Oriental Nippon Development
Back to blog
Insights

Define the Measure Before You Build the Dashboard

In every BI project we've modernised, the hardest work wasn't the pipeline or the charts. It was agreeing what a number means.

When we rebuilt more than thirty legacy reports for a hospital revenue-cycle operator, the visible deliverable was Power BI. The real deliverable was a short document: one definition per measure.

We see the same pattern in almost every BI project. The pipeline and the charts are the easy part. Agreeing what a number means is where the time goes.

The symptom

Two facilities compare their "collection rate" and get different answers. Both are right, by their own report's definition. Multiply that across dozens of reports and leadership stops trusting any of them.

Here is a simplified example of how that happens:

Report"Collection rate" calculated asSame month result
Monthly summaryCollected ÷ total placedLower
Facility packCollected ÷ placed minus cancelledHigher
Agent dashboardCollected ÷ accounts workedHigher still

Illustrative example. The point is that all three are reasonable, and all three are called the same thing.

What we do differently

  • Inventory consumers, not just reports. Who reads it, and what decision does it support? Some reports turn out to have no reader at all.
  • Reconcile definitions before building. On the reporting rebuild this took 2–3 weeks on its own, and every decision went into a log.
  • Agree the dozen measures leadership actually decides on. For a multi-branch staffing group, we agreed placements, revenue, margin and branch profitability before writing a single pipeline.
  • Give every measure an owner and a change process. Governance is what keeps a dashboard trustworthy in year two.

How long each part takes

From the delivery plan of the reporting rebuild:

Delivery timelineIndicative total: 16–24 weeks
  1. 01Inventory

    2 wks

    Every report, consumer and measure catalogued

  2. 02Definition reconciliation

    2–3 wks

    One agreed definition per measure, decisions logged

  3. 03Rebuild waves

    8–12 wks

    Reports rebuilt in consumer groups

  4. 04Parallel run

    2–4 wks

    Old and new compared for a full cycle

  5. 05Automation

    2–3 wks

    Scheduled refresh, export and delivery

Definitions take only a few weeks on the chart. But a definition decided late changes reports already built, so getting them early saves time in the rebuild.

The visible deliverable was Power BI. The real deliverable was one written definition per measure.

Then build

With definitions settled, the engineering is straightforward: scheduled ingestion with failure alerts, a tested model, dashboards designed around the decision, and automated delivery so nobody exports by hand.

A checklist you can steal

  • Does each measure have exactly one written definition?
  • Does each measure have a named owner?
  • Can you trace each number on a dashboard back to source rows?
  • Is there a process for changing a definition, and does it notify readers?
  • Did old and new reports run side by side for a full cycle?

The payoff

When two people compare a number, they are comparing the same thing. That is worth more than any chart. Read the legacy reporting modernisation case study for the full story.

Sources and further reading

  1. Conformed dimensions, Kimball Group
  2. The dbt Semantic Layer, dbt Labs
  3. What is Power BI?, Microsoft Learn

Related case study

30+ Legacy Reports Rebuilt on a Modern BI Stack, with One Definition per Measure