This article is the first in a three-part series on getting started with Artificial Intelligence (AI) in FP&A.

I’m writing this for finance professionals at all levels because I continue to be amazed by the incredible technology now available to us — and by how little many finance teams are actually using it in practice. The 2025 FP&A Trends Survey shows that 53% of FP&A teams still do not use AI in any process today.

Across many conversations with FP&A professionals, I see a clear pattern: most people are aware of AI, but few feel confident enough to start using it in their daily work.

This short series is designed to change that.

The series will cover:
Part 1 — AI tools and why they matter for FP&A
Part 2 — How to prompt and use AI confidently
Part 3 — Practical examples of AI in FP&A 

Introduction

Eighteen months ago, I was an AI cynic. Whilst I was aware of the massive potential, I didn’t think it was real in the moment. I felt that it would be years before the concepts could be turned into reality. Then I started to explore. I had to create an ARR report urgently, and my usual method of ‘Googling’ how to do it was fruitless. So then I turned to a subscription version – ChatGPT Plus. It gave me the logic, how to source the data, and the formulae I needed in Power BI. I was blown away. I have been an AI evangelist ever since.

Over the last few months, I have networked with countless finance leaders. They all know about AI. But they are blissfully unaware of what they can do with it right now. Exactly like I was, eighteen months ago. I have had countless debates with myself. Do I keep this to myself and use it as my superpower? Or do I encourage those around me to learn and take this on?

As you can tell by the existence of this article, the latter argument won.

Imagine this scenario. 

The CFO walks into your office.

 "I need three scenarios modelled by end of day. I need a base case, recession, and aggressive growth. And the board wants to understand the cash runway implications of each." 

Five years ago, that request meant an all-nighter. 

Today, with the right AI tools, you can generate the first scenario structures, draft variance narratives, and even outline the board commentary before lunch.

The 2025 FP&A Trends Survey, which surveyed nearly 3,000 finance leaders, found that 53% of organisations still don't use AI in any FP&A process. Only 10% use it for forecasting or data analytics. Meanwhile, 46% of FP&A time is still spent on data collection and validation, leaving just 35% for generating high-value insights that actually drive the business forward.

There is a large gap between AI adopters and the rest of the population. Organisations using AI and Machine Learning (ML) rate their forecasts as "great or good" 65% of the time, compared to just 42% for the average organisation. That's not a marginal improvement. That's a competitive advantage.

This guide will not tell you how to roll out AI at an enterprise level. You need to learn to walk first. This guide is intended to get you started and improve your productivity. You will learn what these AI tools actually are, which ones matter for FP&A specifically, how to protect sensitive financial data absolutely, and how to write prompts that generate genuinely useful outputs — from forecast narratives to board-ready executive summaries.

1.1. What Is AI? Why FP&A Should Care?

A Large Language Model (LLM) is a complex prediction engine. It has been trained on trillions of words from books, financial reports, analyst commentary, and code. When you give it a prompt, it calculates the most probable sequence of words to respond with.

It doesn't think. It doesn't understand your business. It’s essentially a prediction engine. It guesses the next plausible word based on patterns in its training data.

The best mental model for FP&A: Think of AI as a brilliant intern who hasn't slept in three days. They are incredibly fast and well-read, but if you don't check their work, they will confidently present fabrications as fact. They can draft variance commentary, structure scenario analyses, and write executive summaries in seconds. But they've never actually worked in your business, don't know your specific drivers, and have no access to your data unless you provide it.

AI is effective at the communication and structuring layers of FP&A work. It is not a replacement for your understanding of the business, your relationships with budget owners, or your judgment about what the numbers actually mean.

1.2. The AI Tools for FP&A

I don’t know enough to provide an in-depth strengths-and-weaknesses evaluation. They all have their quirks and unique situations where they excel. Here's what's most relevant for financial planning and analysis work. Note: while 79% of FP&A teams have adopted AI tools to some degree, most are using them for quick operational wins, like Excel automation, rather than strategic transformation. The opportunity lies in going further and deeper.

Figure 1: Key AI Tools Currently Used by FP&A Professionals

The tools shown in Figure 1 represent some of the most commonly used AI platforms currently accessible to finance professionals. Each has different strengths, integrations, and security considerations. For FP&A teams, the goal is not to use all of them, but to understand which tools best support specific parts of the planning, forecasting, and analysis process.

The Big Four: General-Purpose AI Assistants

1. ChatGPT (OpenAI)
   The market leader and most versatile option. Strong at drafting narratives, explaining variances, brainstorming KPIs, and generating Excel formulas. The Enterprise-grade models can analyse uploaded spreadsheets and create data visualisations.

   FP&A sweet spot: Variance commentary, executive summaries, Excel formula generation, explaining financial concepts to non-finance stakeholders, and brainstorming forecast assumptions.

2. Claude (Anthropic)
   Excels at complex reasoning, technical writing, and analysing long documents. Has a larger "context window" — meaning it can hold more information in memory. This is important when you need to feed it an entire budget pack or board presentation for review.

   FP&A sweet spot: Reviewing lengthy strategic plans, synthesising multiple data sources into coherent narratives, detailed scenario analysis write-ups, and technical accounting memos.

3. Google Gemini
   Massive context window for processing multiple documents simultaneously. It has deep integration with Google Workspace. It can work directly with Sheets and Docs. Strong at cross-referencing information across sources.

   FP&A sweet spot: Firms using Google Workspace, research requiring cross-referencing multiple reports, and creating audio summaries of board packs for executives.

4. Microsoft M365 Copilot
   The cynic in me hesitated to include this. Often, it only makes the list because IT forces it on us. However, there are some benefits above the other LLMs.

   It has deep integration with Microsoft 365 (i.e., Excel, PowerPoint, Outlook, Teams). It can automate financial reports, generate Excel formulas, and VBA macros directly within your spreadsheets. It can also draft PowerPoint slides from data. It is also embedded in business platforms in the Microsoft Stack (e.g., Dynamics, Power BI). This is particularly relevant given that 52% of FP&A teams still use Excel as their primary planning tool (FP&A Trends Survey).

   FP&A sweet spot: Teams living in Excel and PowerPoint, automating monthly reporting packs, generating formulas for complex models, drafting board presentations. Copilot Studio is also a recent, useful addition with great potential to automate tasks.
    

The Rising Stars: Specialised & Emerging Tools

5. Perplexity AI
   This is the one I use for research. Unlike ChatGPT, Perplexity searches the web in real-time and provides answers with clickable citations. This is useful for FP&A professionals who need to verify market data, benchmark assumptions, or research industry trends.

   The Deep Research feature spends 2-4 minutes iteratively searching, reading documents, and synthesising findings into comprehensive reports. Perfect for competitive analysis, market sizing, or preparing industry context for board presentations.

   FP&A sweet spot: Industry benchmarking with sources, competitor analysis, validating market assumptions for forecasts, and researching macroeconomic indicators.

6. DeepSeek   
   An open-source model that is starting to make a name for itself in finance. It specialises in mathematical reasoning, structured analysis, and predictive modelling. Its strength lies in step-by-step reasoning explanations. This is valuable when you need to document your methodology or explain complex calculations to stakeholders.

   FP&A sweet spot: Complex financial calculations with documented reasoning, sensitivity analysis, scenario modelling logic, and firms needing on-premises AI for data security.

   Note: Yes, this is a Chinese model, and while I’m not here to pass judgment, be aware that many North American and European IT teams will flag it immediately as a data sovereignty risk.

7. Grok (xAI)
   Developed by xAI, Grok integrates with X (Twitter) for real-time data access. Strong in market analysis and competitive intelligence. Achieved 93.3% on the AIME 2025 math benchmark.

   FP&A sweet spot: Real-time market sentiment, monitoring competitor announcements, tracking industry news for forecast adjustments.

Purpose-Built FP&A & Finance Tools

Beyond general-purpose assistants, specialised AI tools are emerging specifically for financial planning. As this is a vendor-agnostic website, we will not review the products here.

The strategic choice: Start with one general-purpose tool (ChatGPT or Claude) for immediate productivity gains. Add Perplexity for research. Evaluate purpose-built FP&A platforms and point solutions when you're ready to transform your entire planning process. 

Note that 30% of organisations haven't upgraded their FP&A systems in over five years (FP&A Trends Survey). Don't let technical debt hold you back.

1.3. The Platinum Rule — Protecting Confidential Data

FP&A professionals handle some of the most sensitive data in any organisation: forward-looking forecasts, margin structures, pricing strategies, M&A plans, and headcount projections. Before using any AI tool, you must understand this:

THE PLATINUM RULE: Never input raw, unsanitised company data into a public AI tool. Ever.

What AI Companies Do With Your Data

By default, when you use free versions of these tools, your conversations may be used to train future models:

ChatGPT: Conversations used for training unless you disable "Chat history & training" in settings. Even then, retained for 30 days. Some conversations are reviewed by humans.
Claude: Opt-out by default — data used for training unless you disable it in privacy settings. Can be retained up to five years.
Gemini: Conversations used to improve services. Some are reviewed by humans. 18-month default retention.
Copilot: Consumer versions may use data for training; enterprise Microsoft 365 Copilot has stricter protections.
Perplexity: Recent policy confirms outputs not used for training. Enterprise contracts include stronger data guarantees.
Essential for FP&A: Enterprise tiers (ChatGPT Enterprise, Microsoft 365 Copilot for Business) operate under stricter privacy, where business data is not used for training. If your organisation uses these, check with IT about approved use cases. For personal/free accounts, assume everything you type could theoretically become public.

What Counts as Prohibited Data for FP&A?

Highly Sensitive (Never input to public AI):
Actual revenue figures, specific margin percentages, customer names and contract values, employee names and compensation, acquisition targets, pricing strategies, board materials, unreleased financial results, anything that would be material non-public information.

Moderately Sensitive (Anonymise before use):
Budget structures, KPI frameworks, variance categories, planning timelines, and general business context.
 

The Solution: Smart Anonymisation for FP&A

You can absolutely use AI for FP&A tasks, provided you anonymise the data first:

1. Indexing - Convert actual figures to index numbers (base = 100).
   Original: "Q1 revenue was £12.4M vs budget of £14.2M"
   Indexed: "Q1 revenue was 87 vs budget of 100"

2. Percentage-only - Use variances and percentages, not absolutes.
   Example: "Revenue was 13% below budget; EBITDA margin was 340bps lower than plan"

3. Generic categories - Replace specific names with types.
   Original: "Acme Corp contract worth £2.3M"
   Genericised: "Major enterprise customer, large seven-figure annual contract"

4. Hypothetical framing - Present as a teaching scenario.
   Example: "Imagine a SaaS company with these characteristics: ARR growth of 25%, gross margin of 75%, CAC payback of 18 months..."

1.4. The Biggest Risk - Confident Errors

AI "hallucination" is when a model generates information that sounds authoritative but is wrong. For FP&A, this manifests in several dangerous ways:

Invented benchmarks: "Industry average SaaS gross margin is 82%" - plausible, but possibly fabricated.
Misapplied formulas: Generating an Excel formula that looks correct but has subtle errors in logic.
Fabricated sources: Citing analyst reports or studies that don't exist.
Logical leaps: Concluding your data that aren't actually supported.
The rule: Never trust. Always verify. Test every formula. Check every benchmark. Validate every calculation. Use Perplexity when you need citable sources.

This matters because forecast quality directly impacts business decisions. The FP&A Trends Survey shows that 77% of organisations using driver-based models rate their forecasts as good or great, compared to just 27% of those without. AI can help you build better models, but only if you verify its outputs.