtest: hide the last 12 months of actuals, let the tool forecast them from the data before, and compare the result with what really happened and with the forecast your team made at the time. If the AI does not beat your own forecast on the lines that matter, it is a reporting aid, not a forecasting engine.
Check where the data goes while you test. Drivetrain lets customers choose data storage in Frankfurt or in Virginia, states that customer data does not train or fine-tune any model, and uses frontier models from Anthropic and OpenAI under zero data retention terms. Planful names OpenAI as its AI partner and states that customer data never trains external models. Ask every shortlisted vendor for the same three answers: region, model provider and training use.
Finally, keep the forecast explainable. Tools that show drivers, such as Jedox with its explainable forecasts or Valsight with driver trees, make it easier to defend a number in front of management. A black box forecast that beats your team slightly but cannot say why will not survive the first budget meeting. Our FP&A category lists reporting and planning tools side by side if you want to compare wider.
- Pick the linesChoose the 5 to 10 plan lines that drive decisions, such as revenue by segment or personnel costs.
- Hide 12 monthsLet the tool forecast the last 12 months from the data before them.
- ComparePut AI forecast, your team's original forecast and actuals side by side.
- Ask whyCheck whether the tool names the drivers behind each deviation.
- Decide the roleUse the AI as baseline, as a second opinion or only for commentary.
Tools in this article
Frequently asked questions
Is AI forecasting more accurate than a manual forecast?
It depends on your data and business. A machine learning baseline extends patterns in past actuals, so it does well on stable, seasonal lines and poorly after a new decision it does not know about. Run a backtest on 12 months of your own history before relying on it.
How many FP&A teams use AI today?
In the AFP 2025 FP&A benchmarking survey of 362 professionals, 23 percent used AI and 40 percent were testing it. 92 percent still used spreadsheets every day.
Do I have to give up Excel for AI planning software?
No. Vena keeps Excel as its main interface, Cube works inside Excel and Google Sheets, and Jedox offers an Excel experience next to its browser version. Platform-first tools such as Anaplan build models in their own environment instead.
What does FP&A software with AI cost?
Most vendors quote on request. Some publish the structure, such as Jedox with a monthly fee per named user billed annually. LivePlan, aimed at small businesses, publishes $15 or $30 a month billed annually.
Sources
- Planful, AI in Planful
- IBM, Planning Analytics pricing
- Jedox, AI in Jedox
- Corporate Planning, AI features
- Anaplan, role based AI agents
- OneStream, SensibleAI Agents are now generally available (May 19, 2026)
- Valsight, driver based simulation
- CFO.com, Only 23% of FP&A practitioners are using AI (January 2025)
- ai-toolfinder.com dataset, budgeting and forecasting, as of September 26, 2026
- Vena, pricing
- Cube, pricing
- Jedox, locations
- Jedox, pricing
- Board, pricing
- Pigment, free trial sign up
- LivePlan, pricing (as of September 2026)
- Drivetrain, trust and security