Metric Substitution
The metric becomes the mission — and hitting the number works, until the system hollows out
Optimising measurable KPIs improves reported performance, unlocks bonuses, satisfies boards, and creates the appearance of continuous improvement.
Metric substitution is a dark pattern: organisations successfully move the numbers they measure — while the underlying aims (correctness, safety, coherence, genuine growth) erode. The dashboard improves. The system does not.
The pattern
Leadership selects a proxy metric for a complex aim — tickets closed, lines of code, NPS, conversion rate, model accuracy on a benchmark, compliance checklist completion, or an (A)DR falsification trigger kept green — and rewards optimisation of the proxy. Teams respond rationally: they improve the metric by changing what is counted, gaming the measurement window, deferring hard cases, or reclassifying outcomes. That dynamic is the organisational face of Goodhart’s Law — see Charles Goodhart (1975 origin), Donald Campbell (1979 social-indicators form), and Marilyn Strathern (1997 generalisation). When gaming inverts the aim — worse than before intervention — see Cobra effect. Meta-loop control is the counter-discipline when the proxy lives inside closed-loop control.
The pattern works in the short and medium term. Numbers rise. Incentives pay out. Narratives form.
The claimed effect
Boards receive upward trends. Managers gain levers. Engineers and operators receive clear targets. The organisation appears data-driven.
In software systems
Implementation embeds metrics in CI gates, ranking algorithms, and automated prioritisation. Model teams optimise benchmark scores; production calibration drifts. Support systems close tickets by category manipulation. “Quality” becomes measurable artefacts — test coverage percentage without assertion quality, velocity without invariant health.
The telemetry layer is rich; the commitment layer is thin. Metrics enter slides as L0 truth while ground commitments remain unexamined.
In human organisations
Performance systems reward visible outputs over committed outcomes. Sales teams reclassify pipeline; clinicians document to satisfy coding metrics; teachers teach to the test. Research careers reward papers over provenance — publication over provenance under misaligned incentives. Local optimisation dominates because the metric is local and the aim was global.
Document-before-apply inverts: the metric is documented; the mechanism that produces it is not — until audit or harm reveals the gap.
In socio-technical systems
Algorithmic management scales metric substitution. Gig platforms optimise acceptance rates and delivery times while worker dignity and customer safety become externalities. Enterprise “AI transformation” KPIs count deployments, not commitment-boundary quality.
The socio-technical stack aligns on the number. The aims — safety, correctness, growth — are mentioned in strategy decks and absent in architecture.
What it violates
Values: H0: Humane Core — people become optimisers of scores rather than agents pursuing legitimate aims. Truth — the metric stands in for reality. Coherence — dashboards, operations, and accountability tell different stories.
Principles: Immutability of facts and document before apply.
Principles: Immutability requires committed facts, not rewritable aggregates chosen for presentation. Document-before-apply requires the mechanism to be specified before the metric is trusted.
How to protect the organisation
Name the aim before the metric. Link every KPI to an aim entry — coherence, correctness, safety — and review whether movement in the metric advances that aim or substitutes for it.
Separate L2 metrics from L0 commitments. Scores and dashboards are derivation; consequential decisions require commitment boundaries.
Audit for Goodhart signals: rising metrics with rising incidents, overrides, or reconciliation volume — structural evidence of substitution. See Goodhart’s Law for what Charles Goodhart originally said (1975, monetary targets), how Marilyn Strathern generalised it (1997), and why the popular aphorism is often misattributed.
Honest alternative: Commitment boundary — decide what counts as success at a named authority before irreversible action, with evidence attached.
Related aims under threat
Primarily growth and correctness — apparent scale and reported quality without structural health.