The hardest questions at a board meeting aren't random. They come from the numbers, and a financially literate board member finds them the same way every time: by looking for the month that dominates the quarter, the journal entry posted to revenue, the backlog that seems too large, the receivables that are aging faster than the sales.

If those things are in your ledger, someone at the table will find them. The only question is whether you find them first. That's the job of a prompt in the Sonar AI Prompt Library called Board Meeting Prep: The Ten Hardest Questions.

If you're new to this, Sonar AI is an AI agent that runs inside NetSuite. Every prompt in the library is a playbook that I engineered and tested against live NetSuite data, and you run it inside your own account, against your own records.

Deliberately Adversarial

Most reporting prompts are written to describe. This one is written to hunt. It instructs the agent to go looking for exactly the things a skeptical board member would: quarter-end revenue loading, manual journals posted to revenue accounts, intercompany noise inflating the backlog, working-capital asymmetries, accounting periods that are still open. Then it pre-arms an answer to each, rather than letting the meeting discover them.

The method runs in phases. First it orients: it identifies the most recently completed fiscal quarter as the reporting quarter, picks the prior quarter and the same quarter last year as comparisons, and checks whether the reporting quarter's periods are still open. If they are, every figure in the report is flagged as preliminary.

Then it gathers evidence, and only from posting transactions: the P&L by account type for all three quarters, an eighteen-month revenue trend, the top customers by revenue, active-customer and invoice counts, revenue by subsidiary, receivables and payables aging with the top balances, a balance-sheet snapshot, monthly cash movement, and the open sales-order backlog aged by order year.

Drill Down Before Writing a Word

The third phase is where the hard questions come from. For anything surprising in the evidence, the prompt drills to the transaction level before it writes anything.

  • If one month dominates the quarter, it lists the top revenue transactions in that month and classifies them: normal invoices, journals, or one large customer.
  • If any revenue was posted by manual journal, it pulls the journal's full line detail, memo, date, and creator. Manual revenue journals are always a top-three board question.
  • If the backlog looks implausibly large, it inspects the biggest orders. Intercompany customer records routinely inflate backlog on OneWorld accounts, so it quantifies the backlog with them excluded.
  • Every customer and vendor ID is resolved to a real company name before it appears in the report. Never a bare ID.

The math is done in a scratchpad, not in the model's head: margins per quarter, year-over-year and quarter-over-quarter growth, opex growth against revenue growth, organic growth with the anomalies excluded, DSO and DPO, customer concentration, the share of receivables that's current, and the fiscal year-to-date change in cash.

Ten Questions, Each With an Answer

From all of that, the prompt chooses the ten questions a skeptical, financially literate board member would actually ask of these numbers. Not generic ones. Each question is phrased bluntly, the way a board member would say it, and carries a severity tag. Some of them are tagged "low, good story," because a question you can answer well is an opportunity.

For each question, the report explains why it will come up, with the specific figures that provoke it, and then gives a draft answer. The rules for the answer are the part I'd underline: lead with the honest headline, quantify with anomalies excluded where that's relevant, concede weaknesses and pair each one with a concrete remediation commitment, and never spin. Every answer ends with a source line citing the exact tables, filters, and account numbers behind it.

The deliverable is a standalone HTML briefing: a KPI card row, the ten severity-tagged question cards with their evidence tables and draft answers, and a prioritized prep checklist that separates what to do before the meeting from what to present proactively. There are no external assets and no chart libraries. Everything is inline, so the file can be emailed and opened anywhere.

Why I Built It This Way

I've sat in enough of these meetings to know that the preparation usually goes into the slides, and the slides show the story that management wants to tell. The board's questions come from a different place. They come from whatever looks odd, and what looks odd is only visible if you go looking for it with the same suspicion the board brings.

Ten is a framing device. The prompt can be told to produce more or fewer, though fewer than eight tends to skip the working-capital questions and more than twelve dilutes the severity. It can also be pointed at a different quarter, or scoped to one subsidiary.

It's in the paid tier of the Sonar AI Prompt Library, under "Board Meeting Prep: The Ten Hardest Questions."