Building a Repeatable AI FP&A Workflow

You have now used AI across the whole FP&A cycle: forecasting, budgeting, driver modeling, variance commentary, scenarios, and decks. The final step is turning those one-off prompts into a system that runs the same way every month. A repeatable workflow is what separates "I used AI once and it was neat" from "AI saves my team a day every close." This lesson assembles everything into a durable, reusable FP&A operating rhythm.

What You'll Learn

• How to package your best prompts into reusable assets
• How to build a custom GPT or Claude Project tuned to your company
• A month-by-month AI-assisted FP&A workflow you can adopt immediately
• How to keep the system accurate, safe, and trusted over time

Package your prompts as assets

Right now your prompts probably live in scattered chat histories. Collect the ones that worked into a single prompt library, organized by task: a variance-commentary prompt, a forecast-narrative prompt, a board-outline prompt, a scenario-definition prompt. For each, save the full text including your rules, such as materiality thresholds and tone. A prompt you can paste in 10 seconds is one you will actually use; a prompt you have to rewrite each month is one you will abandon.

Store the library somewhere your team can reach it, so the whole function benefits and new analysts inherit a working toolkit on day one.

Build a custom GPT or Claude Project

The next level up is a custom assistant that already knows your context. Both ChatGPT (custom GPTs) and Claude (Projects) let you set persistent instructions and upload reference material, so you stop re-explaining your business every session.

Give yours:

• A role and rules. "You are the FP&A analyst for a mid-size B2B software company. Always keep math in the spreadsheet; never invent drivers; flag unknowns. Use a neutral, factual tone."
• Reference context. Upload your chart of accounts structure, your driver definitions, your house style for commentary, and a sample of strong past commentary. Apply the data rules: use anonymized or templated examples, not live sensitive figures, unless the tool is an approved in-tenant deployment.
• Standard formats. Your variance-commentary format, your deck outline, your scenario template.

Now a single instruction like "write this month's variance commentary from the attached table" produces output already shaped to your standards. This is the highest-leverage setup in the whole course, because it bakes your guardrails and style into every interaction.

Your monthly AI-assisted workflow

Here is a complete rhythm that ties the course together. The spreadsheet does the math at every numbered step; AI does the structuring and writing.

Early in the cycle (actuals just landed):

1. Calculate plan-vs-actual variances in Excel.
2. Add your known drivers as short notes.
3. Run your variance-commentary prompt; confirm flagged unknowns.

Mid-cycle (updating the view):

4. Drop actuals into the rolling forecast; let Excel compute the deltas.
5. Ask AI to flag the biggest misses and propose reforecast methods; apply the chosen numbers yourself.
6. Red-team the updated assumptions with AI from both directions.

Late cycle (communicating):

7. If leadership wants options, define scenarios with AI and recalc them in your driver model.
8. Build the board or management deck: story spine, message headlines, talking points, executive summary.
9. Rehearse the toughest questions with AI before the meeting.

Quarterly and annually:

10. Reuse the budget-build prompts for planning season, and refresh your custom assistant's reference context as the business changes.

Adopt even half of this and you reclaim hours every cycle, while the judgment-heavy parts of the job get more of your attention.

Keep it accurate and trusted

A workflow is only valuable if people trust its outputs. Protect that trust:

• Math stays in the spreadsheet. This is the rule that has run through the entire course. AI writes and structures; your formulas compute. Never ship an AI-calculated number.
• No invented drivers. Keep the "flag, do not guess" instruction in every commentary prompt. A confident wrong reason is worse than an open question.
• Human review before anything ships. AI drafts; you sign off. Your name is on the report, so read every word.
• Data rules every time. Run the Module 1 pre-flight check on every session. Sensitivity does not relax just because the workflow is routine.

Measure the payoff

Track the time you save, in plain terms: hours per close, days per budget cycle. This does two things. It tells you which parts of the workflow are worth refining, and it gives you a concrete story when leadership asks what AI is doing for the finance team. FP&A should be able to quantify its own productivity gains better than anyone, so do it.

Where to go from here

You now have a complete, safe, repeatable system for applying AI across corporate FP&A. The tools will keep improving, but the operating model in this course will hold: clean data in, math in the spreadsheet, AI for structure and language, human judgment on top, and guardrails every time. Build your prompt library, stand up a custom assistant, run the monthly rhythm, and refine it as you go.

Key Takeaways

• Collect your best prompts into a reusable library organized by task, including your rules and tone.
• Build a custom GPT or Claude Project loaded with your role, rules, reference context, and standard formats.
• Adopt the month-by-month workflow: commentary, then reforecast, then scenarios and decks, with math always in the spreadsheet.
• Protect trust with human sign-off, no invented numbers or drivers, and the data pre-flight check every session.