Improving Everything Improves Nothing: Goldratt and the Bottleneck

A packaging manufacturer, 320 people. At quarter close, the executive dashboard was green top to bottom: sales hit its proposal target, engineering cut design time by 18%, the cutting floor raised output per shift, procurement negotiated better terms with two suppliers. Everyone improved.

Average delivery time to the customer stayed exactly where it was: 27 days. It was 27 the previous quarter and it was still 27 after all that effort. The question nobody asked out loud in the meeting was the only one that mattered: if everyone improved, why didn't the customer notice anything?

01The idea: a system has exactly one constraint #

Eliyahu Goldratt, an Israeli physicist, published The Goal in 1984 in the unlikely form of a novel: a plant manager about to lose his factory discovers, with help from an old professor, that he'd been measuring the wrong thing. The book gave rise to the Theory of Constraints.

The argument is almost uncomfortably simple. Any system that turns input into output has, at any given moment, one step that limits the whole. That step is the constraint — the bottleneck. The capacity of the entire system is its capacity, not the average of all capacities.

The practical consequence is harsh: improving a step that isn't the bottleneck doesn't increase delivery. It increases inventory in front of the bottleneck, or idleness behind it. Local gain without global gain is cost dressed up as productivity.

Goldratt also laid out the five-step route: identify the constraint, decide how to exploit it fully, subordinate everything else to that decision, elevate its capacity, and — the step almost everyone forgets — start over, because once the constraint is resolved it moves somewhere else.

02Why this still holds outside the factory #

In 1984 the scene was an assembly line with idle machines and stacks of parts on the floor. Today, in the average mid-size Brazilian company, most work is administrative and the stacks are invisible: proposals waiting on margin approval, orders held for credit analysis, contracts in the legal queue, tickets waiting on the one person who knows how to configure that thing.

The physics is identical. The difference is that steel inventory is visible from the hallway and pending-decision inventory lives in somebody's inbox. Nobody trips over it.

There's a modern aggravating factor too. Since every area has its own metric and its own tool, each optimizes what it can measure. The sum of seven local optima is often one terrible global result — and that's exactly the picture of an all-green dashboard with a frozen delivery time.

03The cost of an ignored bottleneck #

Let's run the math at the scale of that 320-person mid-size Brazilian company (figures in Brazilian reais). The assumptions are explicit and conservative estimates, not field measurement:

Want to see how this looks inside a real operation? Explore the platform.

The math: at R$ 90 million in annual revenue, each day cut from the cycle pulls forward roughly R$ 250 mil of revenue in transit, and four days freed up capacity worth about 1.5% more orders per year with the same headcount. The exact number matters less than the proportion: the gain came from changing one step, not seven.

And the reverse is math too. The improvement projects run outside the bottleneck that quarter consumed people's time, consulting, and systems — and produced, on the metric the customer feels, zero.

04Where this gets stuck in practice #

1. Nobody knows where the queue is. Since each step lives in a different tool, there's no single place showing how long an order waited in each pair of hands. Without that, the bottleneck becomes opinion — and every area's opinion points at the area next door.
2. The bottleneck is a person, and that's awkward. In many mid-size companies the constraint has a name: the director who approves everything, the sole specialist, the partner who signs. Treating that as a system problem rather than a character flaw is half the solution.
3. Utilization gets confused with results. Keeping everyone 100% busy looks like good management and is actually the recipe for inflating queues. Outside the bottleneck, some slack is a design choice, not waste.
4. The constraint moves and nobody re-checks. Once legal is fixed, the queue migrates to implementation. Whoever doesn't redo the diagnosis keeps optimizing last year's bottleneck.

05What data-driven management answers #

Identifying the constraint is, at bottom, a recording problem. If every step of the work has an entry, an owner, an exit, and a timestamp, the queue draws itself and the bottleneck stops being a political argument. It's the same reasoning behind reading metrics correctly — worth reading why variation isn't error before reacting to one isolated queue spike.

Two common traps show up once you start looking at queues. The first is mistaking a pile of open work for capacity — the arithmetic is in why more projects deliver less. The second is assuming the bottleneck gets solved by automating it: sometimes the step shouldn't exist at all, as argued in the piece on automating a bad process.

In practice, that's what the processes and operations modules in the platform record: every step transition becomes data, with date and time, without anyone filling in a spreadsheet. From there the queue map is a consequence, not a project. To see how this behaves by area, the solutions by operation show the most common flows — and if you'd rather look at your own case, a 30-minute conversation beats a generic diagnosis.

The green dashboard wasn't lying. It was measuring seven things that weren't the one holding the company back.