Run the Plant by Its Constraint

The Core Insight

Lou the controller guessed the parts queued at heat treat held ten to twenty thousand dollars of material. Jonah counted throughput instead: about a thousand parts, each blocking a 1,000 dollar product at final assembly, one million dollars waiting to ship. Both men priced the same pile. The gap between their numbers is the whole book.

The Goal is Eliyahu Goldratt's business novel, and the story is the vehicle, never the point. Plant manager Alex Rogo gets three months to save a factory that looks healthy on every local report. Robots raised one department's efficiency thirty-six percent, plant efficiency runs above ninety percent, and orders ship seven weeks late. Jonah, the physicist who plays Socrates, asks whether the robots let the plant sell one more product, employ one fewer person, or carry less inventory. Three answers of no mean the robots raised nothing.

Most operations thinking treats a busy resource as good and an idle one as waste. It prices each work center in isolation and pushes everything to run flat out. Goldratt argues one constraint sets the output of the entire system, and idle time on non-constraints is part of the design. A plant where everyone works all the time is very inefficient, and the whole machinery follows from that inversion.

The Framework

The goal comes first, because productivity means nothing without one. The goal of a manufacturing organization is to make money, and everything else is a means. The finance version moves three dials together: net profit, return on investment, and cash flow. Any one alone can be gamed.

Jonah then hands over three measurements a shift supervisor can run a plant on.

• Throughput is the rate the system generates money through sales, and product built but unsold is not throughput.
• Inventory is all the money the system invests in things it intends to sell.
• Operational expense is all the money the system spends turning inventory into throughput.

The wording breaks with cost convention on purpose. Inventory carries no value-added labor, so no dollar hides between investment and expense, and all labor time, busy or idle, is operational expense. The goal restated: raise throughput while cutting both of the others.

The whole method then compresses to five focusing steps.

• Identify the system's constraint.
• Decide how to exploit the constraint.
• Subordinate everything else to that decision.
• Raise the constraint's capacity.
• If a constraint breaks, go back to step one, and do not let inertia become the new constraint.

The inertia clause is earned inside the plot. Priority tags and make-to-stock padding both outlive the machine constraint they served, and the padding hides twenty percent of bottleneck capacity behind fictitious load. Lou's timing on it: inertia forms in less than one month. The word constraint itself replaces bottleneck once the release policy and then the market each take a turn as the limit.

Key Ideas

Fluctuations Accumulate Down a Chain

Any pipeline has two properties. Dependent events mean one step must finish before the next starts. Statistical fluctuations mean each step's pace varies around an average. Intuition promises the variation cancels out. Down a chain it accumulates. A step can never pass along more than the step before it delivered, so the upside is capped and slowness compounds.

Goldratt proves it with dice at a scout camp. Five bowls sit in a line and matches move down it by die roll, capped at what each bowl holds. Every station has identical capacity, an average of 3.5 a turn, and demand equals that average. Ten rounds promise thirty-five matches through the line. The run delivers twenty, with inventory lurching through the bowls in waves. Nothing broke and nobody slacked. The structure alone did it.

The same experiment later runs on the floor with people and a deadline. Pete's crew works to a quota of twenty-five pieces an hour, feeding a robot welder that does exactly twenty-five. The crew's hourly log reads 19, 21, 28, 32, so it hits 100 by shift end and celebrates beating the robot. The robot, capped at twenty-five, welds what arrives and reaches ninety by the 5:00 truck. The shortfall is ten, exactly the crew's largest cumulative deficit during the shift. The maximum deviation of a preceding step becomes the starting point of the next, and a local quota met on average produced zero shipped product.

A Balanced Plant Runs Itself Broke

A balanced plant matches the capacity of every resource exactly to market demand, and managers across manufacturing chase it as the ideal. Goldratt calls it the closest route to bankruptcy. Trim every resource to demand and fluctuation does the rest: throughput drops while inventory and its carrying cost climb.

The replacement is the first of nine rules Jonah names: balance the flow of product with market demand rather than balancing capacity. The two-resource arithmetic makes it concrete. When demand takes all 600 hours of constraint X, a non-constraint Y feeding it has 450 usable hours out of its own 600. Running Y past that point produces inventory and nothing else. Turning a resource on and getting value from it are different acts. A non-constraint's right level of use is set elsewhere in the system, never by its own potential.

The warehouse audits the old belief. Two-thirds of finished goods are all-non-constraint products built to keep efficiency numbers up. About 1,500 obsolete units built with bottleneck hours sell ten a month at best. Excess inventory is congealed excess capacity.

Price the Constraint Hour at System Expense

A bottleneck is any resource whose capacity is equal to or less than the demand placed on it. Finding one needs no clean data. The routing files listed milling machines sold a year ago, so the team walked to the biggest pile of work-in-process instead. Two stood behind piles: the NCX-10 machine and the heat-treat furnaces, both of them policy artifacts.

The NCX-10 replaced ten older machines with one. Fourteen minutes of work across three machine types became ten minutes on a single machine, so cost per part fell while aggregate capacity fell too. The furnaces ran half empty because expediters forced small urgent loads through them. Efficiency bought the first bottleneck, and expediting starved the second.

The standard rates priced an NCX-10 hour at 32.50 dollars and a heat-treat hour at 21 dollars, as if each work center existed in isolation. Bottleneck output is plant output, so an hour lost at a bottleneck is an hour lost for the entire system. The true rate is total operating expense divided by bottleneck hours: 1.6 million dollars a month over 585 hours is 2,735 dollars an hour. Lou corrects the coefficient overnight, since only 80 percent of products cross a bottleneck, and lands on 2,188 dollars. The paired rule follows: an hour saved at a non-bottleneck is a mirage.

Exploitation is a list of recovered hours, none of them bought. Inspection moves in front of the bottlenecks, because about five percent of parts entering the NCX-10 and seven percent entering heat treat arrive already defective. Dedicated crews cover lunches and setups, because the machine sat idle twenty to forty minutes between jobs. Every batch gets a tag, red for bottleneck-routed work and green for the rest, and red runs first.

Capacity then comes out of hiding. Three retired machines return for the cost of a truck rental and add eighteen percent to output of NCX-10-type parts. A third-shift foreman's batch sorting adds about ten percent through the furnaces. And a fifth of the heat-treat load never needed treating. An efficiency push five years earlier tripled the cutting bite from one millimeter to three, and the brittle metal it produced forced heat treatment. Reversing the cut on machines with spare capacity deletes the load. The team's summary: reduce the efficiency of some operations and make the entire plant more productive.

Halve the Batch and Lead Time Halves

Material spends its life in four states: setup, process, queue, and wait. Setup and process are small. Queue, waiting for a busy resource, and wait, waiting for a sibling part, dominate total time, and the bottlenecks dictate both. Halving batch sizes on non-constraints halves queue and wait, which roughly halves lead time. In the story, lead times fall from three or four months to about two, and the second halving reaches about four weeks.

Cost accounting reports the move as a disaster. A batch of 100 parts with two hours of setup and five minutes of process carries 6.2 labor minutes per part. Halve the batch and the figure reads 7.4, and with burden near three times labor the paper cost jumps. Real operating expense never moved, because the extra setups on non-constraints only convert idle time. The same expense now spreads over more product sold.

The batch logic then wins a sale under fire. Burnside wants 1,000 Model 12 units in two weeks, with about fifty in stock and the bottlenecks producing parts for about one hundred a day. Alex halves batches again and counters with 250 units a week for four weeks, first lot in two weeks. Burnside takes it, prefers the staggered shipments, and the order runs past a million dollars.

The Drum Times Every Release

Drum-buffer-rope arrives as a hiking problem. The troop's speed is the speed of Herbie, the slowest scout, and no eager walker up front changes that. Putting Herbie first stops the line spreading, and emptying his overloaded pack doubles the troop's speed. The lasting fix is a drum and a rope. Herbie sets the beat, and a rope to the lead walker caps how far the line spreads.

The plant version schedules the bottlenecks first, then works backwards. Material takes about two weeks to reach them, so releases hold roughly a three-day buffer, accurate within about a day. The same schedule fixes final-assembly dates and times every other release, so the bottlenecks determine the release of all materials. Releasing faster than the drum had already manufactured a false constraint: milling machines drowned in red-tagged work while green-tagged parts sat three weeks.

The system still relapses once, and the failure teaches the last rule. New orders eat the plant's spare capacity, so after an upstream breakdown the non-constraints cannot refill the buffers in time. Twelve work centers slide into overtime and orders start slipping. Non-constraints must hold protective spare capacity above the drum's pace. More buffer means less spare capacity needed, and the trade runs both ways.

The Last Constraint Is a Policy

The scoreboard settles the argument inside one quarter. Throughput doubles, inventories fall to about forty percent of their level three months earlier, and the bottom line grows seventeen percent against a promised fifteen. A division audit then recomputes the month at 12.8 percent under standard cost policy, because extra setups raised reported product cost. Lou later finds the deeper distortion. Valuing inventory as an asset booked the value-added drawdown as a net loss. Corrected, net profit ran above twenty percent in each of the last three months.

The two verdicts on the same plant name the split. The cost world ranks operating expense first and weighs every link of the chain, so every local saving counts. The throughput world ranks throughput first, then inventory, then operating expense, and measures the chain by its weakest link. The plant's own data picks a side: throughput and inventory moved by tens of percent while operating expense moved under two.

Pricing follows the same split. A French buyer offers 701 dollars for a Model 12 that lists at 992 dollars, below the official product cost. Materials run 334.07 dollars and the plant holds spare capacity, so each unit contributes 366.93 dollars after freight. The deal trades committed yearly volume for delivery inside three weeks, where the European norm runs eight to twelve. The contribution to plant net profit reaches seven digits, and product cost nearly killed it.

At division level the pattern generalizes. Excess capacity sits everywhere, so the constraints are policies. Inventory-profit incentives swelled divisional finished goods by about seventy days. Launch rules held new models back because launching forces obsolescence write-offs. For a policy constraint, exploit and subordinate mean nothing, and replacement is the only move. The manager's job compresses to three questions: what to change, what to change to, how to cause the change.

Practical Applications

Protect the constraint before adding anything. Inspect work before it reaches the constraint, cover its breaks, and move the best people to it. Route around it wherever the work does not need it, the way the plant revived retired machines and reversed the deep cut.

Subordinate the release of work to the constraint's pace. Stop starting work to keep people busy, hold a small time buffer in front of the constraint, and let every other utilization number fall. Say in advance that the numbers will fall, or the first efficiency report reverses the change.

Halve the batch wherever work moves in bundles: release trains, review queues, sprint scopes, invoice runs. Lead time falls with the batch, and a per-unit cost report that worsens is reading the wrong world.

Keep a reserve of spare capacity on non-constraints and size it against your buffers. After every disruption the rest of the system must outrun the constraint to refill its buffer. A pipeline sold to 100 percent load loses that ability first.

When a constraint breaks, rerun the five steps and hunt the leftover rules on purpose. Every tag, quota, and buffer sized for the old constraint is inertia forming, and the plant grew its inertia in under a month.

Who This Is For

Anyone who runs a pipeline gets the most from it: an engineering team, a sales funnel, a support queue. The mapping is direct wherever work moves through dependent stages at uneven speeds.

Skip it if you already run constraint-based operations, because the novel spends its length deriving what a practitioner holds on an index card. The marriage subplot carries no machinery at all.

The evidence deserves its label. The plant, the dice, and every number are constructions built to demonstrate the mechanism, so nothing in the book is field data. The appendix's real company figures are self-reported and unaudited. The setting is a factory, and the transfer to knowledge work is a claim the reader must test.

The Decision

The test for this week is the Herbie hunt. List the stages your work passes through, from request to cash, and find the deepest queue. That stage is your constraint, whatever the dashboard says.

Then price its hour. Divide monthly operating expense by the constraint's working hours and write the number down. Audit next week's calendar against it: every meeting, rework loop, and stray project sitting on the constraint now carries that price. Spend the hour like the most expensive one in the company, because it is.