The Proxy-Metrics Trap

Definition

An activity-based proxy metric measures whether the mechanics of a change happened — how many people were trained, how many teams adopted a tool, how many ceremonies run — rather than whether the change produced the intended business result. Kersten's diagnosis is that these proxies can read as unambiguous success (every activity checkbox ticked) while the underlying business outcome quietly fails, because the proxy was never causally tied to the outcome it was meant to indicate. He pairs this with local optimization: even a genuine improvement, if applied to a part of the value stream that is not the actual bottleneck, produces measured activity without moving the end-to-end result.

In the Book

The book's central cautionary case is Nokia. Consultants used the "Nokia Test" — questions about Scrum ceremony adherence — as proof that Agile scaled to large enterprises, and by every activity metric Nokia's transformation was on track, "right down to the adoption of the Agile tool." But Kersten's interviews found developers suffering severe friction because the Symbian OS's architecture made adding features extremely difficult — a bottleneck the activity metrics never surfaced, because "the tie-in between business outcomes and software production metrics was either not explicit or nonexistent." He states it plainly: "the proxy metrics could deem the Nokia Agile transformation a success" while the actual business result — the inability to shift to a software-and-screen-centric phone in time to compete with the iPhone — failed. He generalizes this as Nokia's Agile rollout being a local optimization: enormous investment went into making development teams more agile, when development was never the bottleneck; the real constraint was upstream, in an architecture and a business-to-engineering feedback loop that no amount of team-level agility could fix. The same pattern reappears with "LargeBank," whose billion-dollar transformation was tracked entirely against cost reduction rather than business outcome — Kersten predicted its failure the moment he saw the metric being used.

Why It Matters

This concept names a specific failure mode of measurement systems: a metric that correlates with success under old conditions keeps being reported as if it still does, even after the causal link to the real outcome has broken (a close cousin of Goodhart's Law, argued here through a concrete corporate autopsy rather than abstractly). It gives a portable diagnostic — before trusting a metric, ask whether it is tracking the activity of doing a thing or the result the activity was supposed to cause, and whether the metric is scoped to the actual system bottleneck or to a locally convenient but non-critical piece of it. That question applies identically to training completion rates in a company, output metrics in a bureaucracy, or vanity engagement numbers in a product.