Almost every analytics project produces a dashboard sooner or later. And almost every dashboard stops being opened after a few weeks. Not because the data is wrong, but because nobody knows what to do with it. A useful dashboard is not a collection of charts: it is a tool built to answer specific questions for specific people, at a specific frequency.
Start from the audience
The first question is not «what data do we have?» but «who will look at it, and when?». The answers change everything:
- whoever sets the budget looks once a month and wants to know whether the direction is right;
- whoever runs the campaigns looks every day and wants to know which ad is wasting money;
- whoever develops the site wants to notice immediately if a release has broken something.
Putting all these needs on the same screen produces a page that serves nobody. Three small dashboards, each with a stated audience, are better than one huge dashboard for everyone.
A few questions, written plainly
Before drawing any chart, write down the questions the dashboard must answer. For a professional services website they might be:
- How many qualified contact requests did we receive this week compared with the average of the last eight?
- Which channels did they come from?
- Which service pages generated the most requests?
- Is there any important page with a clear drop in visits?
Every element of the dashboard must serve at least one of these questions. If a chart answers nothing, remove it, even if it looks nice.
The page hierarchy
A dashboard is read like a newspaper: what matters most at the top, details below. A structure that often works:
- top row: three or four key numbers, each with a comparison against the reference period;
- middle section: the trend of those numbers over time, to spot patterns and anomalies;
- bottom section: breakdowns by channel, page or device, which help explain why.
Someone with two minutes stops at the first row; someone who needs to investigate scrolls down.
Comparisons that make sense
A number on its own says nothing. «Eighty-four requests»: is that a lot or a little? You always need a point of reference, chosen carefully:
| Comparison | When it fits | Risk |
|---|---|---|
| Previous period | stable trends | seasonality ignored |
| Same period last year | seasonal businesses | the site has changed meanwhile |
| Multi-week moving average | noisy data | reacts slowly |
| Set target | KPIs with goals | targets picked at random |
Watch the days of the week too: comparing a period containing two weekends with one containing a single weekend produces fake swings.
Choosing the right chart
You don't need imagination, you need legibility:
- lines for trends over time;
- horizontal bars to compare categories, sorted from largest to smallest;
- big numbers for key values, with the change next to them;
- tables when the user needs to read exact values.
Avoid pies with many slices, 3D charts and dual vertical axes, which make proportions ambiguous. Use colour to draw attention to what is unusual, not to decorate.
Visible definitions
Every number should have its definition one click away: which event it comes from, which filters apply, which time zone is used. When two people read «conversions» and mean different things, the dashboard generates arguments instead of decisions. A short note under each indicator, or a linked glossary, avoids the problem.
Fresh, reliable data
Always show when the data was last updated. If the most recent day is incomplete, flag it or exclude it: an apparent collapse today, caused only by data still being processed, erodes trust in the tool. Likewise, when you change the tracking, add an annotation on that date: whoever looks at the chart six months from now will understand why there is a step.
Maintenance
A dashboard ages. Campaigns end, pages get renamed, goals change. In my work I find a quarterly review useful, with these questions:
- which elements has nobody used?
- which new questions have emerged without an answer?
- are the definitions still correct?
Removing is as important as adding. A dashboard that only grows by accumulation quickly becomes unreadable again.
From reading to action
The final test is simple: after looking at it, does someone know what to do? If a drop in requests from one channel prompts a check of that campaign, the dashboard works. If everyone looks, nods and closes the tab, it needs rethinking, starting again from the questions.
In short
- Design each dashboard for a stated audience and viewing frequency.
- Write the questions first, then choose only the charts that answer them.
- Put a few key numbers with a sensible comparison at the top, breakdowns below.
- Make definitions, update time and tracking changes visible.
- Review the dashboard periodically and remove what nobody uses.
