←BaldanAll skillsLet's talk

Data Analytics & BI · Dashboards

A dashboard is a decision, drawn.

Before the dashboards themselves: how one is built from a question, which kind fits which job, and what separates a good dashboard from a busy one. Everything on the real Online Retail data.

From a question to a dashboard

How a dashboard is built, one decision at a time.

Scroll and watch the real Online Retail sales dashboard put itself together: question, numbers, model, charts, layout, context, interaction, refresh.

SMSales managermonthly review, 10 minutes1. Are we growing?2. Where do sales come from?3. Do customers come back?Sales overviewNovember 2011Country: AllMonth: Nov 2011Revenue (net)SUM(qty × price) − cancelled£1.43m+34%OrdersCOUNT(DISTINCT invoice)2,751+37%Average orderrevenue ÷ orders£519−2%CustomersCOUNT(DISTINCT customer)1,660+22%Revenue by monthDJFMAMJJASONNov: £1.43m (+34% vs Oct)Where it sellsUK: 85% of revenue. Outside the UK:NetherlandsIrelandGermanyFranceAustraliaTop productsRegency cakestand 3 tier£164kParty bunting£98.2kWhite hanging heart t-light holder£97.8kJumbo bag red retrospot£92.2kRabbit night light£66.7kOrders by weekday and hourMonTueWedThuFriSatSun6h → 20h · no orders on Saturdays in this shopCustomers still buyingDec1003732383640Jan1002227233229Feb1001919292725Mar1001525202217Apr1002120212023May1001917172123% still ordering, months 0 → 5dim_date374 daysdim_product3,912 productsdim_country38 countriesdim_customer4,334 (counted)fct_sales522,568 clean lines1234Data up to 9 Dec 2011 · 522,568 clean rows · loads, tests and refreshes on a schedule

Four kinds of dashboards

Not every dashboard does the same job.

Who reads it and how often decide almost everything: how many numbers, how much detail, how often it refreshes. Below are the working examples: three on real public data, one on an IBM sample with fictional customers, and one on synthetic FWA data built so every raw row can be shown.

01 · dashboard type

Strategic

“How are we doing, and where should we look?”

Reader
Management
Refresh
Monthly
Design
Few numbers, big and compared; a trend; little detail.

Our example

Online Retail sales

541,909 real transactions, 38 countries

Open the dashboard →

02 · dashboard type

Experiment

“Did the change work, and should we roll it out?”

Reader
Product and marketing teams
Refresh
Once per test
Design
Groups side by side, lift with uncertainty, a clear verdict.

Our example

Email campaign A/B test

Hillstrom: 64,000 real customers, 3 groups

Open the dashboard →

03 · dashboard type

Operational

“Is everything OK right now?”

Reader
People acting today
Refresh
Daily or live
Design
Status against thresholds, alerts first, the latest values.

Our example

Jakarta air quality

Live PM2.5 from Open-Meteo (CC BY 4.0)

Open the dashboard →

04 · dashboard type

Customer analytics

“Who is leaving, and why?”

Reader
CRM and retention teams
Refresh
Weekly
Design
Segments, cohorts and drivers; more exploration.

Our examples · 2

Telco customer churn

7,043 customers (IBM sample data)

Open the dashboard →

FWA home-internet base health

4,702 customers, 22 months, 7 raw tables incl. surveys (synthetic)

Open the dashboard →

Dashboard makeover

Same data, five fixes, a different dashboard.

The dashboard below shows real numbers but breaks the usual rules. Switch the fixes on one by one and watch what each one changes.

0 / 5 fixes
★ SALES DASHBOARD ★REVENUE (NET)£1.43mORDERS2,751AVERAGE ORDER£519CUSTOMERS1,660REVENUE BY MONTHREVENUE BY COUNTRYUKNetherlandsEIREGermanyFranceAustraliaTOP PRODUCTS (NET REVENUE)Jumbo bag red retrospot£92.2kRegency cakestand 3 tier£164kPaper chain kit 50's christmas£63.7kParty bunting£98.2kRabbit night light£66.7kWhite hanging heart t-light holder£97.8k

Real Online Retail numbers for November 2011, net of cancellations.