Write a Data Narrative from a Dashboard or Metrics Sheet

Tier: Core · Hands-on time: ~35 minutes · Tools: ChatGPT, Google Sheets, Claude

A dashboard answers "what are the numbers?"; a narrative answers "so what?". The gap between them is where analysts earn their keep — and where most status meetings die, with twelve people staring at a chart nobody interprets out loud. This recipe turns the interpretation into a repeatable drill: structured data in, candidate story out, human verification before anything ships.

When to use — and when not to

• Monthly or quarterly business reviewsyou own a metrics sheet and owe a written commentary alongside it.
• The "can you add a summary?" requestsomeone forwards a dashboard screenshot or export and wants three paragraphs of meaning by end of day.
• Board or leadership pre-readsthe same numbers everyone has seen, re-told as movement, cause, and consequence.

And when not to:

• The data itself is unverified. A narrative built on a broken pipeline or a stale export launders bad numbers into confident prose. Confirm the sheet is right first.
• One metric moved and everyone knows why. "Revenue dipped because our largest client churned in March" needs a sentence, not a workflow.
• The narrative requires context the sheet doesn't hold. If the real story is a reorg, a competitor launch, or a pricing change, feed that context in explicitly (step 2) — or the model will invent a plausible-sounding wrong cause.

The recipe

1. 1.

   Export the numbers as text, not a screenshot (5 minutes). From Google Sheets, copy the relevant range — metric names, current period, prior period, plan/target if you have one — and paste it as a table. Text beats screenshots: the model misreads chart images far more often than it misreads a table. Trim to the metrics that matter; forty rows of vanity metrics dilute the story.

2. 2.

   Generate the candidate narrative with movement rules. Paste with this prompt:

   Copy

   You are a business analyst writing commentary for leadership. Below
   is a metrics table for [PERIOD] with prior period and target, plus
   context notes. Write a data narrative with EXACTLY this structure:
   
   1. Headline (one sentence): the single most important movement
      this period and its business consequence.
   2. What moved (3-5 bullets): each names the metric, the direction,
      the magnitude (absolute AND percentage), and versus what baseline.
   3. Why (2-3 bullets): causes — but ONLY causes supported by the
      data or my context notes. If the data cannot explain a movement,
      write "Driver unclear from data — needs investigation."
   4. So what / what to do (2-3 bullets): implications and proposed
      actions, each tied to a movement above.
   
   Rules:
   - Every number must come from the table. Recompute percentage
     changes yourself from the absolutes; do not trust my mental math.
   - Flag any metric that moved >15% either direction even if I seem
     to be ignoring it.
   - No hedging filler ("it is worth noting"), no praise of the data.
   
   CONTEXT NOTES: [ANYTHING THE SHEET CAN'T KNOW — launches, pricing
   changes, one-offs, seasonality]
   
   METRICS TABLE:
   [PASTE TABLE]
   

   A good output reads like this:

   Copy

   Headline: New-customer revenue grew 18% ($214k → $252k) against a
   flat market, but churn in the SMB tier doubled and now offsets
   nearly half of that gain.
   
   What moved
   - SMB churn: 2.1% → 4.3% monthly (+2.2pts vs prior, target 2.5%).
   - New-customer revenue: +$38k (+18%) vs prior period, +9% vs plan.
   ...
   Why
   - SMB churn spike coincides with the March price change (context
     note); enterprise churn is flat, consistent with price cause.
   - Driver unclear from data — support ticket volume rose 22% with
     no corresponding usage growth; needs investigation.
   

3. 3.

   Verify every number against the source range (8 minutes). This is the non-negotiable step. Check each figure and each computed percentage against the sheet — in Sheets, a quick =(B2-C2)/C2 column beside the model's claims settles arguments fast. Models transpose digits and occasionally "improve" a percentage. One wrong number in a leadership memo costs you a quarter of credibility.

4. 4.

   Interrogate the "why" section. For each stated cause, ask yourself: does the sheet or a context note actually support this? Push back in the chat: "You attribute the churn spike to pricing — what in the data supports that over a seasonality explanation?" The model's answer either strengthens the memo or exposes a guess you should cut or soften to "hypothesis".

5. 5.

   Re-pitch to the memo's altitude and ship it. Same conversation: "Rewrite at 150 words for the exec summary slot — headline, two movements, one action. Keep every number identical." Paste into your doc, rewrite the headline in your own voice, and attach the sheet as the source of record.

Make it yours

• ChatGPT with the sheet uploaded: attach the export file and ask it to recompute all deltas with its data-analysis tool before writing — you get a verification table for free, which shortens step 3 considerably.
• Claude for long context: paste twelve months of the table and ask for the narrative plus "which of this period's movements are trend versus noise, given the full year" — trend-versus-blip is the question executives always ask next.
• Google Sheets (Gemini) in place: for a fast first pass, ask Gemini in the side panel to summarise the selected range with the step-2 structure. Weaker at strict formats — paste the full rules, and still do step 3 yourself.

Quality check

• Every figure matches the source rangeabsolutes and recomputed percentages both.
• Each movement names its baseline"up 12% vs prior quarter" not "up 12%".
• Causes divide cleanly into supported and flaggedat least one "needs investigation" is a sign of honesty, not weakness.
• Actions trace to movementsno pet initiative smuggled into "what to do".
• The headline survives alonea reader who sees nothing else still gets the period's true story.