The frozen horizon is the near-term period during which the production plan can no longer be changed without formal approval. Its role: to give the factory a stable target and stop the plan from jumping at every new demand signal. It is the dyke that protects execution from the noise of planning.
Why it matters
Many systems recompute the plan every night. Without a guardrail, the slightest demand change, including a mere forecast revision for next quarter, produces a new plan by morning. Supervisors find a reshuffled production queue, abandon it, and go back to their own priority list. The official plan becomes a fiction no one follows. It is the most common and most costly symptom of planning without a frozen horizon.
This phenomenon has a name, nervousness: a small change at the top propagates and multiplies across the whole BOM, generating dozens of rescheduled orders and unmanageable exception messages. The frozen horizon breaks this propagation at the root. By making the near zone untouchable, it turns a trembling plan into a reliable target production can actually aim at and hit.
Research has quantified a fundamental, often-ignored trade-off: a longer frozen horizon stabilizes the plan but costs more (you forgo recent information), while a shorter one lowers cost but raises nervousness. There is no universal setting, only a balance point specific to each activity.
The expert error to avoid is believing nervousness comes from the system. It comes from the failure to distinguish a real priority change from noise. A genuine customer emergency happens two or three times a week; the system reacts to every micro-variation in forecast. Filtering the noise before it reaches the near zone beats freezing blindly: that is the nuance separating rigidity from stability.
The mechanism
The frozen horizon is best understood by contrast. On the left, a plan with no frozen zone: at each recompute, it diverges from the very first period. On the right, the same plan protected: identical at the start, whatever happens.
On the chart, each curve is a version of the plan at a different recompute. On the left, without a frozen horizon, the three versions diverge from the start: production never knows what it will make tomorrow, because this morning's plan contradicts yesterday's. That is nervousness in action, a self-sustaining chaos.
On the right, the frozen zone (boxed) enforces stillness on the first periods: the three versions are rigorously superimposed there. Production has a stable, reliable target. Only beyond the dyke, where changes cost almost nothing, do the versions start diverging again. The same underlying uncertainty exists in both cases; what changes is only the decision to let it contaminate the short term or not.
Figure 1. On the left, a plan without a frozen horizon diverges at each recompute (nervousness). On the right, the frozen zone keeps the short term identical whatever happens. Illustrative schematic.
The traps
A poorly set frozen horizon quickly tips into one excess or the other.
If the frozen zone is shorter than the lead time, it protects nothing: changes land on orders whose components are already committed. The minimum length of the frozen horizon is the cumulative lead time below which a change hurts.
A total freeze, with no way out, always eventually cracks: facing a real customer emergency, someone bypasses the process and calls the shop floor directly. You need a formal, rare exception channel, reserved for genuine emergencies, rather than a wall people climb over in secret.
Freezing the short term does not remove the need to filter noise upstream. If the system lets every micro forecast revision travel up to the freeze boundary, pressure builds. The real remedy combines a near dyke and a filter that separates signal from noise.
Setting it
Setting a frozen horizon means settling three questions, product by product, by industrial reality.
The lower bound of the frozen horizon is the cumulative lead time: below it, a change can no longer be absorbed without extra cost. Start there, then lengthen if stability comes first, shorten if responsiveness does.
Decide who can unlock a change in the frozen zone, for what reasons, and at what approval level. This channel must exist (or it gets bypassed) but stay exceptional (or the freeze means nothing).
Put in place a rule that distinguishes a real priority change from a mere forecast oscillation, before the signal reaches the short term. It is this filter, more than the freeze itself, that durably stabilizes the plan.
Neighboring concepts
The frozen horizon formalizes a boundary the master schedule and MRP use constantly.
From knowledge to action
A plan the floor ignores is a plan without a frozen horizon. Our Planning, Forecasting & S&OP file installs the dykes and filters that make the plan credible.