Stop 13 · Business Communication & StorytellingTopics · 3/7

Data Storytelling With Visualizations

In short

A chart dump is not a story. People remember one clear point, a surprise and what to do next. One well-built chart can win a meeting that twenty dashboards lost.

Applying data visualization principles specifically to model results — showing predictions versus actuals, feature importance, and error patterns in a way a non-technical audience can follow.

The mascot runs a string puppet show for a group of excited children.

Think of it like this

The kathputli show

A puppeteer doesn't wave every puppet at once. One hero, one problem, one big moment, then the ending. Your charts work the same way: one message per chart, the key bar in colour, everything else in grey.

Watch it happen

Building one chart that tells a story

Message

Chart type

Declutter

Highlight

Point title

Next step

1/6Write the point in one sentence first: 'Late deliveries more than double after 9 pm.'

The big ideas

What you need to know

One chart, one point

If a chart needs two messages, make two charts.

Titles state the finding

Readers skim titles. Let the title say what they should conclude.

Grey plus one colour

Colour is a pointer. Use it on the one thing you want eyes to land on.

Show model results visually

Predicted vs actual, errors by segment and top features are the charts stakeholders actually ask for.

See it in code

Walk through the code

python

1import matplotlib.pyplot as plt2 3hours = list(range(18, 24))4late_pct = [6, 7, 8, 15, 17, 18]5 6colors = ["lightgrey"] * 3 + ["crimson"] * 37plt.bar(hours, late_pct, color=colors)8plt.title("Late deliveries more than double after 9 pm", loc="left")9plt.xlabel("Order hour")10plt.ylabel("% of orders late")11plt.gca().spines[["top", "right"]].set_visible(False)12plt.show()

1/4Small sample data: share of late orders for each evening hour.

Same bar chart, but the colour and title do the talking. Nobody needs a legend to get the point.

Save yourself the pain

Common mistakes

Oops

Titling a chart 'Late deliveries by hour'.

Do this instead

Title it with the finding: 'Late deliveries more than double after 9 pm'.

Oops

Using rainbow colours, 3D pies and every column in the dataset.

Do this instead

Use grey for context and one highlight colour. Show only what supports the message.

Your mission

Give one chart a story

~30 min

Quick check

Your chart shows model error is highest for customers in their first month. What's the best title?

Go deeper

1Build a clear visual comparing a model's predictions to actual outcomes.
2Visualize feature importance in a way that reads clearly to a non-technical audience.
3Practice writing a one-sentence takeaway for each visualization you build.