Building your first dashboard for a portfolio

What a good dashboard is

Not a collection of attractive charts. A tool from which someone understands in a minute what's happening and what to do about it.

The test is simple: show it to someone unfamiliar with the data and ask what conclusion they drew. No answer, or the wrong one, means the dashboard doesn't work however handsome it looks.

Choosing a subject

Take a question, not a dataset. Bad: "built a sales dashboard". Good: "investigated why revenue is growing while profit isn't".

The question dictates the structure: what to show, in what order, what to compare with what.

Choose a field close to you. Sport, transport, property, your city. Domain understanding shows immediately: you ask the right questions and notice oddities in the data.

Which metrics to use

The rule: no more than five to seven headline numbers. Nobody reads a thirty-chart dashboard, including its author.

A structure that works:

Top — the headline figures. Three or four: revenue, volume, share, change against the previous period. Large enough to read in a second.

Below — the trend. How it changed over time. That answers "better or worse".

Below that — the breakdowns. By category, region, channel. That answers "driven by what".

Bottom — the detail. A table for people who need to dig.

Presenting so the conclusion is obvious

Label the meaning, not the metric. Not "Conversion, %" but "Seven of every hundred visitors buy".

Compare. A number without comparison is meaningless. Against last month, against plan, against the market.

Use colour for deviations, not for everything. If everything is coloured, the eye catches on nothing.

Write the conclusion in words. One line above the dashboard: "Revenue up 12%, profit down because acquisition costs rose". That's what the whole thing was for.

Common mistakes

Pie charts with ten slices. The eye can't compare them. Bars read better.

A truncated axis. Starting a bar chart above zero exaggerates differences. It's considered manipulation.

Three axes on one chart. Nobody will untangle it.

Data without a period. "Sales 1200" — over what? A day? A year?

Beautiful but about nothing. The commonest failing of practice dashboards: many charts, zero conclusions.

What to attach

For a portfolio the dashboard alone isn't enough; the account matters:

• what the question was;
• where the data came from and how far it can be trusted;
• what preparation you did: cleaning, joining sources, records discarded;
• what conclusion emerged and what you'd recommend doing.

That last point separates an analyst from someone who draws charts. It's exactly what hiring managers look at.

Where to get data

Open sources: national statistics, city open-data portals, datasets from analysis competitions, public service interfaces.

One tip: avoid the most worn-out teaching datasets. They've been seen hundreds of times, and a dashboard built on them looks like a course assignment rather than your work.