Features
Part of Dashboards: methods, tools and useful context
Dashboards metrics that show whether progress is real
Dashboard metrics worth watching per page and per estate, the signals missing from any usage log, and the one measure that actually settles retirement.
The measures on this page are not about the business. They are about the pages themselves: how you tell whether a dashboard is doing its job, and whether the estate as a whole is getting better or simply getting bigger.
Most teams measure this with view counts, which reward exactly the wrong thing. A page that has to be opened five times a day because it is confusing outscores one that answered the question at a glance.
What to take away
- Views measure traffic, not usefulness. A good page can drive its own view count down.
- The strongest signals are behavioral: exports, questions asked anyway, and manual work that stopped.
- Measure the estate as well as the page. A growing count of dashboards with flat usage is the clearest failure signal available.
Per page: what to watch
| Measure | How to read it | What it is not |
|---|---|---|
| Distinct weekly readers | Whether the intended audience actually arrives | Not total views, which one enthusiast can carry |
| Reader concentration | One person opening it constantly means it is a personal report | Not evidence of value |
| Session length | Very short is good for monitoring, very short is bad for exploration | Not comparable across page types |
| Export rate | High exports means the last step is missing from the build | Not a sign of engagement |
| Filter changes per session | Heavy filtering means the defaults are wrong | Not a sign of a powerful page |
| Time to first look after publish | Whether it reached anyone at all | Not adoption |
Read these against the page's declared purpose. A monitoring page that people leave after twenty seconds is working. An exploratory page with the same profile is not.
The signals that are not in the usage log
The asked anyway rate. Count the questions raised in meetings that the dashboard already answers. This is the single most informative measure on this page and it has to be collected by hand, by someone sitting in the meeting with a tally.
The manual work that stopped. A spreadsheet nobody updates any more, an export that no longer runs, a Monday morning assembly job that disappeared. Name these at launch and check for them a month later.
Screenshots in circulation. If people paste images of your page into messages, they are working around something: no link, no permission, or no way to share a filtered view. Each is fixable and none appears in usage data.
Questions about the number itself. Repeated "is this right" questions are a metadata failure, not a data failure. The definition, source, and freshness are not visible where the doubt occurs.
Per estate: the numbers that matter more
- Published pages, and the direction of that count. Growth with flat readership is accumulation.
- Share opened in the last quarter. In most estates this is far below what anyone expects, and measuring it once is usually enough to start a cleanup.
- Pages with a named living owner. Ownership decays quietly as people change roles.
- Pages defining their own version of a governed concept. Each one is a future disagreement in a meeting.
- Retirements per quarter. A team that never retires anything will spend next year maintaining last year's questions. Any number above zero is a healthier signal than most usage statistics.
- Broken tiles found by readers rather than by tests. If readers are your monitoring, the tests are inadequate.
Freshness and failure, measured honestly
Two measures, both easy and both usually absent.
Time from source data landing to the page reflecting it, tracked as a distribution rather than a promise. Where a delivery commitment exists, measure it the way any service-level agreement is measured: report the breaches, not the average. And the count of intervals where the page served stale data without saying so. The second is the one that destroys trust, because a page that looks fine and is wrong costs more than a page that is visibly broken.
Publish the last updated time on every page and alert on staleness rather than on job failure. Jobs can succeed and still deliver nothing.
The measure that actually settles it
Once a quarter, ask the intended reader one question: what did you do differently because of this page? Then ask them to describe the most recent instance.
A specific answer, with a date, means the page is working. A general endorsement means they like having it. No answer means you have a candidate for retirement, and you now have the evidence to propose one without an argument.
This takes ten minutes per page and outperforms every automated measure listed above. It does not scale to hundreds of pages, which is itself an argument for having fewer.
What not to measure
Do not build a scorecard that ranks dashboards, and do not report page views to leadership as a measure of the data team's contribution. Both create pressure to publish more pages, which is the opposite of the goal, and both are easy to move without helping anyone.
Related reading on this site
For what a page is for and when to retire it, see dashboard design. The declared purpose these measures are read against comes from the one page specification. For the shapes and traps of the underlying business measures, see metric definitions. For reading the request queue as the same kind of behavioral evidence, see business intelligence, and for the fixed cadence artifacts these measures apply to differently, see reporting.
Common questions
Our tool does not expose usage data. What can we still do?
The behavioral signals, which are the better ones anyway: the asked anyway count, the manual work that stopped, and the quarterly question to the reader. None requires access to a log.
Is a low view count grounds for deletion?
It is grounds for asking. Announce an archive, and see who objects. The one person who cared identifies themselves, and you have learned who the real reader was.
How often should we review the estate?
Quarterly is enough. More often and nothing has changed; less often and the accumulation gets large enough that nobody wants to start.
What is a reasonable retirement rate?
Whatever keeps the count of pages roughly stable while the business keeps asking new questions. The target is a steady state, not a number.
