Board decks are the most-time and least-fun deliverable in the CFO calendar. At the $180M SaaS company where I ran finance last year, I spent between 7 and 9 hours on the CFO-narrative sections of a monthly board pack. Not the analyst tables. Not the schedules. Just the narrative, the flux commentary, the KPI story, and the pre-read.

Six months ago I compressed that work to 45 minutes with a single templated prompt chain. The output is not “AI-generated content.” It is a first draft I edit in 30 minutes, then a fact-check pass I do in 15. Total from empty page to shareable draft: 45 minutes.

This is the prompt. Actually four prompts in a chain, with the inputs each needs and the review pattern I use.

The setup

Load these files into a Claude project (I use Claude for the long context; the same approach works in ChatGPT with Projects, but keep an eye on the token limit if your monthly package is over 60 pages):

• Last three months of the full board deck as PDFs
• Current month trial balance and prior-month trial balance
• Current month KPI report (bookings, ARR, gross margin, cash, headcount, pipeline)
• The prior month’s board Q&A transcript, if you have one
• Your original operating plan for the year

The context sits in the project. You do not re-upload it each month. You do refresh the last-three-months of decks quarterly so the tone drift stays honest.

Prompt 1: The flux commentary (about 10 minutes)

“Compare the current-month P&L to the prior-month P&L and to the operating plan for the same month. Return a flux narrative for each of: revenue, gross margin, operating expenses (broken by function), operating income, and cash. For each line, follow this format: ‘[Line] came in at $X, [up/down] $Y vs. plan and $Z vs. prior month. The variance was driven primarily by [specific cause with a number]. Secondary drivers: [list 1 to 2].’ Do not editorialize. Do not use words like ‘strong’ or ‘challenging.’ Every claim needs a number attached.”

The last sentence is the load-bearing one. Without it, the model produces the exact CFO-speak you are trying to escape. With it, you get numbers.

Output is 400 to 500 words. You edit in 8 to 10 minutes. The number-check is fast because every number is claimed against a source line.

Prompt 2: The KPI story (about 10 minutes)

“Using the KPI report and the trailing three months of KPI reports, produce a narrative of what is working, what is not, and what changed this month. Structure: (1) The 2 to 3 KPIs that improved most, with the delta and the likely driver. (2) The 2 to 3 KPIs that deteriorated most, with the delta and the likely driver. (3) The 1 to 2 leading indicators that suggest what will happen next month. Length: 350 words maximum. Every claim needs the number that supports it. Do not use qualitative words without a number underneath.”

The KPI narrative is where the CFO adds the most value in a board deck and where AI-drafted content most obviously fails without discipline. This prompt forces the discipline.

Common edit: I usually reorder to put the leading indicators first because that is what the board actually wants to hear.

Prompt 3: The board Q&A pre-read (about 10 minutes)

“Read the prior-month board Q&A transcript. Identify the 5 to 8 questions the board is most likely to ask this month, given (a) the questions they asked last month and (b) the flux and KPI narratives you produced above. For each question, produce: the question in board-language, a 3-sentence answer with numbers, and any supporting data the CFO should have at hand.”

This is the highest-leverage of the four prompts. Board Q&A is where CFOs make or lose their reputation, and preparing for it manually takes forever. Feed the model the actual Q&A transcript from last quarter and it will predict the current Q&A with something like 70 to 80% hit rate in my experience.

Edit time: usually 5 to 10 minutes to soften the tone and reframe one or two questions.

Prompt 4: The pre-read letter (about 15 minutes)

“Using the flux commentary, KPI story, and predicted board Q&A above, write a 1-page CFO pre-read letter. Structure: (1) The single most important thing that happened this month, in one sentence. (2) The 2 to 3 second-order things, one paragraph each. (3) The one decision or question I am putting to the board. Voice: direct, no hedging, no consulting-speak. Length: 400 words maximum. If you use the word ‘strong’ or ‘challenging’ or ‘headwinds’ or ‘tailwinds,’ start over.”

The pre-read is the entry point for the board. Get it right and the meeting is 45 minutes shorter. Get it wrong and you spend the first 20 minutes of the meeting orienting the board that already read the pre-read.

Edit time: 12 to 15 minutes. This is the piece I edit most because it is what the board reads first.

The full time budget

• Pull data and stage inputs: 5 minutes
• Prompt 1 (flux): 10 minutes generate + edit
• Prompt 2 (KPI): 10 minutes
• Prompt 3 (Q&A): 10 minutes
• Prompt 4 (letter): 10 minutes
• Fact-check pass, cross-reference every number to source: 15 minutes

Total: 60 minutes if you are careful, 45 if you have run it before. Down from 8 hours.

What this replaces

The 8 hours before this stack:

• 2 hours writing flux commentary from a blank page
• 1.5 hours writing KPI narrative
• 2 hours anticipating Q&A (or, more honestly, not doing this and winging it)
• 1.5 hours writing the CFO letter
• 1 hour of re-reading, cleanup, formatting

The prompt chain replaces the writing time. It does not replace the thinking. You still have to know why margin compressed. You still have to know which board member cares about churn cohort curves. The AI does not know your board. It writes the first draft based on what you know.

The counter-argument

“This produces homogenized CFO writing that sounds like every other AI-generated board deck.”

It can. If you paste the prompts as written and never edit the output, yes. The output has a shape and after two decks the board will notice.

The fix is the edit pass. The prompts are load-bearing but so is the 30 minutes you spend making the draft sound like you. The 12% CFOs I mentioned in Why Most CFOs Are Using AI Wrong all do this edit pass. The 88% do not.

The second objection: “The prompts hallucinate numbers.” I have not seen this happen once when the inputs are attached and the model is told to reference them. It happens frequently when the model is asked to reason about numbers you did not give it. Solution: give it the numbers.

The third objection: “This is one workflow. What about the tables and schedules?” Correct, this covers the narrative, which is 60 to 70% of my board deck effort. The tables come from FP&A tools like Cube or Mosaic that already automate them. If your tables still take five hours, that is a different problem.

Where this fits

The board pack workflow is one of six I run each month. The others are in The AI-Assisted CFO Weekly Cadence, which covers what happens Monday through Friday when the deck is not due.

The full templated prompts, plus the prior-month Q&A parsing script and the CFO letter voice guide, are in The AI-Native CFO Prompt Pack Pro.

For the analytical foundation this all sits on, The 2026 AI CFO Benchmark has the data on where CFOs are actually spending their AI budget in 2026.

Note: Company details in this piece have been anonymized. Any figures drawn from Spencer’s advisory work with middle-market and PE-backed finance teams are directional.