Proxy Metric

A proxy metric is a measurable stand-in for an outcome that is hard, slow, or expensive to measure directly. It is useful exactly to the extent that it moves with the outcome, and that link is an assumption to state and check, not a property of the number.

A stand-in with a stated link

A fictional support team wants customers to wait less. It cannot measure “waiting” as customers feel it, so it tracks the median business minutes to a first human reply. The proxy is cheap and prompt, and it can fall while the outcome does not improve: if agents send quick holding replies, the median drops and customers still wait for an answer. Pairing it with the share of tickets answered within 60 business minutes and with a reopen guardrail does not close that gap: a holding reply improves the share too, and a ticket left open is never reopened. The team therefore separates three things, first human contact, a substantive answer, and resolution, keeps the share as a measure of first contact and says so, and checks substance separately by reviewing a sample of replies against a written rule. If the goal is resolution, resolution time and customer confirmation are the measures, not faster first contact. GOV.UK’s service guidance makes the same point in general terms: choose metrics that show whether the service is meeting its purpose, and be explicit about what each one indicates.

Activity is the commonest false proxy

Tickets closed per hour, features shipped, hours logged, and pages viewed measure activity. They are easy to count and easy to increase without improving anything, and they are often reported as if they were results. Ask what outcome a proposed metric stands for, how the link could break, and what would be seen if the proxy improved while the outcome did not. If no one can answer, the metric is not yet a proxy for anything.

Revisit proxies when they become targets: once a number is what people are judged on, the behaviours that move it without moving the outcome become more likely, and the link the proxy relied on weakens.

Reference: GOV.UK Service Manual: How to set performance metrics for your service. Examples here are illustrative.