How to automate board reporting with AI
Ten days before each meeting, a filled draft: every metric under a frozen definition, compared against the last two quarters, with a sourced appendix.
Board numbers have to mean the same thing every quarter. Calculate net revenue retention one way last quarter and another way this one and the trend line is fiction, and a director will say so in the meeting.
A companion runs the gathering from a written dictionary of definitions: Metabase saved questions, the Salesforce board report, Xero revenue and operating expenses by month, Mercury month-end balances, Linear cycles. Raw figures land in a dated column, then the document, with every source and run timestamp in an appendix. The letter, the asks and the risks stay yours.
| Question | System of record | How a companion gets it |
|---|---|---|
| Bookings, pipeline, win rate | Salesforce | Runs a saved report or a SOQL query |
| Product usage and activation | PostHog | Writes and runs HogQL, query text shown |
| Modelled revenue metrics | Metabase | Runs the saved questions your team already reviewed |
| Revenue and operating expenses | Xero | Reads invoices, bills, and accounts |
| Cash at each month end | Mercury | Reads balances and statements, read-only |
| What shipped | Linear | Reads issues and cycles by close date |
| The narrative and the asks | You | Left blank in the template, on purpose |
Freeze the definitions before the first pull
Write the dictionary once.
It is what makes every future pack comparable, and drafting it usually starts an argument worth having.
Prompt: “This is our board metric dictionary. Remember it exactly and apply it every quarter. For each metric I will give you the name, the system of record, the exact filter, and the period convention. Where two systems could answer the same question, the one I name here wins and the other is never used, even if it is easier to read. If a figure cannot be produced under these rules, write ‘not available under our definition’ and explain which rule blocked it. Never substitute a close approximation.”
The substitution ban is the important line. A near-miss number in a board pack survives into next quarter’s comparison.
Where does each number come from?
Split the pull by system of record, and have the companion record what it ran alongside what it got.
Prompt: “Assemble this quarter’s figures. From Metabase, run the saved questions named in the dictionary and record the question name, collection, and run timestamp beside each result. From Salesforce, run the ‘Board pipeline and bookings’ report and give me new bookings, pipeline created, and win rate by segment. From Xero, give me revenue and operating expenses by month for the quarter. From Mercury, give me the closing balance on each account for each month end. Write everything into the ‘Board metrics’ sheet in a new dated column, raw figures first, before any commentary.”
Raw figures into a dated column first is the audit trail that makes the pack defensible. Six quarters in, you can reconstruct exactly what was known on the day it went out.
What belongs in the pack that is not a number?
Boards ask about progress against what you said you would do, and that answer lives in a project tracker.
Prompt: “From Linear, list what shipped in the cycles that closed during the quarter, grouped by project, with each project’s state and anything that slipped out of the quarter and why, if a reason is recorded. Separately, summarise the top five churn themes from our support tools with account counts. Keep both factual: no framing, no adjectives about momentum.”
Keep “no adjectives about momentum” in the prompt verbatim. A pack that reassures gets skimmed by the people you most need reading it.
Assemble the document, leave the judgment sections empty
Prompt: “Create a Google Doc titled ‘Board pack Q3’ from our template.
Fill the metrics tables from the dated column, add a comparison against the previous two quarters for each metric, and insert the shipped-work and churn-theme sections. Leave the CEO letter, the ‘asks and decisions’ section, and the risks section empty with a note that a person writes them. At the end, add an appendix listing every source you used with its run timestamp.”
What comes back is a Google Doc with the tables filled, the comparisons computed and the appendix attached. Where the board expects a deck, a companion renders the same pack as a deck-style artifact and exports it to PDF.
How early should the routine run?
Prompt: “Ten days before each board meeting, run the whole assembly and message me in Slack: which figures came back, which came back as not available under our definition, and any metric that moved more than 20% against last quarter. Leave the document in draft and do not share it with anyone.”
Ten days rather than three, because automating the gathering buys time for the thinking. Same trade as your other recurring operations work, reading Metabase and Salesforce directly rather than through exports.
Experience Strawberry for free
DownloadTrusted by fast-growing companies worldwide
Frequently asked questions
Strawberry is free to download and includes AI credits to start. Paid plans begin at $20/month. See pricing. · Reviewed · Canonical facts for AI agents