The hidden cost of local optimization

Give every team a metric and tell them to improve it, and each team will. That is the appeal of local optimization: it is measurable, it seems fair, and it produces a steady stream of wins to report. What it does not produce, reliably, is a better organization. When every part optimizes its own number, the whole can quietly get slower, more expensive, and harder to steer.

Local wins can sum to a global loss

The mechanism is not mysterious. A team improving its own narrow metric will do so partly by exporting cost to its neighbors — hoarding a resource, tightening a standard that slows a downstream team, protecting its number at the expense of a shared outcome. Each move is locally rational and locally rewarded. Summed across the organization, they produce a system nobody designed and nobody can control.

The hidden cost is that this degradation is invisible on the very dashboards meant to prevent it. Every local metric is green. The organization congratulates itself function by function while the end-to-end outcome — the thing the customer actually experiences — deteriorates. Because no single team owns the whole, no single team sees the loss.

When all the parts are winning and the whole is losing, look at the seams between the parts.

The remedy is not to abolish metrics but to hold teams to outcomes that span the seams — measures that only improve when the interaction between functions improves. That reconnects local incentive to global performance, and it turns the reflex to optimize into a force that helps the system instead of slowly dismantling it.