Free resource in French · Word and Excel Kit
AI Finance Kit: 24 use cases and 48 prompts for finance teams
A four-stage course for management controllers, accountants, treasurers and FP&A analysts to develop their skills independently, from the first variance comment to closing agents, with an Excel dataset to practice without exposing their figures.
Conceived by Flavien Besset, founder of NextStart.AI · updated October 26, 2026 · 5,350 professionals trained · satisfaction 9,2/10
The AI Finance Kit, in brief
- What it is
- A self-training course in generative AI for finance teams: 24 use cases, 48 quick-to-copy examples, a challenge solved in stages and a training dataset.
- For who
- Management controllers, accountants and closing managers, treasurers and credit managers, FP&A analysts, internal auditors, CFOs and finance managers of large organizations.
- Format
- A 71-page Word document and a five-tab Excel workbook (read me, monthly profit and loss statement, supplier invoices, accounts receivable aging report, legacy forecast to be audited), with entirely fictitious data.
- What it contains
- A positioning test, a "Where to start" selection, the rules of the game (data, verification, which tool for which task, risks specific to finance, AI Act and GDPR), the prompt finance method in eight elements, 24 sheets in four levels, four corrected challenges, career paths, a 30-day deployment plan, a glossary and a progress log.
- Tools
- Copilot Chat and Microsoft 365 Copilot (often called "Copilot Premium"), Copilot in Excel, ChatGPT, Claude, Gemini, Vibe by Mistral (formerly Le Chat), and AI integrated into ERP and EPM systems. Prompts are tool-independent; each entry indicates the recommended tool and the differences between them.
- What is it based on?
- The 2026 surveys on AI in finance departments (CFO Connect, Consero), the Corporate Finance Institute's finance prompt libraries, and our practice of AI workshops in companies.
- Price
- Free, in exchange for a professional email address. No unsolicited emails.
- Update
- Version 2 of October 26, 2026.
- What it doesn't replace
- Neither your organization's usage charter nor the validation by your IT department of the tool and authorized data.
- Go Further
- AI training for finance teams, AI Measurement and ROI Kit, AI charter template.
Many finance departments no longer doubt the value of generative AI; they hesitate on the starting point and how to control its output. This kit addresses both of these concerns with a structured learning path. Four levels, six fact sheets per level, a challenge at the end of each level, and a fictitious dataset for practice without revealing any of your company's figures.
1. Why a path, and not just another list of prompts?
There is no shortage of prompt lists for finance. Almost all of them have two flaws. They treat a variance comment and a closing agent as equally valid, as if they required the same maturity. And they rarely explain how to verify the result, whereas in finance, a plausible answer is not necessarily the correct one: a reversed sign, a cumulative total that extends beyond a month, or thousands read as units can easily go unnoticed in a well-phrased sentence.
The 2026 surveys on AI in finance departments (CFO Connect, Consero) all point to the same starting point: management reporting and variance analysis, tasks that every finance team performs monthly. That's why the toolkit now offers a "Where to Start" selection of six fact sheets (1.2, 1.6, 2.1, 2.2, 2.5, 2.6) for a useful first month.
This is the order of the kit. We start by writing and commenting on a few handwritten figures, without any sensitive data. We then move on to anonymized files, then reusable prompts and shared assistants, and only at the end to agents who perform multiple steps on cases. At each stage, the checkpoints are as important as the prompt.
2. Four levels, from drafting to agents
A ten-question test at the beginning of the kit indicates each person's starting level. Allow one and a half to two and a half hours per level; in a team, the most effective pace is one level per week with one hour of group practice.
| Bearing | What you learn | Case examples | Duration |
|---|---|---|---|
| 1. Writing and understanding | Frame a prompt, work on text and some aggregated figures, check an answer. | Commenting on a discrepancy, simplifying a financial statement, writing graduated reminders. | 1 hour 30 minutes |
| 2. Analyze your own data | Attach an anonymized file, check its quality, prioritize discrepancies, comment on a report. | Compare actual, budget and forecast; estimate the landing; check an invoice. | 2 h |
| 3. Structure and reuse | Write variable-based prompt templates, chain the steps, create a persistent assistant shared by the team. | Rolling forecast, budget and arbitration, budget review, review of auditors. | 2 h |
| 4. Delegate to agents | Describe a mission, impose a quality protocol, maintain human validation at each stage. | Closing assistant, collection monitoring, invoice controller, Excel template auditor. | 2 hour 30 minutes |
3. The 24 use cases, family by family
Each level covers the six work families of the finance function, those we use in our AI finance training. The same family can therefore be found at four levels of maturity: reporting begins with the popularization of a financial statement and ends with an agent who produces the report with commentary from beginning to end.
Each sheet follows the same template: the objective, when to use it, the estimated gain, the necessary data and those to never paste, the recommended tool, the prompt to copy, a variant, the control points of the result, the classic trap and the bridge to the next level.
For each case, here is the objective, the estimated gain, the first point to check, and the deliverable obtained. The complete prompt and its variant are in the kit, and Sheet 1.2 is reproduced in full below..
Writing and understanding
Making a financial statement accessible to a non-financial audience
- A profit and loss statement explained in ten lines to a site manager
- Estimated gain: 30 to 45 minutes per note
- To be verified: each discrepancy mentioned, its meaning and its basis
Deliverable: a ten-line memo
Commenting on a discrepancy based on aggregated figures
- Observation, probable causes, and question to ask, based on five lines of work completed and budget.
- Estimated gain: 15 to 20 minutes per commented post
- To check: the gaps recalculated before gluing
Deliverable: one discrepancy comment per item
Draft graduated payment reminders
- From a simple reminder to a formal notice, with the right tone for the right client.
- Estimated time saved: 20 to 30 minutes per case
- To verify: number, amount including tax and due date in the ERP
Deliverable: three coherent follow-ups
Documenting the assumptions and risks of a budget
- The page everyone's asking for before the Executive Committee meeting, but no one has time to write.
- Estimated saving: one to two hours per budget
- To be verified: each assumption must be identical to the budget file.
Deliverable: one page outlining assumptions and risks
Prepare the file validation rules before opening it
- The checks to be performed on each column, before discovering errors in the meeting
- Estimated saving: 30 to 45 minutes per file
- To be verified: each rule can be executed as is in Excel
Deliverable: a reusable checklist
Summarize a regulatory change or an internal procedure
- What's changing, who's affected, and when does it take effect?
- Estimated time saved: 45 minutes to one hour per text
- To be verified: each date and each threshold in the source text
Deliverable: a one-page document for the team
Analyze your own data
Checking a dataset and ensuring the reliability of a legacy file
- Empty cells, duplicates, signs, formats and orders of magnitude reviewed
- Estimated saving: 30 to 60 minutes per file
- To be verified: the expected number of lines, recalculated manually
Deliverable: a data quality report
Compare actual results, budget, and forecast, and prioritize the variances.
- Differences ranked by impact in euros, with the five to be commented on first.
- Estimated gain: one hour per closing
- To check: the direction of the discrepancies in revenue and expenses
Deliverable: a prioritized gap analysis table
Estimate the year-end landing based on the actual
- By site and by position, with an explicit methodology and budget variance
- Estimated gain: half a day on the first version
- To be verified: completed by date plus remaining to be done equals the landing
Deliverable: a reasoned annual landing
Detect spending anomalies
- Unexplained spikes, trend breaks and persistent overshoots highlighted
- Estimated gain: one to two hours per month
- To verify: the percentage is based on the budget, not the actual amount.
Deliverable: the list of lines to check
Produce a monthly report with commentary
- Three to five factual sentences per site, with the exact figures from the file
- Estimated time saved: 45 minutes to one hour per report
- To be verified: each figure cited, referenced in the source table
Deliverable: the comments page for the monthly pack
Check a supplier invoice before validation
- Calculations, dates, order, receipt and supplier verified in two minutes
- Estimated time saved: five to ten minutes per questionable invoice
- To be verified: Price excluding VAT plus VAT recalculated, rounded to the nearest cent
Deliverable: a completed control grid
Structure and reuse
Automate the preparation of monthly data
- An assistant that always applies the same checks, plus a Power Query step
- Estimated gain: one to two hours per month starting from the second month
- To check: the rows counted before and after Power Query
Deliverable: a ten-minute check-up routine
Update a rolling forecast with scenarios and sensitivity
- The same quick-model chained every month, two scenarios and one sensitivity
- Estimated gain: half a day per update
- To check: the cumulative total recalculated using SUMIFS
Deliverable: an updated landing plan and its scenarios
Prepare a budget review with an operational manager
- Five key gaps, quantified, and open questions that are generating discussion
- Estimated saving: one hour of preparation per review
- To be verified: the figures in the script, down to the last euro
Deliverable: a 30-minute review script
Prepare the annual budget and make informed decisions.
- A proposal per site and per position, with written assumptions, in relation to the target margin
- Estimated gain: two to three days on the first version
- To be verified: the budgeted margin recalculated site by site
Deliverable: a budget and its arbitration note
Transforming a dashboard into recommendations and a presentation
- A clear message per indicator, with an assistant who knows your KPIs
- Estimated time saved: two to three hours per committee
- To be verified: two KPIs recalculated from the source
Deliverable: a page of recommendations and a presentation plan
Prepare the auditors' review
- Up-to-date parts list, explanations of fluctuations, and compliance note
- Estimated gain: one to two days of preparation
- To be verified: each request from last year must be renewed or justified.
Deliverable: an audit preparation file
Delegate to agents
Monthly closing assistant
- Following up on the checklist, morning progress update, and draft follow-up messages, without sending anything on my own
- Estimated gain: 30 to 45 minutes per closing day
- To check: the point of day, predicted by hand and then compared
Deliverable: a daily progress update
Vigie debt collection and customer receivables
- Each week, the aging report is reviewed, the customer risk summarized, and follow-up notices prepared.
- Estimated gain: one to two hours per week
- To be verified: three known customers, categorized as you would have done
Deliverable: a summary of the risk and follow-up actions to be validated
Supplier invoice controller
- Supplier, amounts, order and receipt reconciled, exceptions routed, no payment
- Estimated time saved: one to two hours per batch of approximately one hundred invoices.
- To be verified: a test on five invoices, two of which are fraudulent.
Deliverable: a controlled batch and its exceptions
End-to-end annotated monthly report generator
- File check, discrepancies, comments in our style and support, two validations
- Estimated gain: half a day per month
- To check: the consolidated margin calculated manually before launching
Deliverable: the report and its presentation, ready for review
Excel Model Auditor
- Inconsistent formulas, hard-coded values, broken references, and identified loops
- Estimated gain: half a day per workbook
- To be verified: two intentionally introduced defects, both of which were found.
Deliverable: an audit report of the workbook
Rolling forecast agent
- Integrated execution, recalculated landing, three scenarios, each version retained
- Estimated gain: two to three hours per month
- To be verified: the consolidated landing, to the nearest thousand euros
Deliverables: a versioned forecast and a report for the CFO
4. An example of a worksheet: commenting on a discrepancy
Here is form 1.2, as it appears in the kit. It does not require any files: five lines of handwritten information (actuals and budget) are sufficient. When to use it: On the third day of closing, the CFO asks why purchases exceed the budget on a site and you have ten minutes to offer an initial comment. Estimated profit: 15 to 20 minutes per post commented on, and a framework of causes that prevents forgetting a lead.
Act like an experienced industrial management controller preparing the monthly closing review.
Here are five lines of actual/budgeted data [site] to [period], in euros:
[Position 1] : accomplished [Rising], budget [Rising], gap [Rising] or, [percentage]. [Positions 2 to 5: same structure, one line per position].
What I already know: [two or three facts, for example: 5% rise in the price of steel in September, sales volume 7% higher than budget].
For each position where the gap exceeds [threshold, for example 3% or €20,000]write a comment in three parts: numerical observation, probable causes ranked from most likely to least likely, question to ask the operational person to confirm.
Always distinguish between volume effect, price effect and exceptional effect in the causes, and indicate which one is confirmed by a fact that I gave you.
Format: one paragraph per post, 60 words maximum, difference in euros and percentage in the first sentence.
Do not propose any cause that contradicts the facts provided, and never assert as certain a cause that I have not confirmed: mark it "to be verified".
If the discrepancies I gave you do not correspond to what was achieved under budget, please point this out before making any comments.
The outcome control pointsbefore transmitting anything:
- Recalculate the five variances before pasting: actual minus budget, and the percentage based on the budget. A sign error in the input results in an inverted comment.
- Cross out any cause that the AI presents as certain when it is not part of your facts: it remains a hypothesis until an operational person has confirmed it.
- Check that the required threshold is respected: a post below the threshold should not be commented on.
- Add up the differences in the commented posts and compare them to the site's margin difference: if they do not explain the essentials, a post is missing.
The classic trap: The AI suggests plausible causes with the assurance of a senior controller, even without any facts. "Rise in raw material prices" comes up almost every time, whether it's true or not. The comment is only valid if each cause is linked to a fact or marked "to be verified." At level 2, form 2.2 performs the same task across all positions and sites using a file, and prioritizes the discrepancies for you.
5. A risk-free training dataset
The kit comes with an Excel spreadsheet for Solveo Industries, a fictional mid-sized industrial company with €48 million in revenue and three sites that also serve as three business units (Lyon, Nantes, Lille). The file can be attached as is to any AI tool: it contains no confidential or personal data. It serves as the basis for the challenges of levels 2, 3, and 4.
- PL_monthly: the 2025 management profit and loss statement, one line per month, site and position (360 lines), in actual, budget and revised mid-year forecast.
- Supplier invoices: 121 lines of invoices received in the last quarter, including 48 in the status "To be validated", as extracted from a dematerialization tool, with order, receipt and truncated IBAN.
- Balance_agee_clients: 60 clients as of December 31st, with seniority brackets, disputes, final reminder and comments from the credit manager.
- Forecast_herite: new tab of version 2, a legacy landing workbook whose formulas read PL_monthly, to be audited with the variant of sheet 2.1 and sheet 4.5.
The workbook contains 25 anomalies and risk signals intentionally inserted into the three data tabs (data entry errors, duplicates, VAT inconsistencies, IBAN changes, and clients in receivership), plus three construction flaws in the Forecast_herite tab. In version 2, dated October 26, 2026, the surge in purchases in Lille is concentrated in the second half of the year, as indicated in the challenges, and the ten invoices with anomalies fall within the agent's scope; review the kit and the workbook together. The challenges involve identifying these anomalies using AI, then comparing what your controls caught with what they missed. The answer key, with the expected figures, is included as an appendix to the kit.
6. The rules of the game, before the first prompt
The kit dedicates its first part to what makes the use of AI acceptable in a financial department. Three categories of data first: what is public or fictitious and is used everywhere, what is internal but anonymizable and only goes into a company tool, and what never goes into an AI tool without written agreement (individual remuneration, IBAN, mergers and acquisitions, unpublished results).
Five verification steps, repeated in the checkpoints of each sheet: manually recalculate key figures, request the source line by line, check the sign, period, unit, and entity, test on a known case before launching a full process, and have all text originating from the finance department proofread. Finally, a list of "never": closing accounts, making an accounting decision, or drafting regulated financial communications independently remain outside the scope of AI, regardless of your level, and no IBAN changes or payment releases are made based solely on a message, call, or video conference. Two new sections complement these rules: which tool for which finance task (Copilot Chat, Agent mode and function =COPILOT() in Excel, Analyst and Python in Excel, ChatGPT, Claude, AI of ERP and EPM), and the seven risks specific to finance with their countermeasures (calculation error, invented figure, CEO fraud by deepfake, separation of duties, instructions hidden in documents, audit trail).
Since February 2, 2025, Article 4 of the European AI Regulation has addressed staff proficiency in using AI. Since Omnibus Regulation (EU) 2026/1744, which came into force on July 27, 2026, it no longer sets a "sufficient level" but requires measures to support this proficiency among employees and contractors who use AI on your behalf. Credit ratings of individuals classified as high risk (bonds as of December 2, 2027) remain outside the scope of this toolkit. Following the toolkit and completing the progress log allows a finance team to demonstrate what it has learned. To formalize these rules across the company, see our AI charter template.
7. What the kit contains
Version 2, updated October 26, 2026: the Solveo dataset is more reliable (Lille drift in the second half of the year, new Forecast_herite tab, numerical corrections), the October 2026 tools are clearly defined (Copilot Chat or Microsoft 365 Copilot, often called "Copilot Premium", Agent mode and =COPILOT() in Excel, Analyst), and an eighth methodological element, Proof, mandates calculation by code or formula. The kit is a 71-page Word document accompanied by a five-tab Excel workbook. It contains, in order: a ten-minute start with an initial prompt, the "Where to Start" selection, the path map, and the placement test; the rules of the game, including which tool for which task and the risks specific to finance; the eight-element finance prompt method, with a before-and-after example, six habits, and an independent review prompt; the four levels, each with its introduction, its six sheets, its challenge and its checklist "you are ready for the next step if"; the recommended paths for seven finance jobs; a 30-day deployment plan for the team leader, with the weekly ritual, the shared prompt library and the indicators to follow; and six appendices: index of the 24 cases and the 48 prompts, description of the dataset, solutions to the challenges, glossary, progress log and sources.
Download the AI Finance kit (Word and Excel, free)
71 pages, 24 use cases, and 48 prompts organized into four levels, with the Solveo Industries training dataset (five tabs). Written by NextStart.AI based on the 2026 surveys on AI in finance departments and the best finance prompt libraries.
- Positioning test and rules of the game
- 24 cards: prompt, variant, checks, trap
- Four challenges solved using fictitious data
- Career path by profession and 30-day deployment plan
8. Frequently Asked Questions
Do you need to know about AI to use the kit?
No. Level 1 starts from scratch: a first prompt in ten minutes, without files or sensitive data. The placement test at the beginning of the kit allows those already familiar with the material to start at level 2, 3, or 4.
What AI tool is needed?
The one your organization allows. The prompts work with Copilot Chat and Microsoft 365 Copilot (often called "Copilot Premium"), ChatGPT, Claude, Gemini, and Mistral's Vibe (formerly Le Chat); a section of the toolkit indicates which tool to choose for which finance task. Each sheet indicates the most suitable tool and what changes from one tool to another, particularly for file analysis functions and Tier 4 agents.
Can we use the prompts with our real data?
Yes, starting at level 2 and within a business tool, after anonymization: names are replaced by codes, and IBANs and personal data are removed. Each sheet specifies what should never be pasted. To learn, start with the fictitious dataset provided with the kit.
How long does it take to go through the kit?
Approximately eight hours of practice in total: one and a half hours for level 1, two hours for levels 2 and 3, two and a half hours for level 4. In a team, one level per week with one hour of common practice works well.
Is the kit suitable for an entire team?
Yes. Its final section is aimed at the team leader: a 30-day plan, a weekly one-hour ritual, a shared prompt library, and five indicators to measure before and after. The job-specific career path chart prevents everyone from having to complete all 24 scenarios.
Can the kit be used without assistance?
Yes, it is designed for self-directed skills development. We intervene when a team wants to practice on its own projects with a facilitator, during a AI finance training half a day.
9. Going further
The kit facilitates skills development; it does not replace guided practice on your own cases. Our AI finance training Get your controllers, accountants, and analysts working in pairs on their real-world cases for half a day, using the tool validated by your IT department. For banks and insurers, see also the AI training for banking and financial services.
Read also : Measuring the uses, value and ROI of AI: the 2026 toolkit, AI Charter in Business: The 2026 Model (Downloadable), AI systems governance: the 2026 toolkit, AI Maturity Grid: Assess your organization's readiness in 90 minutes et AI Ambassador Kit: Build your internal network in 90 days.
The entire series of free kits. To drive adoption: AI charter template, AI training plan template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, AI-powered time-saving kit, Soft Skills Kit for AI, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit, AI Governance Kit, Cowork kit, 50 Copilot M365 prompts, Copilot Studio kit, Deploy Microsoft 365 Copilot kit, Copilot Chat Kit, Copilot Excel kit, Copilot Teams and Outlook kit, Copilot Word and PowerPoint kit, Configure Claude kit, Deploying Claude in the Workplace Kit, Claude Agents Kit, Claude Skills kit, Claude's Marketing Skills, Skills Claude HR, Claude Plugins Kit, Claude Code kit, Claude Design kit et Vibe coding kitFor the AI project sponsor: AI Act kit, GDPR kit, Security and data kit, Shadow AI kit, AI and Social Dialogue Kit, frugal AI kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI kit for lawyers, AI kit for salespeople, AI toolkit for project managers, AI innovator kit, communicator's AI kit, AI Procurement Kit, AI Kit for Researchers et AI kit Customer service.
10. Sources
The figures and references cited on this page and in the kit are based on the following sources.
- CFO Connect, State of AI in Finance 2026: report findings and what they mean for CFOs, 2026.
- Consero, 2026 CFO Survey, survey conducted in the first quarter of 2026 among 102 financial executives.
- Corporate Finance Institute, 15 Best AI Prompts for Finance: FP&A, Treasury, and More, 2026.
- Corporate Finance Institute, 18 Best AI Prompts for Accounting: Workflows, Examples, and Guardrails, 2026.
- CFO Connect, 25 Claude Prompts for Finance Teams: CoworkCode & FP&A, 2026.
- Jaydai, AI use cases in finance: 12 examples for CFOs, 2026.
- JMCO, How Finance Teams Are Using Claude, Copilot, and ChatGPT Inside Excel, 2026.
- DFCG, AI and digital transformation at the service of management controllers, 2026 program.
- European Union, Regulation (EU) 2024/1689 on artificial intelligence, Article 4 "Control of AI", June 13, 2024, amended by Regulation (EU) 2026/1744 (omnibus IA), in force since July 27, 2026.
- CNIL, Artificial intelligence: CNIL recommendations, accessed in September 2026.