Notifications
Last 30 Days
No notifications
Last 30 Days
No notifications
The most valuable skill in data analytics isn't writing SQL or building dashboards — it's communicating insights that drive decisions. Data storytelling combines data, visuals, and narrative to make complex findings clear and actionable.
| Phase | Focus | Deliverable | ||||
| 1. Context | What's the business question? Who's the audience? | Problem statement | ||||
| 2. Insight | What did the data reveal? What's surprising? | Key findings (2-3 max) | ||||
| 3. Action | What should we do? Why now? | Recommendations with ROI | Choosing the Right Visualization | Message | Best Chart | Avoid |
| Trend over time | Line chart | Pie chart for time data | ||||
| Part-to-whole | Stacked bar, treemap | 3D pie charts | ||||
| Comparison | Bar chart (horizontal) | Radar charts for >5 items | ||||
| Distribution | Histogram, box plot | Tables of raw numbers | ||||
| Relationship | Scatter plot | Dual-axis with different scales | ||||
| Geospatial | Choropleth map | Too many colors on map |
# Bad: "Here's a chart of our sales"
# Good: "Sales dropped 23% in Q3 due to supply chain delays"ax.annotate("Supply chain\ncrisis began",
xy=(crisis_date, crisis_value),
xytext=(offset_x, offset_y),
arrowprops=dict(arrowstyle="->", color="red"),
fontsize=12, color="red", fontweight="bold")
| Pitfall | Problem | Fix |
| Truncated Y-axis | Exaggerates small differences | Start axis at 0 for bar charts |
| Cherry-picking dates | Shows only favorable trends | Show full timeframe with context |
| Dual Y-axes | Implies false correlations | Use two separate charts |
| Too many colors | Cognitive overload | Limit to 5-7 distinct colors |
| 3D effects | Distorts perception | Use flat 2D charts always |
Storytelling is the part nobody teaches in stats class but decides whether your work matters. A correct analysis nobody acts on changed nothing. The job of the analyst's last mile is to take charts + numbers and turn them into a *decision* a busy executive can make in 60 seconds.
Three ingredients:
1. Data — the numbers you've cleaned and verified. 2. Visuals — the chart(s) that show the pattern at a glance. 3. Narrative — the words that say *what it means and what to do about it*.
Missing any one of these turns a story back into a report. A great analyst makes the narrative the headline.
| Audience | What they want | What to skip |
| Executives | the answer + 1-2 reasons + ask | methodology, p-values |
| Product / Ops | the metric breakdown + actionable segments | model details |
| Other analysts | methods, code, caveats | high-level framing |
Same analysis, three different decks. Always know who's in the room.
1. Burying the lede. Recommendation on slide 30. Executives stop reading at slide 3 — put the answer first. 2. Methodology dumps. "We used a Chi-square test with α=0.05…" No business decision needs this in the deck. 3. One slide, ten charts. Pick the *one* chart that tells the story. Move the rest to the appendix. 4. Jargon. "Churn rate increased 2σ" → "We're losing customers faster than usual". 5. Charts without takeaway titles. "Sales by Region" is a label, not a story. "West region drove 60% of Q1 growth" is a story. 6. No call to action. Every analysis must answer: *what should we do, and who decides?*
The single most useful framework for executive communication:
Recommendation
/ | \
Reason 1 Reason 2 Reason 3
/ \ / \ / \
Data Data Data Data Data DataMECE = Mutually Exclusive, Collectively Exhaustive. Every segment cuts the world cleanly:
When revenue drops, you don't guess. You decompose:
Revenue = Traffic × Conversion × AOVRevenue ↓ 15%? Check each lever:
Replace descriptive titles with takeaway titles:
| Before | After |
| "Q1 Sales by Region" | "West region drove 60% of Q1 growth" |
| "Conversion rate over time" | "Conversion dropped 4pp after the April redesign" |
| "Cost vs Revenue" | "We turned profitable in February" |
If an executive only reads the title, they should still get the message.
Pre-attentive cues — the brain processes them in milliseconds:
Executive-ready deck pattern:
1. Title slide: the recommendation. 2. Executive summary (1 slide): answer + 3 reasons + ask. 3. Each reason gets one slide: takeaway-title chart + 2-3 bullets. 4. Risks / counter-arguments: 1 slide. Shows you stress-tested it. 5. Next steps + owners: who does what by when. 6. Appendix: methodology, raw tables, sensitivity analysis.
Many teams now prefer a one-page memo (Amazon-style) over slides for high-stakes decisions — same structure, denser prose.
When someone challenges a number:
1. Acknowledge — "good question, I had to dig into that too". 2. Locate it — go to the appendix slide; show the data. 3. Bound it — "even if X were 20% off, the recommendation doesn't change because…". 4. If you don't know: "I'll get back to you by EOD" — never bluff.
1. Take a recent chart you made. Rewrite the title as a *sentence* that states the takeaway. 2. Pick a metric (revenue, signups, NPS). Decompose it into 2-3 multiplicative drivers and write a one-paragraph diagnosis if it dropped 10%. 3. Restructure a 15-slide deck into Pyramid form: 1 recommendation slide + 3 reason slides + appendix. 4. Write a 1-page Amazon-style memo for the same analysis. Notice which slides become unnecessary.