Topic 25 of 64· Governmental performance and delivery metrics

Public Sector KPIs

A key performance indicator (KPI) is a chosen, tracked measure that stands in for whether a public service is doing its job well. In government the choice of KPI is never neutral: because KPIs attach to budgets, league tables, and careers, the act of selecting one shapes the behavior of everyone downstream of it, often more than the policy that created the service.

Why it matters

Charles Goodhart's 1975 observation on monetary policy — later popularized by Marilyn Strathern as "when a measure becomes a target, it ceases to be a good measure" — is the single most important warning label in public-sector performance management. A KPI chosen to describe a system starts to distort that system the moment resourcing, pay, or political survival is tied to it. The canonical illustration is NHS ambulance response times: when the eight-minute Category A response target became binding, some trusts were shown to have "stacked" ambulances just outside the response-time clock, or reclassified calls, to hit the number without changing patient outcomes. The UK National Audit Office's guidance on choosing and using performance indicators — set out across its value-for-money reports and its "Performance Measurement by Regulators" and "Choosing the Right FABRIC" framework (Fit for purpose, Appropriate, Balanced, Robust, Integrated, Cost-effective) — exists precisely because departments kept picking indicators that were easy to report rather than indicators that were hard to game. A software engineer who ships the dashboard that a minister or director will be judged against is, whether they intend it or not, designing the incentive structure of a public institution.

The math

KPI design is a framework-shaped topic, but the evaluation of a candidate KPI is a repeatable checklist, not a formula:

For each candidate KPI, score against:
  Fit for purpose  — does it measure the outcome, or a proxy several steps removed?
  Appropriate      — does it belong to the people who can actually influence it?
  Balanced         — is it paired with a counter-metric that catches gaming?
  Robust           — can it survive audit, or is it self-reported and unverifiable?
  Integrated       — does it fit the wider set, or push against another KPI?
  Cost-effective   — does collecting it cost more than the decision it informs?

Leading vs lagging split:
  Leading indicator  → predicts future outcome, but often gameable (e.g. calls answered <60s)
  Lagging indicator  → confirms the outcome happened, but arrives too late to steer (e.g. annual
                        satisfaction survey)
  A defensible KPI set pairs at least one of each per objective.

Worked example

Ambulance trust: a trust reports a Category A (life-threatening) response-time KPI of "75% of calls responded to within 8 minutes." In one quarter, 6,000 Category A calls come in; 4,500 are met within 8 minutes, giving 75.0% — apparently on target.

Headline KPI = 4,500 / 6,000 × 100 = 75.0%  (meets the 75% threshold)

But a Goodhart audit adds a counter-metric: mean response time for the slowest 10% of calls.

Slowest decile mean response = 34 minutes (up from 19 minutes two years earlier)

The trust is hitting the target while the tail — the calls most likely to be genuinely life-threatening once triage is imperfect — has got much worse, because crews are being prioritized toward calls close to the 8-minute cliff-edge rather than toward clinical urgency. The single KPI told a false story; the paired KPI told the true one.

Software engineering connection

Engineers building performance dashboards for government are, functionally, designing the incentive API of the organization. Practical implications: instrument the denominator as rigorously as the numerator (a KPI reported as a bare percentage invites denominator gaming — see cost-per-transaction for the same trap in digital services); build counter-metrics into the same dashboard rather than a separate report nobody reads, so gaming is visible at the point of decision; and version the KPI definition, because a silent redefinition (changing what counts as a "call," a "case," or a "completion") is functionally equivalent to changing the target without announcing it. A public value scorecard is one structured way to stop a single KPI from being read in isolation, and outcomes-based accountability is the discipline of choosing population-level KPIs that a single team cannot unilaterally distort.

Pitfalls

  • Picking the easy-to-collect metric over the meaningful one: call-answer time is trivial to log; whether the call solved the citizen's problem is not — but only the second one is the outcome. Resist defaulting to what the system already emits.
  • No counter-metric: any KPI attached to money or reputation will be gamed at the margin; ship it with a paired metric that catches the likely gaming vector before publishing it.
  • Redefining the metric without a changelog: swapping "calls received" for "calls answered" to flatter a trend destroys the time series' credibility the moment it's discovered — always publish a definitions changelog alongside the numbers.
  • Confusing activity with result: counting inspections completed is an output; counting premises brought into compliance is closer to the outcome (see outcomes vs outputs).

Sources

  • National Audit Office, "Choosing the Right FABRIC: A Framework for Performance Information." https://www.nao.org.uk/
  • Marilyn Strathern, "'Improving Ratings': Audit in the British University System," Social Anthropology, 1997 (formulation of Goodhart's law as commonly cited).
  • National Audit Office, investigations into NHS ambulance service performance reporting. https://www.nao.org.uk/