Data Storytelling

Headline Formula

Every data finding needs a headline: [Specific Number] + [Business Impact] + [Actionable Context].

"Readmission risk scores now flag 23% more high-risk patients, saving $2.1M annually." Not "We improved our model performance." Not "Results were statistically significant."

The number makes it concrete. The impact makes it relevant. The context makes it actionable.

Narrative Structures

Pick the structure that matches the situation:

| Structure | When to Use | Shape | |-----------|------------|-------| | Problem-Solution | Pitching new work | "Here's the gap, here's what we built" | | Trend | Status updates | "Here's what changed and why it matters" | | Comparison | Trade-off decisions | "Here are two options with costs" |

Narrative Arc for Presentations

Hook (the surprise or gap) → Context (what the audience needs to know) → Evidence (the data, 2-3 charts max) → Implication (so what?) → Recommendation (now what?)

Start with the finding, not the methodology. Executives care about the answer. They'll ask about the method if they want it.

Chart Selection

Match the metric type to the right chart:

| Metric Type | Chart | Example | |-------------|-------|---------| | Counts | Bar chart | Monthly patient encounters | | Rates over time | Line chart | 30-day readmission rate by quarter | | Part-of-whole | Stacked bar | Claim denials by category | | Distribution | Histogram or box plot | Length of stay distribution | | Correlation | Scatter plot | Cost vs complexity score | | Ranking | Horizontal bar | Top 10 diagnoses by volume |

NEVER use pie charts. Stacked bar does everything a pie chart does, with readable labels.

NEVER use dual Y-axes. Two metrics, two charts. Dual axes let you imply any correlation by scaling the axes.

ALWAYS label data directly on the chart. A legend across the room is useless.

Presentation Anti-Patterns

Charts without a "so what." Every chart needs a headline that states the finding. "Figure 3: Revenue by Region" tells the audience nothing. "Northeast revenue dropped 12% after formulary change" tells them what to see.

Data dumps disguised as analysis. 47 metrics on one slide helps no one. Pick the 2-3 numbers that matter and make them big.

Leading with methodology. "We used a logistic regression with 47 features and cross-validated using..." Save it for the appendix. Lead with the result.

Hiding bad news in an appendix. If the data tells an inconvenient story, put it up front. Credibility compounds faster than any dashboard metric.

Cross-References

For stakeholder alignment process (getting agreement on what to build), see stakeholder-alignment. For metric definitions (what exactly does this number mean?), see metrics-definition. This skill covers how to present the results.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: hollandkevint
• Source: hollandkevint/data-product-operator
• License: MIT
• Homepage: https://www.kevintholland.com/

Install and usage instructions live in the source repository linked above.