AI for Finance & FP&A
Every case follows the same loop: a worked example, an assignment you complete on realistic sample data, and AI review against the rubric below. Each case takes 30–45 minutes. This is the working syllabus for the founding cohort (early 2027); founding members can suggest changes.
Module 1. Clean Data, Fast: From Messy Exports to Analysis-Ready Tables
Clean a messy GL export
Your ERP dumped 8,000 GL lines with merged headers, text-formatted numbers, inconsistent account names and blank rows; the CFO wants spend by cost center by noon.
Your assignment
The cleaned table (or Power Query steps / Python script), plus the prompt chain used and a 5-line log of what was fixed and what was excluded.
What the AI reviewer checks
- Totals reconcile to the raw export (control total shown)
- Data types fixed (dates, numbers, signs for debits/credits)
- Account/cost-center mapping is consistent and documented
- No confidential data pasted into a non-approved tool (anonymized or used enterprise tool)
- Steps are repeatable next month (query/script, not manual edits)
Tools
ChatGPT/Claude (Advanced Data Analysis), Copilot in Excel, Power Query
Write and debug complex Excel formulas
You need a dynamic lookup across three tabs (actuals, budget, mapping) and a broken nested IF that returns #N/A on half the rows.
Your assignment
Working formulas (XLOOKUP/LET/LAMBDA or SUMIFS), a short explanation of each in plain English, and 3 test rows proving edge cases.
What the AI reviewer checks
- Formula returns correct values on provided test cases
- Handles errors/blanks explicitly (IFERROR/IFNA used deliberately, not to hide bugs)
- Uses ranges/tables robustly (no hard-coded cell refs that break when rows are added)
- Explanation would let a colleague maintain it
- Learner verified AI output rather than pasting blindly
Tools
ChatGPT/Claude, Copilot in Excel
Month-end close checklist & reconciliation review
Close is due in 3 days; you have a bank-to-GL reconciliation with 9 unmatched items and a 40-step close checklist.
Your assignment
Classify each reconciling item (timing, duplicate, mis-posting, error), write the follow-up list with owners and dates, and tighten the close checklist.
What the AI reviewer checks
- Every reconciling item classified with evidence
- Aged items (>30 days) escalated, not labelled timing
- Amounts tie out to the reconciliation
- Follow-ups have owners and due dates
- No real customer or bank data used in the prompt
Tools
ChatGPT/Claude, Copilot in Excel
Module 2. Variance Analysis & Management Commentary
Monthly budget-vs-actual variance commentary
Close is done; you have a BvA table for 12 cost lines and need board-ready commentary on the material variances by tomorrow morning.
Your assignment
Variance commentary for lines above a materiality threshold you define, with the prompt used and the threshold logic.
What the AI reviewer checks
- Materiality threshold stated and applied consistently
- Each comment explains driver (price/volume/timing/one-off), not just restates the number
- Distinguishes timing vs permanent variances
- Numbers in text match the table exactly (no hallucinated figures)
- Tone and length fit the audience (CFO/board): concise, action-oriented
Tools
ChatGPT/Claude, Copilot in Excel/Word
Price-volume-mix (PVM) bridge for revenue
Revenue is up 6% vs prior year but gross margin fell; the VP Sales and CFO want a PVM waterfall explaining why.
Your assignment
A PVM calculation (Excel or Python) with a waterfall chart and a 5-bullet executive summary.
What the AI reviewer checks
- PVM components sum exactly to total change
- Method (e.g., price effect at current volume) stated and applied consistently
- Mix effect computed at correct level of granularity
- Chart labels are clear and sorted logically
- Summary names 1-2 business actions, not just math
Tools
ChatGPT/Claude (data analysis), Copilot in Excel
Flux analysis on the balance sheet
Audit asks for explanations of every balance-sheet account that moved more than 10% and $50k quarter-over-quarter.
Your assignment
Flux table with flagged accounts and one-line explanations, plus list of items needing follow-up with owners.
What the AI reviewer checks
- Thresholds applied correctly (both % and $)
- Explanations are specific and verifiable, with supporting evidence noted
- Items AI could not explain are flagged, not invented
- Follow-up list has owner and question for each
- Format usable by auditors
Tools
ChatGPT/Claude, Copilot in Excel
Module 3. Forecasting & Driver-Based Models
Build a driver-based revenue & opex forecast
Leadership wants a 12-month rolling forecast driven by headcount, pipeline conversion and pricing instead of last year +5%.
Your assignment
An Excel driver model (inputs, calculations, outputs tabs) and a one-paragraph description of the key drivers and assumptions.
What the AI reviewer checks
- Inputs separated from calcs; no hard-coded numbers in formulas
- Drivers are logically linked to outputs (traceable)
- Assumptions documented with source/rationale
- Model balances and checks included (e.g., sum checks)
- Learner challenged at least one AI-suggested assumption
Tools
ChatGPT/Claude, Copilot in Excel
Scenario & sensitivity analysis
The CEO asks: what happens to EBITDA and cash runway if churn rises 2 points and hiring is delayed one quarter?
Your assignment
Base/upside/downside scenarios with a sensitivity table (data table or tornado) and a half-page memo.
What the AI reviewer checks
- Scenarios change drivers, not outputs directly
- Sensitivity ranges are realistic and justified
- Cash impact (not just P&L) addressed
- Memo leads with the answer and the decision implied
- Clear about limits and assumptions
Tools
ChatGPT/Claude, Copilot in Excel
Module 4. Reporting, Storytelling & Stakeholder Communication
Turn the monthly pack into an executive summary
The 25-slide management pack is ready; the CEO reads only the first page.
Your assignment
A one-page executive summary (top 3 messages, KPIs, risks, asks) generated with AI and edited by you.
What the AI reviewer checks
- Top messages are prioritized by business impact
- Every number traceable to the pack
- Risks and asks are explicit
- Plain English, no jargon
- Fits on one page
Tools
ChatGPT/Claude, Copilot in PowerPoint/Word
Answer an ad-hoc question from a business partner
The Head of Marketing asks why CAC rose 30% and whether to cut paid search; you have campaign spend and lead data.
Your assignment
A short analysis (table/chart) and a reply email with recommendation and caveats.
What the AI reviewer checks
- Correct metric definitions (CAC components)
- Analysis answers the actual question
- Recommendation is specific and tied to data
- Caveats and data gaps stated
- Email tone is collaborative, not defensive
Tools
ChatGPT/Claude (data analysis)
Module 5. Your AI Career Asset: Workflow, Policy & Proof
Build your personal finance AI playbook and prove the impact
You want to show your manager measurable impact and ask for a raise or promotion.
Your assignment
A reusable prompt library for 5 recurring tasks, a before/after time log, and a 1-page impact brief for your manager.
What the AI reviewer checks
- Prompts include role, context, format and verification steps
- Time savings measured, not guessed
- Respects company AI/data policy (what data can go where)
- Impact brief quantifies value in hours or $
- Ask (raise/scope/promotion) is concrete
Tools
ChatGPT/Claude/Copilot (whatever is approved)