Regime Change

Definition

The moment the world your data was learned from stops applying: a lost flight route, a new 200-room competitor, a pandemic, an office-market shift. There is plenty of historical data — but the relationships inside it are no longer valid.

What it tells you

Models keep calculating through a regime change as if the old rules still held, because nothing in the database says otherwise. Only a human can declare “it works differently from now on”, because only the human knows the cause. Miss the call and every forecast and rate recommendation inherits a systematic bias; make it, and you know to lean on judgement while the new pattern builds up in the data.

How to track it

Watch for persistent one-sided forecast errors (the model keeps missing in the same direction), overrides that consistently beat the model, and outside-the-data news that changes demand structurally. At a suspected regime change, put machine recommendations under tighter human review until accuracy recovers.

Where it fits

One of the three classic failure modes of the machine layer — alongside unprecedented events and contextual decisions. The free Academy covers it in depth: AI and RM — what the machine sees and what it doesn’t.