Analytics & BI Tools · Chapter 8 of 18
Making analytics actually get used
The failure that no tool prevents: dashboards nobody opens, and numbers nobody trusts.
Most BI programmes do not fail on tooling. They fail because the output is not trusted, not found, or not actionable — and each of those has a specific remedy.
Not trusted
Trust breaks the first time two reports disagree in a meeting, and it does not recover on its own. The remedies: one definition in a modelled layer; a visible certification marker so users can tell an owned dashboard from someone's experiment; a named owner on every certified asset; and freshness shown on the dashboard itself, because a stale number presented confidently is worse than no number.
Not found
In most organisations there are thousands of reports and no way to tell which is current. The remedies are unglamorous: a catalogue with search that people actually use, aggressive archiving of anything unopened for six months, and a naming convention. Archiving is the one that gets resisted and the one that helps most.
Not actionable
A dashboard that shows what happened without indicating what to do gets opened twice. The remedies: design around a decision rather than around available data; include a comparison so a number has meaning; and alert on exceptions rather than expecting people to check.
The metrics that tell you whether it is working
Report: proportion of certified assets with a named owner, weekly active consumers rather than total accounts, and the number of reports archived. That last one is the health indicator nobody tracks — a BI estate that never removes anything is accumulating distrust.
Do not report: number of dashboards created. It rewards exactly the behaviour that produced the mess.
Why dashboards go unused, and what predicts the ones that don't
The usual explanation for low adoption is training. It is almost never training. The dashboards that get used share a small number of properties, and none of them are about the tool.
| Property | The failure it prevents |
|---|---|
| It answers a question someone already asks weekly | Building for a question nobody had. The most common cause of an unused dashboard and the hardest to admit |
| The numbers reconcile with the source people already trust | The credibility failure. One mismatch against the finance number and the dashboard is dead — not distrusted, dead, permanently |
| It is fast | People will not wait. A dashboard over about five seconds is one people stop opening, and they will not tell you why |
| It says what to do, not just what happened | A wall of numbers with no threshold and no comparison leaves the reader to do the analysis, which is the work they wanted done |
| Someone owns it | Slow decay. An unowned dashboard drifts out of correctness and nobody notices until it is quoted in a meeting |
The reconciliation problem is the whole game
A dashboard that disagrees with the number someone already has is worse than no dashboard. It does not create doubt about the other number; it creates doubt about yours, permanently, and no correction fully recovers it.
And the disagreement is usually legitimate on both sides — different grain, different date boundary, different treatment of returns or internal transfers, different currency conversion date, different definition of an active user.
So the work is not making the numbers match. It is making the difference explicable. A documented, agreed reconciliation — ‘ours is 3% lower because it excludes internal transfers, and here is that line’ — converts a credibility problem into a definition, which people accept immediately.
Do this before launch, not after the first challenge. Doing it afterwards reads as a defence; doing it beforehand reads as rigour, and it is the same work.
The related habit: when a metric has more than one legitimate definition, put the definition on the dashboard next to the number. It costs a line of text and prevents the argument entirely.
Check yourself