Storytelling with Data
Numbers find their voice
Analyses succeed as narratives: context, progression, a decisive turn, and a recommendation. The same chart can be placed to persuade or confuse — narrative placement is editorial power.
What you will learn
- Structure analysis as context → insight → recommendation
- Place charts to carry the arc, not decorate it
- Tune chart emphasis to audience decisions
What is this?
Storytelling with data is sequencing: open with context the audience already believes, present the surprise the analysis discovered (with the chart that proves it), and close with a recommendation tied to the decision. Charts are placed as evidence at the moment they're needed, emphasized by color/labels focusing one takeaway each.
Why do you need it?
- The same numbers persuade when edited and erode trust when dumped.
- Recommendations land only after the context and evidence steps.
- Viewers remember stories, evidence, and names — not forty data rows.
- Senior analysts are judged on how decisions move, which is narrative craft.
Mental model
A trial lawyer's closing: scene, proof, ask
Analytics storytelling
The lawyer sets the scene (context), exhibits the evidence with one clean exhibit each (charts), then states the verdict and asks for it (recommendation). No exhibit is unadorned; no point is made twice. Data stories run under the same discipline.
How it works
- Structure: context (1-2 slides) → analysis (evidence slides, one claim each) → recommendation (decision + next step).
- One takeaway per slide, written as a headline, not a title.
- Emphasis: the single chart element the audience must see is color/label focused.
- Place charts exactly at the claim they support; never rez-in ahead.
- Conclude with the ask: 'approve X' or 'monitor Y' — else the analysis dangles.
The report arc.
Examples
Headline-first arc
deck = [ ("Churn rose to 3.2% in Q2", "context that shakes the board"), ("Drivers cluster in legacy accounts", "evidence chart, one highlight"), ("Cutting onboarding gaps reverses 1.1pts (est.)", "the recommendation"),]for i, (headline, purpose) in enumerate(deck, 1): print(i, headline, "|", purpose)Each slide carries one headline-claim and a purpose tag, so the story reads like a decision path, not a file dump.
In production
- An exec review leads with 'NRR up', then the driver chart, then 'approve rebooking offers'.
- A pricing story places the elasticity chart only where the recommendation needs it.
- A churn post-mortem opens with the trust baseline before revealing the failure curve.
Best practices
- Write the headline before the chart — the claim dictates the visual.
- One claim per slide; supports support, never wander off-thesis.
- Directly label what matters; let the rest be quiet background.
- End with an explicit recommendation and an owner.
Common mistakes
- Starting with raw methodology and losing the audience before evidence.
- Ten nearly identical charts no one can rank by importance.
- Recommendation without takeaway — a deck that ends, not concludes.
- Stating correlations as certainties the narrative can't carry.
Debugging tips
- If a reviewer asks 'so what?', the deck skipped the recommendation step.
- If two charts carry the same claim, one is redundant — cut the weaker.
- When a story feels flat, check placement: is the surprise positioned where momentum peaks?
- Run the 'headline test': read only the slide headlines; it must still tell the story.
Interview checkpoint
Quick quiz
The strongest deck structure is:
A slide should carry more than one headline claim.