---
slug: local-optimization
type: antipattern
summary: "Improving individual parts of a system in ways that do not improve, and often actively harm, the performance of the whole."
created: 2026-05-21
updated: 2026-06-02
related:
  five-focusing-steps:
    relation: corrupts
    note: "Local optimization is the failure mode the Five Focusing Steps are designed to prevent: improvement effort is directed at whichever resource looks busiest instead of at the resource that limits the system."
  subordination:
    relation: prevents
    note: "Subordination is the named refusal of local optimization: non-constraints are run at the rate the constraint can consume, not at their own local optimum."
  improving-non-constraints:
    relation: produces
    note: "Improving Non-Constraints is the operational form local optimization takes when an organization tries to act on the trap; the diagnostic move is the same and the corrective is the same."
  breaking-wrong-constraint:
    relation: produces
    note: "Breaking the Wrong Constraint is what happens when local optimization escalates to capital spending: the busy-looking non-constraint is elevated, and the true system constraint remains untouched."
  constraint:
    relation: depends-on
    note: "Local optimization is unintelligible without the constraint concept: the trap is to optimize a resource that is not the constraint, which presupposes the distinction the constraint vocabulary names."
  goal:
    relation: depends-on
    note: "The trap is defined relative to the system's goal: a local optimum is a maximum of a local measure that does not move the rate at which the system produces goal units."
  throughput:
    relation: prevents
    note: "Local optimization prevents throughput from rising even as local efficiency reports improve; throughput is the global measure the trap routes effort away from."
  inventory:
    relation: produces
    note: "A non-constraint run at full local utilization produces work-in-process inventory the constraint cannot absorb; the inventory pile is the visible signature of the trap."
  operating-expense:
    relation: produces
    note: "Local optimization tends to add operating expense (overtime, hires, machine hours, energy) at non-constraints, with no corresponding gain in throughput."
  cost-world-thinking:
    relation: complements
    note: "Local optimization is the operational behavior; Cost-World Thinking is the measurement-and-paradigm frame that rewards it. The two travel together and reinforce each other across the implementation."
  throughput-accounting:
    relation: prevents
    note: "Throughput Accounting prevents the trap at the measurement layer: T, I, and OE are global measures, and a non-constraint's local efficiency is not one of them."
  cost-throughput-thinking:
    relation: corrupts
    note: "Local optimization is the operational tell that the cost-world paradigm is still running the decisions, regardless of what the implementation plan declared."
  antipatterns-index:
    relation: part-of
    note: "Listed in the Antipatterns Index sub-page of Foundations alongside Improving Non-Constraints, Breaking the Wrong Constraint, and the Performance Measures and Project Management antipatterns."
---

# Local Optimization

[Common Trunk (Foundations)](learning-tracks.md#common-trunk): [prev: Exploitation Without Elevation](exploitation-without-elevation.md) -> Local Optimization -> [next: Improving Non-Constraints](improving-non-constraints.md)

> **Antipattern:** A recurring trap that causes harm — learn to recognize and escape it.

*Improving individual parts of a system in ways that do not improve, and often actively harm, the performance of the system as a whole.*

> "A system of local optimums is not an optimum system at all; it is a very inefficient system."
>
> Eliyahu Goldratt and Jeff Cox, *The Goal*, Ch. 25 (Jonah, in dialogue with Alex Rogo)

Local optimization feels responsible at the point of action. A supervisor keeps a machine busy, a department raises its utilization score, or a project team clears its own queue faster. The trap appears when that local gain bypasses the constraint: the part improves, the whole system does not, and the measures still congratulate the part.

## Context

The trap is foundational because every major TOC application is built to refuse it. [Drum-Buffer-Rope](drum-buffer-rope.md), [Critical Chain Project Management](critical-chain.md), [Replenishment Pull](replenishment-pull.md), and the [Mafia Offer](mafia-offer.md) all route improvement effort away from local optima and toward the resource that actually limits the system. Without that routing discipline, the inherited behavior returns: every part tries to look efficient on its own.

The antipattern presupposes three priors. [The Goal](goal.md) supplies the standard against which "improvement" is judged: a move is an improvement only if it raises the rate at which the system produces goal units. [Constraint](constraint.md) names the resource or policy a system-level improvement has to act on. The [Five Focusing Steps](five-focusing-steps.md) supply the discipline whose absence permits the trap. When the loop is not running, local incentives fill the vacuum and improvement effort drifts wherever it is loudest. [Subordination](subordination.md), the third focusing step, is the corrective for the same situation: non-constraints run at the rate the constraint can use, not at their own local optimum.

The *TOCICO Dictionary*'s entry on the local optimum is short and precise: the maximization or minimization of a measure of *a part* of a system, with a usage note that traditional cost accounting treats such actions as having global impact while TOC does not. The *TOC Handbook* spreads the treatment across chapters. Goldratt's Ch. 1 establishes the systemic frame. Srikanth's Ch. 8 on DBR names local efficiency as the antagonist the rope releases the plant from. Budd's Ch. 13 on traditional finance measures explains the accounting mechanism that rewards the trap. Barnard and Immelman's Ch. 16, Holistic TOC Implementation, documents cases in which the measurement frame had to be retired before the implementation could hold.

## Problem

A practitioner trying to improve a system inherits a set of measures that point everywhere except at the constraint. Departmental efficiency targets, machine utilization percentages, individual productivity scores, on-time delivery rates at each handoff, and per-unit cost trends are calculated for every resource the organization runs, regardless of which resource limits the system. Each measure has a sponsor, an audit trail, a place in the quarterly review, and a story about how raising it represents progress. Most of the measures are improvable through familiar moves: a faster machine, a more skilled operator, a tighter changeover, fewer interruptions, more hours. None of them, considered in isolation, asks whether the resource being optimized is the system's constraint.

The result, repeated across forty years of TOC case literature, is an organization in which most resources have measurable local efficiency gains, the executive dashboards trend upward, and the system's throughput is flat or declining. The pile of work-in-process inventory ahead of the constraint grows; lead times extend; the customer waits longer; finished goods accumulate in the wrong mix; cash conversion slows; the cost-per-unit reports improve as the denominators (units produced) climb without the corresponding sales. The local measures all clear their targets. It's the system that underperforms, and nobody on the management dashboard can see why.

The diagnostic question is not "which resources can run faster?" but "which one resource is the limit, and what would actually change if a non-limit ran faster?" Until that question is asked and answered, every improvement project is a candidate for the trap.

## Trap

The trap has a stable shape across domains. Five operational moves are the visible expression.

*Running non-constraints at full utilization.* A non-constraint resource is given a 100% utilization target, treated as wasted capacity when idle, and rewarded for sustained throughput against its own local rate. The non-constraint produces, accurately and on time, units the constraint cannot absorb at the rate they arrive. The units accumulate. The non-constraint's utilization report looks excellent. The system's throughput does not move.

*Equating local efficiency with system improvement.* Operations meetings are organized around local-efficiency reports. A department's improvement from 78% to 84% utilization is treated as a 6-point gain. Whether the department is the constraint, whether the additional output passes a constraint at all, and whether the extra utilization corresponds to additional sales are not asked. The local report displaces the global question.

*Eliminating "waste" at non-constraint resources.* A Lean-style audit identifies idle time, slack, and protective capacity at non-constraints and characterizes them as waste. Headcount is reduced, shifts are consolidated, schedules are tightened. The resources lose the protective capacity TOC reads as the system's catch-up reserve. The next disruption at the constraint cannot be recovered from because non-constraints no longer have the spare capacity to run faster while the constraint catches up; the implementation's resilience falls; the lead-time variability rises.

*Investing in capacity at non-constraint resources.* A capital project upgrades a busy-looking but non-constraint resource. The upgraded resource now produces faster, which moves more material to the queue in front of the constraint, which does not move throughput. The capital is sunk; the constraint is unchanged; the inventory grows; the eventual diagnosis is that the wrong investment was made. The [Breaking the Wrong Constraint](breaking-wrong-constraint.md) antipattern is the escalation of this move from operating decisions into capital decisions.

*Rewarding people, departments, and managers for local achievements.* Bonuses, performance reviews, promotions, and public recognition track local measures. The people closest to the work have their careers tied to the trap's continuation. The implementation is fighting not against ignorance but against the present alignment of incentives; when the implementation plan asks a department head to deliberately let their utilization drop so the constraint can pull more, the answer requires the department head to absorb a measured loss the scorecard will publish.

Two structural points run through all five moves. First, every move is locally rational: the operator's, supervisor's, department head's, and CFO's reasoning is internally consistent given the measures they are accountable for; the irrationality lives at the level above. Second, the moves are mutually reinforcing: the measurement frame produces the operating behavior; the operating behavior produces the inventory and lead-time signals; the cost-accounting reading of those signals recommends moves that intensify the original behavior. Goldratt's Jonah, in the dialogue with Alex Rogo in *The Goal*, names the dynamic in the epigraph: a system composed of local optima is not an optimum system. The aphorism reads as paradox to a cost-world ear; it reads as a property of throughput-world arithmetic to a TOC ear.

## Consequences

**Throughput is flat or declining despite favorable local reports.** The single most diagnostic signature. Departmental efficiency, utilization, and unit-cost measures trend in the directions the implementation plan wanted; the rate of goal-unit production through sales doesn't move. The discrepancy is the antipattern's tell. Read the executive dashboard against the income statement: if the dashboard improves while T (revenue minus totally variable costs) doesn't, the dashboard is reporting on local optima rather than on the system.

**Work-in-process inventory grows in front of the binding resource.** A non-constraint running at full local rate generates the inventory pile a constraint cannot absorb. The pile is visible on the shop floor, in the project management software's backlog of feeder-chain tasks, in the queue depth of code review, in the unscheduled patients waiting for assessment. The pile is the operational signal that release rates upstream of the constraint are tied to upstream local capacity rather than to the constraint's pace.

**Lead times rise and become more variable.** Inventory ahead of the constraint is time the unit will sit and wait. The expected wait is roughly the queue depth divided by the constraint's rate, and the queue depth grows with the upstream-downstream rate mismatch. As the queue grows, so does the average lead time; as the queue varies (which it always does under stochastic arrivals), so does the lead time's variance. Customers wait longer and the wait is less predictable; both effects show up in service-level reports, both can be misdiagnosed as forecasting problems or planning problems before the antipattern is identified.

**Operating expense rises without throughput compensation.** Overtime at non-constraints, additional shifts at non-constraints, capital spent on non-constraints, energy and consumables spent on producing material the constraint will not absorb: all show up as operating expense. The corresponding throughput gain is absent because the constraint, which sets the system's rate, has not been changed.

**The implementation reads as failed.** Eighteen months in, the operations VP is asked what the program produced. The local-efficiency improvements are difficult to defend without the corresponding revenue gain; the customer-service metrics are unchanged or worse; the executive sponsor is asking different questions than at kickoff. Mabin and Balderstone's cross-domain meta-review of more than 80 documented TOC implementations records, on this point, that the implementations that completed Steps 2 and 3 in order produced systematically better outcomes than the implementations that ran Step 2 alone or inverted Steps 2 and 4. The cases that *appear* to have failed are, on inspection, the cases that ran the non-Step-3 program against the local-optimization default.

**The measurement system regresses.** Even when the implementation succeeds at the operating level, the underlying measurement frame, if unchanged, pulls the operation back toward the antipattern over time. Promotions are awarded against the old measures, new hires are oriented to the old measures, the quarterly close runs against the old measures, and within two to three years the old default has re-asserted itself. Barnard and Immelman's Holistic TOC Implementation chapter (Cox & Schleier 2010) documents cases including First Solar and Hawthorn Medical in which the measurement-set change had to be made structurally before the operational gains could hold.

## Correction

The corrective is not a single move; it is the discipline of running the Five Focusing Steps and refusing to skip Step 3. Five operational counter-moves track the trap's five expressions.

**Read every "improvement" against the constraint's exploitation rate.** Before any improvement project is approved, the question is whether it raises the constraint's effective throughput. If the project's effect on the constraint is zero or negative, the project is a local optimization candidate; the discipline is to redirect or refuse it. The TOC field's name for this audit move is [Continuous Improvement Auditing](continuous-improvement-auditing.md); the operating discipline of running it is described in Barnard's TOCICO paper and in Ch. 16 of the Handbook.

**Retire or recalibrate non-constraint local-efficiency measures.** Local efficiency at non-constraints is not a system property and should not be a non-constraint's primary measure. Replace it with buffer-status reports, throughput-per-constraint-hour for the constraint, and global T-I-OE for the system. The change is structural: rebuild the dashboards, retrain the line managers, and rewrite the bonus formulas. This is the change Goldratt names in the *Haystack Syndrome* and which the *TOC Handbook*'s measurement chapters (Budd Ch. 13, Smith and Herman Ch. 14, Barnard and Immelman Ch. 16) develop at length.

**Read non-constraint idle time as protective capacity, not waste.** A non-constraint that runs at 100% utilization has no capacity to catch up when the constraint stops or slows. Subordination requires non-constraints to hold protective capacity (the *TOCICO Dictionary*'s named capacity bucket alongside productive and excess) so they can run faster than the constraint when the constraint needs to be re-fed quickly. The corrective is to budget for and protect that capacity in the operating plan rather than auditing it as opportunity for headcount reduction.

**Direct investment at the constraint, after exploitation has been completed.** [Exploitation Without Elevation](exploitation-without-elevation.md) is the discipline that exploitation precedes elevation. Capital spent on a non-constraint produces a faster non-constraint and no throughput; capital spent on a constraint whose exploitation has not been completed funds the elevation of unused capacity. The correction is the order: exploit first, subordinate, and only then elevate, with the elevation directed at the constraint.

**Align incentives with system measures.** The bonus, the review, and the promotion need to track throughput at the system level (or at minimum the non-regressing protective capacity at non-constraints) rather than local utilization. Without this move, the implementation fights the alignment of careers against itself; the operating discipline cannot hold against the personnel system. The structural fix is to bring HR, finance, and the operating leadership into the same room and rewrite the relevant policies in one pass.

A practical signal that the correction is taking hold: visible idleness at non-constraints, named and defended by the people running them. An operator who can say "I'm idle because the buffer is full and the constraint can't yet absorb the next batch", and who isn't pressured to find make-work, is operating inside a subordinated system. The 95-to-100-percent figure Srikanth cites (non-constraints constitute the overwhelming majority of resources in any non-trivial system) is what makes the indicator useful: there should be many such operators, not few.

> **⚠️ Common misapplication:** The correction is *not* "stop optimizing." Plenty of optimization is appropriate: at the constraint, on the constraint's setup time, on the constraint's queue sequence, on the constraint's exploitation rate. The discipline of TOC reroutes optimization effort, it doesn't abandon optimization. Practitioners who hear "local optimization is the antipattern" and conclude that all efficiency work is suspect have over-corrected; the rate-matching principle of subordination uses non-constraint capacity hard during catch-ups and lightly when the constraint can't absorb the output. The signal is selectivity, not abstention.

The corrective is hard to hold across the accounting close. The cost-world reading of an idle non-constraint as inefficiency keeps reappearing in the CFO's quarterly review; the recommendation that follows ("reduce headcount, consolidate shifts, eliminate the idle hours") is the readout that breaks subordination and re-installs the antipattern. Holding the corrective requires [Throughput Accounting](throughput-accounting.md), the [Cost-World vs. Throughput-World Thinking](cost-throughput-thinking.md) frame, and the explicit naming of the cost-world recommendation as a misreading of the protective capacity the system needs to subordinate well.

## Sources

- Cox, *TOCICO Dictionary* (2nd ed., 2012), entries "local optimum," "non-constraint," "global measures," "protective capacity," and "excess capacity". These are the certification-grade definitions cross-checked here. The Dictionary is distributed through TOCICO membership rather than the trade book channel and does not carry an Open Library record.
- Goldratt and Cox, [*The Goal*](https://openlibrary.org/works/OL2463051W) (1984; revised 1986, 1992). The originating treatment is dramatized through Alex Rogo's plant. The epigraph is from Ch. 25, in dialogue between Jonah and Alex; the entire Bearington-plant subplot stages the dynamics in which local efficiencies trend upward while system performance does not.
- Goldratt, [*The Haystack Syndrome*](https://openlibrary.org/works/OL4455381W) (1990). Formalizes the T-I-OE global measures and contrasts them with the local measures the antipattern is built on; the chapters on the global-measure-versus-local-measure question are the operational counter-position the corrective rests on.
- Goldratt, [*What Is This Thing Called Theory of Constraints and How Should It Be Implemented?*](https://openlibrary.org/works/OL4186015W) (1990). Names the Five Focusing Steps as the operational discipline that prevents the antipattern by ordering identification and exploitation ahead of any improvement move at a non-constraint.
- Cox and Schleier, [*Theory of Constraints Handbook*](https://openlibrary.org/works/OL18346923W) (2010), Ch. 1 (Goldratt, "Introduction to TOC — My Perspective"), Ch. 8 (Srikanth, "DBR, Buffer Management, and VATI Flow Classification"), Ch. 13 (Budd, "Traditional Measures in Finance and Accounting"), and Ch. 16 (Barnard and Immelman, "Continuous Improvement and Auditing the Process"). The contemporary canonical treatment is distributed across the volume rather than collected in a single chapter. Srikanth names local efficiency as the principal antagonist of every DBR implementation he has worked; Budd explains the accounting mechanism through which the trap is rewarded; Barnard and Immelman document the measurement-set change as the structural move that prevents regression.
- Mabin and Balderstone, [*The World of the Theory of Constraints*](https://openlibrary.org/works/OL13537037W) (2000). The meta-review of more than 80 documented TOC implementations is the empirical basis for the claim that the antipattern, not the framework, accounts for most apparent implementation failures when cases run Step 2 alone or invert Steps 2 and 4.
- Dettmer, [*Goldratt's Theory of Constraints*](https://openlibrary.org/works/OL1827619W) (1997). Extended treatment of the local-optimum trap as it operates at policy and paradigm constraints; the Thinking Processes are positioned as the diagnostic toolkit for cases in which the antipattern is sustained by a belief or a rule rather than by a measurement.
- Smith, [*The Measurement Nightmare*](https://openlibrary.org/works/OL8447843W) (2000). Monograph treatment of the measurement frame the antipattern rides on, with the operational and reporting consequences of replacing local-efficiency measures worked at chapter length.

---

- [Next: Improving Non-Constraints](improving-non-constraints.md)
- [Previous: Strategic Constraint](strategic-constraint.md)