ChatGPT Prompts for Finance: The RACEF Framework

Finance professionals who have tried ChatGPT and found it underwhelming almost always share the same story: they typed a vague question, got a generic answer, and concluded that AI is not ready for serious finance work. The problem is rarely the AI. It is the prompt. Poorly structured prompts produce poor outputs. Well-structured prompts, using a consistent framework, produce outputs that are genuinely useful for financial analysis, reporting, and commentary.

The RACEF framework was developed through extensive work with finance teams across the UK and MENA. It addresses the five most common prompt failures and gives you a repeatable structure that produces high-quality financial outputs across FP&A, reporting, audit, and analysis tasks. This guide explains the framework in detail, then gives you 20 ready-to-use prompts you can use today. For context on the broader landscape of AI tools, see our complete guide to AI use cases in finance.

Why Most Finance Prompts Fail

Most finance professionals approach AI prompting the same way they would type a search query: short, unstructured, and without context. “Write a variance analysis” or “explain why revenue is down” are typical examples. These prompts fail for predictable reasons.

No role definition. When you do not tell the AI what role it should adopt, it defaults to a generic response. It does not know whether it is talking to a CFO, an FP&A analyst, or a junior accountant, and it calibrates its response for none of them specifically. A senior finance professional needs a different level of analysis than a graduate trainee.

Ambiguous action. “Help with my cash flow forecast” is not a task - it is a category. Does it mean build the model, review the assumptions, identify the risks, or draft a board commentary? Vague actions produce vague outputs. The AI fills in the ambiguity with its best guess, which rarely matches what you actually needed.

Missing context. AI has no knowledge of your company, your industry, your current period performance, or your specific circumstances unless you provide it. Without context, every output is generic. With context, outputs become specific, relevant, and immediately usable.

No examples. Showing the AI an example of the output format or style you want dramatically improves consistency. If you want variance commentary written in a specific style - direct, numbered, maximum two sentences per item - showing the AI one example is more effective than describing it in words.

Undefined format. Without format instructions, AI outputs are often too long, insufficiently structured, or formatted incorrectly for your needs. Specifying the format - bullet points, table, three paragraphs, maximum 200 words - gives you outputs that are immediately usable rather than outputs that require significant reformatting.

The RACEF Framework Explained

RACEF addresses each of the five failure modes above. The framework has five components: Role, Action, Context, Examples, and Format. Every high-quality finance prompt includes all five, though the depth of each component will vary by task.

R - Role

The Role component tells the AI what perspective to adopt. In finance, this means defining the professional role (FP&A analyst, CFO, audit manager, treasury analyst) and often the seniority level. The Role component sets the tone, depth of analysis, and vocabulary of the response.

Example: “You are a senior FP&A analyst with 10 years of experience in manufacturing sector finance.” This single sentence changes the output significantly - you will get responses that use the right terminology, adopt the right level of technical depth, and approach the problem from the right professional angle.

A - Action

The Action component defines precisely what you want the AI to do. Use active verbs and be specific. Not “help with reporting” but “draft a two-paragraph board commentary explaining the revenue variance for Q4, identifying the three primary drivers.”

Strong action statements include: what you want created or analysed, how many items or how much detail, and the specific output the task should produce. Think of this as the job description for the specific task you are delegating to the AI.

C - Context

Context is where most prompts are shortest and where they should be longest. Context provides everything the AI needs to give you a relevant, accurate answer: your industry, your company size, the current period, the specific data, and any relevant background the AI needs to understand the situation.

Example context for a variance analysis prompt: “We are a UK-based professional services firm with £45m annual revenue. Q3 revenue came in at £10.2m versus a budget of £11.8m - a £1.6m shortfall. The main drivers are: lower-than-expected project wins in our legal sector practice (£0.8m), a delayed project start in our tech practice (£0.5m), and FX headwinds on our MENA contracts (£0.3m).” With this context, the AI can produce analysis that is immediately recognisable as relevant to your situation.

E - Examples

Examples are the single most underused element of finance prompting. Providing one example of the output you want - even a short one - dramatically improves consistency and quality. You can use examples to define tone (formal vs conversational), structure (numbered vs narrative), technical depth, and length.

For variance commentary, you might include: “Here is an example of the commentary style I want: 'Revenue underperformed budget by 14% (£1.6m adverse). The primary driver was lower legal sector project wins (£0.8m), attributable to elongated client procurement timelines in H2. This is expected to normalise in Q4 with three confirmed project starts.'” The AI will match this style for the rest of its output.

F - Format

The Format component specifies exactly how you want the output delivered. This includes structure (table, bullet points, numbered list, narrative paragraphs), length (word count or number of items), and any specific requirements (include a summary section, end with recommendations, avoid jargon).

Without a format specification, AI tends to produce long, essay-style responses even when you need a concise table. Always include format instructions. A simple addition like “Present as a table with three columns: variance driver, amount (£000), and recommended action” transforms the usability of the output.

20 Ready-to-Use Finance Prompts

The following prompts are pre-built using the RACEF framework. Each is ready to use - paste directly into ChatGPT, Claude, or Microsoft Copilot, replace the bracketed placeholders with your data, and you will receive a high-quality output. See our article on how to use AI in FP&A for more context on integrating these workflows.

FP&A Prompts

Reporting Prompts

Audit Prompts

Analysis Prompts

Advanced Prompt Techniques

Once you are comfortable with RACEF, two additional techniques will significantly improve your results: prompt chaining and iterative refinement. These are especially powerful for complex finance tasks such as building multi-scenario models or producing long-form analysis reports. The AI forecast commentary case study shows how chaining was used to automate a full board pack in under 30 minutes.

Prompt Chaining

Prompt chaining involves breaking a complex task into a sequence of simpler prompts, where the output of each prompt becomes the input for the next. This is more reliable than trying to do everything in a single prompt, and it allows you to review and correct at each stage.

A chained workflow for a budget variance report might look like this: Prompt 1 extracts and categorises the variances from raw data you paste in. Prompt 2 takes that categorisation and identifies the three most material drivers. Prompt 3 takes the drivers and drafts the narrative commentary. Prompt 4 takes the draft commentary and refines it to match your house style. Each prompt is simple and manageable; the combined output is a polished variance report.

Chaining also gives you more control over quality. If the output of Prompt 2 is not quite right, you correct it before passing it to Prompt 3. This is far more efficient than trying to get a single, complex prompt to produce a perfect output first time - and it mirrors how experienced finance professionals delegate work: breaking complex tasks into discrete steps and reviewing at each stage.

Iterative Refinement

Iterative refinement means treating the first AI output as a draft, not a final product. After the initial output, you give targeted feedback and ask the AI to revise specific aspects. This is analogous to the editorial process: you draft, you review, you refine.

Effective refinement prompts are specific. Rather than “make this better,” use: “The second paragraph is too hedged - rewrite it to be more direct and definitive” or “The tone is too casual for a board audience - revise to be more formal without losing clarity.” Specific feedback produces specific improvements.

You can also use refinement to add information you forgot to include in the original prompt: “Actually, also consider that the MENA revenue shortfall is temporary - a major project starts in January. Update the outlook paragraph to reflect this.” The AI will incorporate this new context while maintaining the existing structure.

For teams looking to build these skills systematically, Module 4 of the AI for Finance Leaders course covers prompt engineering for finance with 15+ hands-on exercises, including guided practice with RACEF and prompt chaining across real finance scenarios. Our AI consulting team also runs bespoke prompt engineering workshops for finance teams.

Frequently Asked Questions