The Haystack Syndrome

Three questions, three measures. Imagine buying a money-making machine. You would ask how fast it generates money, how much money it traps inside, and how much it costs to run. Goldratt formalizes these as:
1. Throughput: the rate the system generates money through sales (selling price minus what you paid vendors for what went into the product).
2. Inventory: all money invested in things you intend to sell, valued at raw-material price only, with no added value.
3. Operating Expense: all money spent turning inventory into throughput.

Never optimize one alone. Fire everyone and operating expense plummets, but throughput dies too. Net profit is throughput minus operating expense; return on investment divides that by inventory. Because the three interlock, only their relationships, not any single number, can judge a local action's impact on the goal.

A brilliant solution that killed itself. When invented, cost accounting was genius. Direct labor was paid per piece and dwarfed overhead tenfold, so allocating overhead by labor gave a workable way to split a company product by product. But technology inverted the ratios: overhead now runs five to eight times direct labor, and workers are paid for showing up, not per piece. Both founding assumptions collapsed.

Product cost, product profit, product margin: delete them. You never paid money to a product, only to categories (workers, banks, utilities). Goldratt cites a nine-billion-dollar company whose CEO outsourced every part cheaper to buy outside; each round made surviving parts absorb more overhead, triggering more cuts, until the firm shrank to a third and changed its name. He calls activity-based costing a futile effort to rescue a corpse.

The P and Q quiz. Goldratt gave over 10,000 managers a puzzle with two products. P sells for $90 (materials $45), Q for $100 (materials $40). Every cost lens, selling price, material, or labor effort, flags Q as more profitable. But one machine, resource B, lacks capacity. The real question is throughput per constraint minute. P earns $45 over 15 minutes ($3/min); Q earns $60 over 30 minutes ($2/min). Push the cost-favored Q first and the company loses $300 a week. Push the despised P first and it profits $300.

Only one in a hundred solved it. Managers obey training, not intuition. The lesson bites hardest for sales: paying higher commission on the higher-margin product drives salespeople to push exactly the wrong item.

The focusing engine. Because every chain has exactly one weakest link, Goldratt distills management into five steps:
1. Identify the system's constraint (the thing you lack enough of).
2. Decide how to exploit it, squeeze maximum value from it, no waste.
3. Subordinate everything else to that decision (non-constraints should feed the constraint, no more, no less).
4. Elevate the constraint (add capacity) only after steps two and three.
5. If a constraint breaks, return to step one.

The warning clause. Step five carries a caution: do not let inertia create a new constraint. He notes he almost never finds true physical constraints in companies; he finds policy constraints, rules that made sense decades ago and outlived their reasons.

Two worlds of thinking. In the cost world, every worker, machine, and scrap pile is a leak of money, so almost everything matters. The Pareto 20-80 rule applies because variables are treated as independent. But a company is a chain of dependent variables synchronized to produce a sale. In this throughput world, the rule shifts dramatically: a tiny fraction of variables, the constraints, govern nearly all output.

The paradigm switch. Goldratt frames this as the biggest shift a scientist can make, from independent to dependent variables. Managers stuck in the cost world describe themselves holding dozens of ping-pong balls underwater, spread thin across equally urgent fires. Recognizing that only the constraint truly limits the goal collapses that chaos into a handful of decisive leverage points, which is why TOC, JIT, and TQM all preach that local optima do not sum to a global optimum.

Measurement is destiny. Goldratt's most quoted line warns that illogical metrics produce illogical behavior, and complaining about the behavior is pointless. A plant manager who slashes inventory (a stated goal) gets punished because accounting reclassifies inventory as an asset, so the cut shows as a loss. Measure by efficiency and workers overproduce non-constraints to look busy.

A better local yardstick: dollar-days. For deviations he proposes throughput-dollar-days: multiply an order's selling value by the number of days it is late, and assign that penalty to whichever department currently holds the stuck task. Late work becomes a hot potato nobody wants, triggering self-expediting. Sloppy work that fails downstream gets its dollar-days reassigned back to the source, driving quality at the source without an investigation bureau.

Murphy is a permanent resident. Disruptions (breakdowns, absentees, scrap, late vendors) cannot be eliminated, only managed. The traditional response buffers every task with stock; TQM says attack every defect. Goldratt blends both: accept Murphy exists today while fighting to reduce it, and place protection only where loss of throughput is at stake.

Drum-buffer-rope. The constraint sets the pace (the drum). Material is released a time buffer ahead of when the constraint needs it (the rope), so disruptions upstream get absorbed without starving the constraint. Buffers are measured in time, not inventory, because the composition of work matters, not mere busyness. Watching for holes in the buffer (tasks that should have arrived but did not) tells you exactly which resources to expedite and which processes to fix first, generating a live Pareto list for improvement.

Powerful solutions self-destruct. Goldratt argues there are no ultimate solutions, only powerful ones, and the more powerful a solution, the faster it can render itself obsolete by changing the environment that demanded it. Cost accounting is his prime example. When constraints break, the rules built around them usually survive, quietly hardening into policy constraints.

Dig for the buried why. He almost never finds true bottlenecks or vendor constraints in real companies; he finds marketing, purchasing, and production policies whose original justification vanished thirty years ago. He also skewers technical inertia: MRP systems split product data into bills of material and routings only because 1960s magnetic tape forced it, yet the split persisted long after disks made it pointless, bloating computation a thousandfold. Progress demands revisiting inherited assumptions as ruthlessly as you attack visible waste.