- Design dashboards around decisions, not around available data.
- Agree on metric definitions before building visualisations.
- One trusted number beats ten conflicting ones.
- Data quality monitoring is as important as uptime monitoring.
The dashboard graveyard
Most organisations have a dashboard graveyard: dozens of reports built with enthusiasm, opened twice and then quietly ignored. The usual diagnosis is “people don’t use data”. The real cause is almost always that the dashboards were built around the data that was available, rather than the decisions people actually need to make.
When a regional manager opens a report on Monday morning, they are not looking for charts. They are trying to answer a question: Which stores need attention this week? Which customers are at risk? Where should I put my team?
Start with decisions
A decision-intelligence programme starts with interviews, not tools. For each leadership role, we document the five to ten recurring decisions they make, how often they make them, what information they use today and what a better decision would be worth.
- What decision are you making?
- How often, and by when?
- What would change your mind?
- What happens if you get it wrong?
This list becomes the backlog. Every chart, metric and alert must trace back to a decision on it.
One definition of truth
Nothing destroys trust faster than two reports showing different revenue figures. Before building anything visual, agree on a metric dictionary: what exactly counts as an active customer, when revenue is recognised, how returns are treated. Each metric gets an owner, a written definition and a single implementation in the data model.
“Leadership will forgive a slow dashboard. They will not forgive a wrong one.”
A modern data foundation
The technical foundation does not need to be exotic. A typical modern stack ingests data from operational systems into a cloud warehouse, transforms it with version-controlled models, tests it automatically and serves it to a BI tool through a governed semantic layer.
What matters most is that transformations are code — reviewed, tested and deployed like any other software. That is how you guarantee that “revenue” means the same thing in every report, every time.
Monitor data quality like uptime
Pipelines break silently. A source system changes a field, a nightly job fails, a currency conversion goes missing. Automated tests on freshness, volume, uniqueness and accepted values catch most issues before anyone sees a wrong number. When something does go wrong, a visible banner on the dashboard is far better than silent inaccuracy.
From reporting to action
The most valuable analytics rarely look like dashboards at all. They are alerts sent when a threshold is crossed, ranked lists of accounts to call today, forecasts embedded in planning tools. Push insight to where decisions happen, and measure success by decisions improved — not reports viewed.
Start with one leadership team, one set of decisions and one trusted dataset. Earn trust there, then expand.