Choosing performance measures a department will actually use

Reference article · 5 min read

A city manager asks for a quarterly performance report. A board adopts a strategic plan with reporting language in it. A grant condition requires results twice a year. The department picks measures under time pressure, usually from whatever the current system exports. Two budget cycles later, half the list is stale and the report exists because one person remembers how to build it.

A measure set that lasts is small, defined in writing well enough that a new analyst can reproduce last quarter's value, and built on data the department already produces during normal work. Its review schedule matches decisions someone actually makes, which in local government usually means the budget calendar. None of that takes a new reporting team. It takes fewer measures, defined properly.

Four kinds of measures, and what each answers

Performance frameworks sort measures into the same few families, and a report made entirely of one kind tells you little.

• Inputs. Resources consumed: budgeted dollars, staff positions, vehicle hours, licensed seats. Usually in the budget document already, and a description of capacity rather than results.
• Outputs. Units of work completed: permits issued, inspections performed, service requests closed, lane miles resurfaced. Under the department's control and useful as context, but volume is not value.
• Outcomes. The condition the work is meant to change, or the experience of the person receiving the service: pavement condition rating, median days from application to decision, share of requests resolved without a repeat call. Partly outside the department's control, and the only kind that answers what a governing body is asking.
• Efficiency. Resources against results: cost per inspection, staff hours per case, cost per lane mile treated. These work only when input and output are both defined carefully.

Aim for one or two outcomes per major service, a few outputs for context, and an efficiency measure where cost per unit is a live budget question.

Write the definition, not just the name

"Average response time" means several different things depending on when the clock starts. The fix is a one page definition sheet per measure, stored where the next person will find it.

1. What the measure tracks, in plain language, and what decision it informs.
2. The exact calculation, including numerator and denominator for any rate or average.
3. The source system, and the specific report, query, or field the value comes from.
4. Inclusions and exclusions. Does a reopened case count once or twice? Do weekends count in elapsed days?
5. The reporting period, the data cutoff, and how late-arriving records are handled.
6. The owner by position title rather than by person's name, plus a backup title.
7. A revision log: the date of each definition change and the reason for it.

Then run the turnover test. Hand the sheet to someone who has never produced the number and ask for last period's value without answering questions. If the result does not match, the definition is not finished. That afternoon is what keeps a measure from quietly changing meaning after a retirement.

Spotting measures that drive the wrong behavior

Any measure that carries weight changes behavior, which is the point. The risk is changing it in a direction nobody intended.

• Volume without quality. Counting closures rewards closing records that are not resolved, so pair throughput with a rework measure.
• Speed without completeness. Cycle time pushes work toward whatever step stops the clock, so define the endpoint as the point of value to the resident, not an internal handoff.
• Threshold cliffs. A target stated as the share of cases under a limit creates pressure at the boundary, and the distribution tells you more than the percentage.
• Measures that punish honest reporting. When counts of a problem are read as failure, reporting falls before the problem does.
• Measures outside the unit's control. A number driven by weather or another agency produces defensive reporting.

Before adopting a measure, ask how a reasonable person would make the number look better without doing better work. If the answer comes easily, add a balancing measure or pick another one.

Tie the review cadence to the budget calendar

Reporting on its own schedule gets skipped. Reporting that feeds a meeting someone already attends survives.

• Monthly. A short check by the measure owner, for data quality and early warning.
• Quarterly. A review with the department head: trend, exceptions, and definition changes recorded in the log.
• Annually, with budget preparation. Measures go into the submission next to the request they support, and any measure nobody used all year is retired.

Grant conditions carry their own due dates, so map them onto that calendar rather than running a parallel process, and use the same numbers in every report.

How to start, and what to require of the systems

Start with the services the department would defend first in a budget hearing and write one outcome measure for each. Pull a year of history first, so the report opens with a trend rather than a point, and publish definitions next to numbers.

When the data lives in a vendor system, the contract decides whether the measure is sustainable. Worth asking during evaluation and putting in writing:

• A current data dictionary for the fields behind the measures, and notice before report logic or field definitions change.
• Read access to the underlying data, or a scheduled export in a documented open format, not only the vendor's fixed reports.
• Confirmation that ad hoc queries and exports carry no extra fee.
• Retention of historical values, and whether a configuration change restates prior periods.

HQT works with state and local agencies on data and reporting systems, including the unglamorous part: defining the measure, finding the source of record, and keeping the number reproducible after the person who built it moves on.

Frequently asked questions