Y

Data visualisation and storytelling · Intermediate

Design a dashboard people open every morning

3 min read

Why it matters

Most analyst roles own at least one recurring dashboard. Job descriptions ask for "building and maintaining dashboards" and "self-serve reporting". A dashboard that nobody opens is wasted work, and portfolio reviewers can tell within seconds whether a dashboard was designed for a user or just filled with charts.

The concept

Design in this order:

  1. Audience and decision. Who opens it, how often, and what do they decide? "A city operations manager, every morning, deciding where to add drivers."

  2. KPIs. Three to five numbers, each with a comparison (vs yesterday, vs target, vs same day last week). A number without a comparison is not a KPI.

  3. Layout. People read top-left first. KPIs across the top, the main trend below, breakdowns underneath, details last.

  4. Interaction. Filters that match how the user thinks (city, date range), not every column in the data.

  5. Performance. Under three seconds to load, or people stop opening it. Pre-aggregate, limit visuals per page, avoid high-cardinality slicers.

  6. Trust. Show "data as of" time and a link to definitions.

Worked example: a taxi operations dashboard spec

Before opening any BI tool, write the spec. For NYC yellow taxis in January 2025:

dashboard_spec.yaml
audience: Fleet operations lead, checks at 08:00 daily
decision: Where to position cabs today; which zones are under-served
kpis:
  - name: Trips yesterday
    comparison: same weekday last week
  - name: Median fare
    comparison: 28-day median
  - name: Airport share of trips
    comparison: 28-day average
  - name: Trips with zero or negative fare
    comparison: threshold 5%, shown red above it
main_trend: Daily trips with 7-day average, last 60 days
breakdowns:
  - Trips by pickup borough (sorted bar)
  - Top 15 pickup zones with change vs last week
  - Heatmap of trips by hour of day and weekday
filters: [Borough, Date range]
footer: "Source: NYC TLC trip records. Data as of {last_refresh}. Definitions: link"
performance_budget: first view under 3 seconds on a laptop

Then build it in Power BI, Tableau or Looker Studio. The numbers you should see for January 2025: 3,475,226 trips, median fare 12.11, Midtown Center the busiest pickup zone, and a clear weekday morning and evening peak in the heatmap.

Review it with one real user before you share it widely. Ask them to answer their morning question out loud while you watch. Where they hesitate is what you fix.

Common mistakes

  • Designing for yourself. Analysts love detail; managers want the answer. Put detail on a drill-through page.

  • Too many visuals. More than about eight per page slows loading and splits attention.

  • KPIs without context. "91,250 trips" means nothing without "up from 84,832 the day before".

  • No definitions. Two teams will define "active customer" differently and argue about your numbers.

Interview angle

"Tell me about a dashboard you built" is a standard behavioural question. Structure it: who used it, what decision it supported, what you left out and why, and one change after user feedback. Numbers on adoption (weekly users) make the answer much stronger.

Practice task

  1. Write the spec for a dashboard on the Online Retail data for a head of sales.

  2. Build it in the BI tool of your choice with no more than six visuals.

  3. Ask a friend to answer two questions using it, and note where they got stuck.

Further reading