Board reports get written every quarter by the most senior people in the business. They take 6–10 hours of senior time to produce. AI compresses that to 90 minutes — but only if you use it for the right parts and write the parts AI can't handle yourself. Here's the split.
One-time setup. Create a Claude Project ("Board Reports — [Company]") and load:
— Last 4 quarters of board reports (anonymized if needed)
— Your standard report structure (sections, length, tone)
— Your KPI definitions and historical baselines
— Board member context (what each member typically asks about)
System prompt: "You are drafting board materials. Voice: formal but direct. Always cite the numbers. Never invent data. If a data point is missing, write [TK: data needed from X]. Match the length and depth of prior reports in the knowledge base."
This is where AI is most valuable. Paste in the actual quarter's numbers + commentary from the team about why each line moved. Ask Claude to turn it into board-ready prose.
Here are this quarter's revenue numbers vs. plan: [paste]. Here's commentary from the team on each line: [paste]. Draft the "Revenue Performance" section of the board report. Match the structure and depth of last quarter's. Keep it to 350 words.
This single section typically takes a CFO 90 minutes. AI cuts it to 15 minutes of editing.
For each KPI on the board deck — NRR, CAC, pipeline coverage, customer count, headcount — provide Claude the number and a sentence of context, and ask for the commentary. Generates parallel paragraphs in seconds.
Then read each one. Maybe 30% need real editing. The rest are 80% ready to ship.
The CEO summary, the strategic framing, the "what we're betting on next quarter" section — these are not AI work. They require judgment, context the AI doesn't have, and a level of accountability the AI can't carry.
The right split: AI does 70% of the prep work (numbers commentary, variance narratives, KPI sections), you do 30% (the strategy, the open questions, the asks of the board). The 30% is what the board actually reads.
Pass 1 — Numbers. Every cited figure gets verified against your source-of-truth dashboard. AI will occasionally invent or transpose digits.
Pass 2 — Named claims. Anywhere the report claims you did something, achieved something, or signed something — verify. AI sometimes invents wins.
This adds 20 minutes. It's non-negotiable for any document going to a board.
1. Letting AI write the strategic narrative. AI can write "what happened." It cannot write "what we should do next." Don't let it try.
2. Trusting numbers without verification. AI will occasionally transpose digits or invent figures. Always verify against source-of-truth dashboards.
3. Using ChatGPT in a fresh chat each quarter. Without prior reports loaded, you get generic board-report-shaped content. Use a Claude Project with knowledge base loaded.
4. Pasting raw financials into consumer AI. For sensitive financials, use Claude Enterprise or API with zero retention. Or anonymize.
For most CFOs, 5–7 hours per quarter. The variance narrative and KPI commentary sections are where 90% of the savings happen.
Claude. Specifically Claude Team or Enterprise. The Projects feature lets you load prior reports as reference, which dramatically improves consistency quarter-over-quarter.
Use Claude Team or Enterprise with audit logs enabled. Document your AI policy. Most boards don't object to AI-assisted drafting; they object to unverified AI-generated facts.
Generate a first draft, then heavily rewrite. The CEO letter is voice-and-judgment work. The first draft saves you 15 minutes; the heavy edit is the actual work.