Skip to lesson content

Practice Academy / Data & decision governance

From dashboard
to decision.

Measure what matters, understand what changed, and assign the next action.

General physician business education • Prepared November 15, 2026

Executive summary

A key performance indicator earns its place on an owner dashboard by informing a decision, not by being available in a report. Define the measure before calculating it: population, period, formula, exclusions, source, limitations, and accountable owner. Pair results with process measures and balancing measures so an apparent improvement does not hide new problems elsewhere. Compare consistent populations and examine patterns over time rather than treating one favorable change as proof that a new workflow caused success. This lesson uses a fictional denial-rate example to demonstrate the difference between arithmetic and inference. The goal is a repeatable governance loop: validate the signal, investigate the explanation, choose a proportionate action, and review what happened.

Learning outcomes

  • Build a measure definition another reviewer can reproduce.
  • Separate process, outcome, and balancing measures.
  • Interpret a changing rate without confusing association with causation.

Where this fits: Connect financial statements, revenue-cycle operations, and practice improvement. These are proposed management methods, not universal reporting requirements or benchmarks.

Start with the decision

What would you do differently?

Before adding a metric, name the question it informs. Is the practice deciding whether to investigate a cash shortfall, stage a new hire, resolve a vendor backlog, or examine a workflow? A red tile without an owner and a response is a display, not a control.

AHRQ’s measure-development guidance emphasizes specifications such as population, time period, numerator, denominator, exclusions, and data sources. Its source context is child-health quality measurement; the definition discipline below is a stated adaptation for business governance, not a validated financial measure (Agency for Healthcare Research and Quality [AHRQ], 2020).

Definition fieldQuestion to answer
PurposeWhich decision, objective, or risk does the measure inform?
Formula and unitWhat exactly is counted: claims, lines, encounters, dollars, people, or time?
Population and periodWho or what enters the calculation, and on what date basis?
Exclusions and missing dataWhich records are excluded, pending, or unresolved?
Source and validationWhich system is authoritative, and how is the result reproduced or reconciled?
Ownership and cadenceWho reviews it, how often, and who can authorize action?
Interpretation limitsWhat lag, small sample, changing mix, or incomplete follow-up matters?
ResponseWhat condition triggers investigation, and how is follow-up documented?

Maintain the definition with a version and effective date. If the formula changes, label the break in comparability instead of drawing one uninterrupted trend as if nothing changed.

A family of measures

Results, mechanisms, and tradeoffs.

IHI distinguishes outcome, process, and balancing measures: what resulted, whether intended work occurred, and whether changes produced unwanted effects elsewhere (Institute for Healthcare Improvement [IHI], n.d.-a). An owner dashboard should not reward one visible result while hiding the work or cost that produced it.

Leading is not the same as proven predictive

In this lesson, a leading indicator means an earlier signal hypothesized to influence a later result; a lagging indicator records a realized result. The label depends on the question and horizon. Completing eligibility checks may be an early signal for later claim performance, but completion alone does not prove accuracy or improved collections. A process measure is not automatically a validated predictor.

Review meaningful segments when an aggregate conceals differences. IHI discusses stratifying data to identify potential systemic inequities and avoid widening equity gaps (IHI, n.d.-a). Payer, location, or service segmentation here is a separate business adaptation, not a claim that the source validates these business measures; sensitive demographic analysis requires appropriate purpose, privacy safeguards, and clinical governance.

Fictional example • Not an effectiveness claim

The rate fell. What can you conclude?

Earlier period: 20 denied claims ÷ 200 adjudicated claims = 10%.
Later period: 5 denied claims ÷ 100 adjudicated claims = 5%.

The observed rate is five percentage points lower, and half the earlier rate, for the stated populations. That arithmetic does not establish why it changed.

The later period might have a different payer mix, service mix, or set of claims that reached adjudication. Pending claims may not yet be represented. The fact that a new workflow began between the periods is not enough to attribute the change to it. NIST distinguishes association from causation and explains that other factors can account for observed relationships (NIST/SEMATECH, n.d.).

Knowledge check: can the practice say its new eligibility process cut denials in half?

Not from these two figures alone. It can describe the observed rate difference with the definitions and periods. A causal claim requires stronger evidence and consideration of alternative explanations.

Questions before celebrating or blaming

  • Were definitions, exclusions, and source systems unchanged?
  • Are these submission cohorts followed long enough, or snapshots based on adjudication dates?
  • Are pending claims and missing data shown separately?
  • Did payer or service mix change?
  • Did rejections increase while denials fell?
  • Did staff workload or scheduling rework increase?

These are analytical questions, not reasons to ignore a promising result. The next step is to improve the evidence, not exaggerate the conclusion.

Observe over time

One before-and-after pair is not a trend.

IHI’s run-chart guidance supports plotting measures over time and annotating changes to understand patterns and whether improvement persists (IHI, n.d.-c). A chart is useful only when its scale, dates, definition, and missing observations are clear. Do not imply causation merely by drawing a line at the date of an intervention.

Small sequential tests can support local learning: plan a change, predict what will happen, collect observations, compare them with the prediction, and adapt before broader implementation (IHI, n.d.-b). This methodology does not grant permission to change clinical care, employment conditions, or regulated processes without the relevant authorization and professional oversight.

Build a meeting around decisions

  1. Validate: establish that the reported signal is real and reproducible.
  2. Investigate: identify plausible explanations and evidence still needed.
  3. Decide: choose a proportionate response within the decision owner’s authority.
  4. Assign: record the accountable person, next action, and review date.
  5. Review: compare what happened with what was expected, including unintended effects.

External benchmarks can inform questions only after comparing definitions, populations, specialty, size, payer mix, period, and operating context. A target is not a statistical control limit, and a local action threshold is not automatically a national standard.

A red tile should trigger inquiry, not automatic blame or an employment decision. Clinical outcomes and care-process changes remain under qualified clinical leadership. Administrative coordination can organize definitions, evidence, and follow-up within the engagement’s scope.

Next: diagnose turnaround needs · assign vendor accountability · Practice Academy.

References & source notes

Agency for Healthcare Research and Quality. (2020, June). Develop your own? AHRQ measurement guidance. Originally published September 2012; reviewed June 2020.

Institute for Healthcare Improvement. (n.d.-a). How to improve: Model for improvement: Establishing measures. IHI guidance.

Institute for Healthcare Improvement. (n.d.-b). How to improve: Model for improvement: Testing changes. IHI guidance.

Institute for Healthcare Improvement. (n.d.-c). Run chart tool. IHI tool.

NIST/SEMATECH. (n.d.). Experiments and experimental design. In e-Handbook of statistical methods (§ 3.1.3.6). NIST handbook.

Sources checked November 15, 2026. These are foundational methodological sources, not recent empirical evidence of MedCBO effectiveness. Examples and business adaptations are labeled educational synthesis. General education only; no universal benchmark, reporting-rule, clinical, or employment recommendation.