Turn a dashboard into a short decision note
The dashboard has twelve charts. Your director needs to know what changed and whether they have to decide anything, ideally before the meeting rather than during it.
What you'll make
a note of up to six sentences for a named reader, with every number traceable to the dashboard and any proposed decision clearly labelled as a proposal.
What you'll need
a report you can open, its measure definitions and refresh time, and ChatGPT, Claude, Gemini or Copilot in your work account. You can paste a small table of checked figures instead of connecting the report. About 20 minutes.
Example result.
Please decide whether to hold Supplier K’s November order while we investigate its returns. In September, 112 of its 1,000 shipped units were returned (11.2%), against 4.1% for the other suppliers. The total return rate was 5.52%, up 1.22 percentage points on September 2025. We don’t yet have K’s figure for last year, so we can’t say how much of that rise is K’s. The cause isn’t established, and the hold is a proposal.
1. Choose the reader and the question
Write down what the reader needs to understand or decide. An update with no decision is fine. Don’t manufacture an approval request because a template has a box for one.
Pick the two or three measures that answer that question. For each, note the reporting period, units, the denominator for any rate, filters and the comparison period. Include a deadline only if one has been agreed.
Why this works: a note with a named reader and one question has an obvious first sentence. A note “about the dashboard” doesn’t.
What you should see: one sentence naming the reader and the question, and a short list of measures with their periods and denominators.
2. Gather a small, checked set of figures
Open the relevant visuals and copy the figures with their titles or links. Record when the data was refreshed. Use underlying values where rounding would change the comparison, and check that both periods use the same definition. A screenshot alone can hide definitions, so paste a table where you can.
If your organisation has Copilot in Power BI (it needs a paid Fabric F2 or Premium P1 capacity or higher, and a Pro licence alone isn’t enough), a report summary can help you find the right visuals. Follow its citations, and know where it stops looking. Copilot picks up to 20 visuals and summarises those, so on a busy report it can miss some. It only considers visuals with fewer than 30,000 rows. Tables paginate after 100 rows, and only visible rows are summarised. Page and report filters are supported, visual-level filters aren’t, and summaries don’t apply filters just because your prompt mentions one. None of these produce an error message. They produce a confident summary of part of the data.
Why this works: if the dashboard itself is wrong, a summary of it is wrong in better prose. Checking a few figures first is cheaper than checking a finished paragraph.
What you should see: a short table: measure, value, period, denominator, filter and source link for each figure.
3. Ask for a six-sentence note
Draft a note of at most six sentences for [reader] about [question].
Use only the evidence below. Put the decision request first if there is one, otherwise lead with the main finding.
Include the reporting period and comparison. Keep percentage changes separate from percentage-point changes.
Do not infer causes, deadlines, completed actions, approvals or commitments. Label hypotheses and proposals as such.
If a needed fact is missing, list it under Questions rather than inventing it.
After the note, map each number or factual claim to its source.
Evidence:
[checked figures and agreed context][reader] is the person the note is for. [question] is your sentence from step 1. [checked figures and agreed context] is your table from step 2, plus any decision request someone has actually made.
A good note usually takes this shape, one sentence each:
- The decision you need, if any.
- The headline number, with its period and comparison.
- The one change that matters.
- The likely cause, labelled as belief or evidence.
- What you’ve done so far.
- The deadline, and what happens if it slips.
Drop any sentence you have nothing true to put in. A missing sentence 4 is fine. An invented one isn’t.
Why this works: the claim map makes review quick, and the shape forces the model to choose. Directors aren’t troubled by uncertainty. They’re troubled by discovering it later.
What you should see: a short note, a claim-to-source map and a Questions list if anything was missing.
Check the result
- Every number matches your table, with the same period and denominator.
- Percentage-point and percentage changes aren’t mixed up.
- A pattern that two things share isn’t presented as one causing the other.
- A proposed action is still a proposal, and the reader has the authority to decide it.
The usual failure: the note promotes a comparison into a cause and a proposal into an action. “K’s return rate is high” becomes “K caused the increase”, and “consider a hold” becomes “we have paused the order”. Spot it by reading sentences 4 and 5 against the evidence on their own.
Revise the note. The evidence supports [checked finding], but it does not establish [unsupported cause or action]. Remove that inference. Keep [request] as a request. Preserve the numbers, their periods and the remaining uncertainties, and update the claim-to-source map.[checked finding] is what the figures show. [unsupported cause or action] is the overreach. [request] is the proposal as it was made. If the data itself is inconsistent, stop and ask the report owner. A more elegant paragraph won’t reconcile conflicting denominators.
Worked example: a supplier return rate
Fictional example. Operations wants a decision on Supplier K before the November order goes in.
Returns report, September 2026, same unit-based definition throughout:
Supplier K: 1,000 units shipped; 112 returned; return rate 11.2%.
Other suppliers combined: 4,000 shipped; 164 returned; return rate 4.1%.
Total: 5,000 shipped; 276 returned; return rate 5.52%.
September 2025 total: 5,000 shipped; 215 returned; return rate 4.30%.
September 2025 supplier-level figures: not available.
Operations: requests approval to hold Supplier K's November order while returns are investigated. No hold approved. No cause found yet. No decision deadline given.The first draft said:
Returns rose 28% this year because of Supplier K, whose rate is nearly three times everyone else’s. We have put K’s November order on hold pending investigation.
- The failure. The 28% is a relative change (5.52 ÷ 4.30). Without K’s 2025 figure, nobody can say K caused it. And no hold was approved.
- The repair. The revise prompt, with “K’s rate is higher than other suppliers this September” as the checked finding and “K caused the annual rise” and “order on hold” as the overreach, produced the note in the example result.
- The arithmetic. 112 + 164 = 276 returns, and 276 ÷ 5,000 = 5.52%. The rise is 1.22 percentage points. The Questions list asks who decides and by when, rather than inventing a date.
Try it
Counts and rates tell different stories
Choose the reporting view. Keep its denominator beside each value.
Supplier K: 112 count (112 of 1000); Other suppliers: 164 count (164 of 4000).
Task kit for this job: Download (ZIP) · Reusable instructions
Go deeper
Optional training: update the note when the missing figure arrives
The report owner sends K’s September 2025 figures: 40 returns from 1,000 shipped, a rate of 4.0%. That leaves the other suppliers at 175 of 4,000 last year (4.4%), now 4.1%.
Update the note with this new evidence: [finding and source]. Say exactly which uncertainty it resolves and which remain. Keep unapproved actions as proposals. Do not imply the new evidence was known when the earlier note was written.[finding and source] is the new figure and where it came from. The update can now say the year-on-year rise comes from K: its returns rose from 40 to 112, while the others fell. It still can’t say why. The cause and the hold decision stay open.
Optional training: give the reader a way back to the evidence
Send the report title, links to the visuals you used, the filters and the refresh time with the note. If the reader can’t open the dashboard, attach an excerpt they’re allowed to see. A source link the reader can’t open needs a better handover.
Optional training: use Tableau Pulse as raw material
Tableau Pulse, on Tableau Cloud, sends digests about the metrics you follow, and builds an insight summary once you follow two or more. Open the metric page and check its scope and filters before borrowing a figure. It’s a different surface from a Power BI report summary, but the note and the checks are the same. For a progress update from narrative notes rather than figures, use the weekly-update guide.