Board-ready financial summaries drafted from your raw numbers

The financial narrative for your board deck, drafted from your actual statements in your usual format, so you spend your time on judgment and fact-checking instead of staring at a blank page.

What you'll have when you're done

A first draft of the financial section of your board update, written in your format and your voice, built from the numbers you uploaded. Highlights, variances, the story behind the quarter. You move from authoring (the slow, blank-page part) to editing and fact-checking (the part where your judgment actually adds value). The weekend-eating task becomes a focused hour.

Writing the board narrative is the part that always slips to Sunday night

The numbers are done. The board meeting is Tuesday. And somewhere between those two facts is the job I always underestimate: turning the financials into a narrative, what happened, why, what it means, what we are doing about it, in the format the board expects. It is not hard work exactly, but it is slow, and it has a way of landing on Sunday night.

The leverage here is specific and it comes with a sharp warning. AI is genuinely good at the drafting: given your statements and an example of your past memo, it will write a clean, structured financial narrative in seconds. But a board document is the one place a hallucinated number does real damage to your credibility, so the rule for this workflow is stricter than any other in the finance set: the AI writes the words around your numbers, it never generates the numbers. Get that boundary right and this is a weekend back. Get it wrong and you read a wrong figure aloud to your board.

What you need first

• Your financial statements for the period: P&L, balance sheet, and the prior period for comparison, exported clean.
• A template: last quarter's board memo, or your standard deck's financial section, so the AI matches your format and voice.
• A Claude Project on a business plan. Board financials are about as sensitive as company data gets, never on a personal account, and strip or avoid any customer or employee names that do not belong in the summary.

Step-by-step

Step 1Assemble the inputs

Gather the period's P&L and balance sheet, the prior period for comparison, and your template memo. The template is what makes the output usable, without it you get a generic summary; with it you get your summary. Upload all of it to your finance Project.

Step 2Set the hard boundary first

Before you ask for anything, set the rule that makes a board document safe:

You are drafting the financial narrative for my board update. Absolute rules:
- Use ONLY the numbers in the files I uploaded. Never calculate a figure I didn't
  give you unless I explicitly ask, and when you do, show the arithmetic.
- If a number I'd expect isn't in the files, flag it as [MISSING: x] and do not fill it in.
- Match the structure, tone, and length of the template memo.

This is the same reviewer-discipline from the rest of the finance set, dialed up, because the stakes are highest here.

Step 3Draft the narrative

Prompt it to write the section:

Draft the financial section of my board update for [period], matching the template.
Cover: revenue and margin versus prior period and plan, the 3 most important variances
with the why, cash position and runway, and 2-3 risks or watch-items. Pull every number
from the uploaded statements and cite which statement each came from.

What comes back is a structured draft: the highlights, the variance explanations, the cash story, in your format. Illustrative, in the structure a template would set:

Q2 financial summary

Revenue of $4.62M came in 7% ahead of Q1 ($4.32M) and 2% above plan ($4.53M), led by the enterprise segment (Income Statement, p.1). Gross margin held at 61%, essentially flat to Q1's 62%.

The three variances worth the board's attention:

1. Sales and marketing rose to $1.18M from $0.94M (+26%) against a 7% revenue increase. This is the pipeline investment approved last quarter, tracking to plan, but worth naming because it compressed operating margin this quarter.
2. Hosting and infrastructure grew 19% on the platform migration; expected to normalize in Q3.
3. [MISSING: prior-period headcount cost] the comparison figure is not in the uploaded files; add it or I will leave this variance out.

Cash position: $2.1M, roughly 9 months of runway at current burn (Balance Sheet, p.2; burn from the trailing-three-month average, arithmetic shown below).

Two things to notice. Every figure carries its source, so verifying is a lookup, not a recompute. And the model surfaced a gap as a [MISSING:] flag instead of inventing a number to fill it, which is exactly the behavior the Step 2 boundary buys you.

Step 4Fact-check every number against the source

This is the non-negotiable step. Go through the draft and check each figure against the statement it came from. The AI is reliable at structure and language and must be verified on every number. Asking it to cite the source statement per figure (Step 3) makes this fast: you are confirming, not recomputing. Any [MISSING:] flags are your cue to add the data or cut the claim.

Here is what verifying the excerpt above actually looks like. The $4.62M revenue cites "Income Statement, p.1," so you open p.1 and find the cell. Two seconds, match. The 61% margin you eyeball against the same statement. The line to slow down on is the runway figure, because it admits "burn from the trailing-three-month average, arithmetic shown below." That is the AI doing a calculation, which is precisely where it goes wrong. So you check the working, not just the inputs: did it average the right three months, did it use net burn or gross? The reason this matters: on a real draft I once caught a runway number that was several months too generous because the model had averaged a one-time financing inflow in as if it were operating cash. It looked right. It was very wrong. Catching the cited arithmetic, not just the pulled figures, is the whole difference between confidence and an embarrassing correction in the room.

Step 5Own the judgment calls

The draft will explain what happened. The parts that need you are the so what and the what we are doing about it, the strategic read, the decisions, the framing of a miss. Rewrite those in your own words. The AI gets you to a complete, accurate draft; you supply the operator judgment that makes it a leader's update rather than a report.

Make it yours. Boards differ, and the template you feed it is how you encode yours. A venture board often wants a tight narrative with growth, burn, and runway up top and the detail in an appendix. A bank or a family-office board may want the full statements with a conservative, plain-spoken commentary and no spin on a miss. A board with one famously detail-oriented member may want a "questions you might ask" section that pre-empts the obvious follow-ups. Whichever it is, give the model your actual last memo rather than a generic finance template, and it will match the register your specific board reads in.

How you'll know it's working

The board-prep task stops eating your weekend, and the financial section stops being the bottleneck. The quality tell is that your board's questions shift, from "what does this number mean?" (which the narrative now pre-answers) to the strategic questions you actually want to discuss. If board prep went from a Sunday to an hour, it is working.

When it breaks

• A number is wrong. This is why Step 4 is non-negotiable. If you catch a wrong figure, it almost always means the AI calculated something instead of pulling it. Re-state the "only use numbers from the files" rule.
• It made up a figure you did not provide. Same fix, harder enforcement. Add the [MISSING:] instruction so gaps surface as flags instead of fabrications.
• The tone is off. Your template is missing or too thin. Give it a real past memo, not a description of one.
• It is editorializing on a miss. The strategic framing is your job (Step 5). Let it state the variance; you write the interpretation.
• You are tempted to skip the fact-check because it looks right. Do not. "Looks right" is exactly how a wrong number reaches your board. Verify every figure, every time.
• It cites a page or statement that does not exist. It is guessing at structure, which means your upload was ambiguous. Re-upload with clear filenames ("Q2-income-statement.csv") and a one-line note telling it what each file is, so its citations point at something real.
• It buries the one number your board always asks about. The model weights by dollar size, but your board may care most about runway, CAC, or a single segment regardless of size. Name your board's hot buttons in the prompt ("lead with runway and net revenue retention") so the draft opens where the questions will.

Where this fits in your harness

This is the capstone of the finance chain: it draws on clean, closed books, the understanding you built interrogating your P&L, and the forward view from your cash-flow forecast. One finance Project, fed once, now closes your books faster, answers your questions, projects your cash, and drafts your board update, which is what "AI leverage in finance" actually looks like for a CEO: not one trick, a chain.