Data Storytelling: Turning a Chart into a Decision

McTaba runs coding and data bootcamps in Kenya. This guide is written to be useful whether or not you ever join one.

Picture a Monday meeting. The shop owner has two minutes before a supplier arrives. You slide over a laptop with a dashboard of nine charts. She looks at it, says "nice", and goes back to her gut feeling.

That analysis changed nothing. Data storytelling is the skill that stops this happening. In our data science career guide for Kenya it is listed as a core skill, not a decoration, because employers hire analysts to change decisions.

Use this test on every piece of work. Can the reader, in one minute, say what you want them to do and why? If not, the story is not finished.

Write the memo before you build the chart. It forces you to know your point. Use five short parts:

1. The decision. One sentence. "Move the sugar order from weekly to twice weekly."
2. The evidence. Two or three sentences, with one chart.
3. The cost of being wrong. What do we lose if the finding is false?
4. What we do not know. Honest gaps, in plain words.
5. The next step. Who does what, by which date.

Notice that the recommendation comes first. A reader who stops after two lines still has the answer. Many analysts do it the other way round, building up to the point like a detective story. Busy readers never reach the last page.

Keep the memo to one page of plain sentences. Skip jargon. If you must say "median", say it once and add "the middle day".

Choose the chart by the comparison you want the eye to make, not by what looks interesting.

The comparison	Use	Avoid
Over time	Line chart	Pie chart
Between categories	Horizontal bars, sorted	3D bars
Part of a whole, two or three parts	One stacked bar or plain numbers	Ten-slice pie
Two measures related	Scatter plot	Dual-axis lines
Exact lookups	A small table	A chart

Choosing and presenting charts is practised weekly in the Data Science to AI Engineering programme, which has a weekly project and a team capstone.

Four habits make any chart easier to read:

• Title the chart with the finding: "Saturday sales are twice a weekday", not "Sales by day".
• Sort bars from biggest to smallest.
• Start bar charts at zero. A cut axis makes small gaps look huge.
• Use one accent colour for the thing you want noticed, and grey for the rest.

The charts themselves come from your code. Our guide to the first 30 days of Python ends with one.

For example, imagine a small shop that sells groceries from 7 AM to 9 PM. The owner asks, "Should I hire a second person for the evening?" Made-up daily sales, in shillings, are in the lists below.

Before

The analyst sends a dashboard. It shows a line of 90 daily totals, a pie chart of product categories, and a table of every till entry. The title says "Sales analysis, Q3". The owner sees noise.

After

The analyst groups sales by hour and finds the pattern first. For example, imagine these made-up averages per hour:

• For example, 7 AM to 12 noon: about KES 1,200 an hour.
• For example, 12 noon to 5 PM: about KES 900 an hour.
• For example, 5 PM to 9 PM: about KES 2,600 an hour, with one cashier and a queue.

Now the one-page memo:

Decision: Add a second cashier from 5 PM to 9 PM.

Evidence: Evening hours take the most money, and the queue is longest then. One bar chart, sorted, with the evening bar in colour.

If wrong: The cost is one extra wage for four hours a day. Try it for one month.

What we do not know: Whether customers leave without buying when the queue is long. The till does not record it.

Next step: The owner picks a cashier by Friday. We compare evening takings after four weeks.

The chart title reads "Evenings take the most money, and the queue is longest then". Three bars, one colour, no gridlines. The owner decides in one minute.

The analysis was the same in both versions. Only the packaging changed. That is the whole skill.

1. Take any analysis you have made. Write the decision it supports in one sentence.
2. Delete every chart that does not support that sentence.
3. Rewrite each remaining chart title as a finding.
4. Read the memo aloud to a friend. If they cannot repeat the decision, cut more.

Stories also depend on numbers you can defend. Read the statistics you actually need for data science to know when a gap is real, and SQL for data analysis to pull the numbers cleanly.

The Data Science to AI Engineering programme ends in a team capstone shown at a public demo day, so presenting is practised, not skipped. It starts on 11 January 2027 and costs KES 140,000, or KES 120,000 if paid in full upfront.