Sign in

Blog · Board and management reporting

A board narrative with a citation on every number: how to draft the commercial pack from reconciled figures

A method for writing the quarterly commercial narrative so every figure carries a citation to the calculation that produced it: the reconciled roll-up as the source, the narrative structure, the citation format that survives export, and why the model writes sentences but never numbers.

The short answerDraft the narrative from a reconciled roll-up, and give every figure in it a citation to the function and inputs that produced it: net revenue from the revenue calculator for the quarter, growth from the same calculator against the prior quarter, contribution from the region split. The citations survive export into the slide notes and an appendix. The model may write the sentences; the numbers come from the tools, and the two never disagree because the chart on the next slide called the same functions.

The first question at any board is where a number came from. In most companies the honest answer is a chain of spreadsheets that ends at an analyst who has since gone home. This guide sets out how to draft the commercial narrative so every figure carries its own answer, and why that is only possible when the numbers were never typed.

The principle

Every figure in the narrative comes from a function run on the reconciled roll-up, and the narrative records which function. The prose can be edited freely. The figures cannot be edited at all; they change only when the data changes, and the citations change with them.

The source

Before any sentence is written, the roll-up for the quarter is asserted against the number finance confirmed:

reported revenue = Σ regions = Σ teams = Σ people = Σ accounts

If the identity fails, the narrative waits. A board pack drafted on a roll-up that does not reconcile is the pack that gets contradicted in the room.

The structure

A commercial narrative has the same shape every quarter, which is what makes it draftable.

  1. The headline figure and its movement. Net revenue for the quarter, against the prior quarter and the same quarter last year, on the same definition.
  2. What drove it. Contribution by region, by product, by new against existing customers.
  3. Coverage and pipeline. How much of the territory the team reached, what that means for the forecast.
  4. What needs attention. The accounts at risk, valued, and what is being done.
  5. The plan. What changes next quarter, in figures.

Each paragraph has two to four figures, each cited.

The citation format

In the narrative: a superscript after the figure. In the appendix and the slide notes: one line per citation.

# Figure Source Reconciled
1 Net revenue $14.8m revenue_calculator(period=Q3 2026) to finance, 3 Oct
2 +9.2% on Q2 revenue_calculator(period=Q3 2026, vs=Q2 2026) same
3 Northeast 38% of growth contribution(region=NE, period=Q3 2026) same
4 Midwest coverage 71% coverage(region=MW, window=90d) to CRM export, 30 Sep

A reader who wants to check figure 3 knows exactly which calculation to rerun and on which data.

A worked paragraph

Net revenue reached $14.8m¹, up 9.2%² on Q2, with the Northeast contributing 38%³ of growth. Coverage in the Midwest fell to 71%⁴ after two territory vacancies, which accounts for the full shortfall against plan.

Four figures, four citations, one paragraph. The chart on the next slide, revenue by region, was drawn by the same contribution function, so the 38 percent in the prose and the bar on the chart cannot disagree.

What the model does and does not do

A language model is good at the sentences and bad at the numbers. So it is given the tool outputs and asked to write the narrative around them, with the citations attached to each figure. It cannot introduce a figure that no tool produced, because the draft is checked: every number in the prose must match a cited tool output, or the draft is rejected. That check is mechanical, and it is the reason the pack can be trusted.

Where it goes wrong

Figures typed into the prose. One hand-typed figure breaks the guarantee for the whole pack. The check catches it; the fix is to cite it or remove it.

Definitions that differ from finance. "Revenue" in the narrative and "revenue" in the finance pack must be the same definition, or the reconciliation line has to say what the difference is.

Citations lost in export. PowerPoint drops superscripts on copy. Keep the citation table in the notes and the appendix, and the superscript is a convenience.

Prior-period figures recomputed. Last quarter's number in this quarter's pack must be the number that was reported, not a fresh computation on data that has since changed. Freeze reported figures with their reconciliation date.

Drafted by Thursday

Mapped once, the quarter's roll-up is asserted the day finance confirms it, the tools run, and the draft with its citation table is ready a week before the board. Covirage does this on the reconciled roll-up. The board reporting page describes it.

Questions people ask

What does a citation look like in a board pack?

A superscript number on the figure in the narrative, and a line in the appendix: 1, net revenue, Q3 2026, revenue_calculator(period=quarter), reconciled to finance 3 October. In the slide notes, the same line. A board member who asks where the 9.2 percent came from gets the answer from the page.

Can the model change a number?

No. The numbers are computed by tools and locked before the narrative is drafted. The model writes the prose around them, and an editor changes the prose. The numbers change only when the data does, and then the citations change with them.

How early can the draft exist?

The day finance confirms the quarter. The roll-up is asserted against the confirmed number, the tools run, and the draft is written from their output. Thursday before the board, a week ahead, is comfortable.