Sylvics Apply for the beta
📖 Goldratt on bottlenecksLiterature note
LITERATURE NOTE

🏭Goldratt on bottlenecks

🚦 Status🟢 Processed
🏷️ Topics
BusinessProductivity

🧵SYNTHESIS

Paraphrase, not quotation: roughly the first two thirds of 📕 The Goal.

A plant manager, Alex Rogo, has three months to save his factory. A physicist called Jonah asks him questions and refuses to answer them. The idea underneath: a plant is a chain of dependent steps, each with its own random variation, and delays pile up while gains do not. So the output of the whole chain is set by its slowest resource, the bottleneck.

Two consequences follow. An hour lost at the bottleneck is an hour lost for the entire plant, so its true cost is the hourly revenue of the whole business, not of one machine. And an hour saved anywhere else buys nothing, because that capacity was already waiting.

He finds his two bottlenecks by walking the floor and looking for the biggest pile of unfinished work. I did the same with our board. The pile sits in front of QA.

✨KEY IDEAS

On the scout hike, the line can only move as fast as its slowest walker, and the gaps between boys grow because slowness accumulates and speed does not.
The Herbie chapters. Clumsy fiction, perfect diagram. Put the slow walker at the front and share out his pack.
The bottleneck is found by looking for where the work in progress piles up.
Not by asking who is busiest. Everyone is busiest.
An hour lost at the bottleneck is an hour lost for the whole system.
So a bottleneck that breaks for lunch costs the plant, not the department.
Inspect quality before the bottleneck, so that it never spends time on parts that will be scrapped.
We send QA builds that design has not looked at. This is that.
Once a bottleneck is found, everything else is paced to it.
The middle step of 🧭 Five focusing steps, and the one that hurts.

📝READING NOTES

The book, briefly

It is a business novel from 1984 and it reads like one: a marriage in trouble, a boss with a deadline, a mentor who appears in airports. I read it because Priya has said bottleneck in every planning meeting since November and I wanted to know whether we meant the same thing. We did not. I meant slow. She meant the one resource that sets the rate.

The argument in four moves

  1. The goal. A plant exists to make money, and Jonah gives three measures for that: what it sells, what it has tied up in stock, and what it spends turning one into the other. Local efficiency is not one of them.
  2. Dependent events plus variation. The dice game with matchsticks shows it. Every station averages the same number, and still the last bowl gets fewer than average, because a bad roll upstream starves you and a good one cannot be banked.
  3. The bottleneck. Any resource whose capacity is at or below the demand placed on it. See 🧩 Bottleneck.
  4. What to do. Keep it running, feed it only good work, move work off it, and make everyone else follow its pace.

Finding ours

Stage Waiting on Friday
Design almost nothing
Build a normal week's worth
QA the largest column on the board, and growing
Release empty

Nadia has one test environment and two people. In January I asked Tomasz for three more empty states, because design had capacity and I dislike seeing idle capacity. That was feeding the front of the hike.

What I do not accept from the book

Jonah's plant has stable bottlenecks: a machine and a furnace. In software the constraint moves, sometimes within a sprint, and a person is not a furnace. Calling Nadia's team the bottleneck in a meeting would be accurate and unkind. The constraint is the single environment, which is a thing, and can be discussed as one.

What it produced

💎 A bottleneck sets the pace for everything upstream of it, written today.

Added 19 March: in the retro Priya drew this table from memory. Nobody argued with it. The second note from the book is 📖 Goldratt on local optima.

To Try At Work

  • Done: Count work waiting at each stage every Friday for a month
  • Done: Stop starting design work that QA cannot reach this sprint
  • Done: Ask Lucía what a second test environment would cost
  • Still open: Move the design review in front of QA, not after it

🗣️IN SHORT

  • In : The Goal, the chapters from the scout hike to the first results at the plant
  • Goldratt argues that : the output of a chain of dependent steps is fixed by its bottleneck, so time there is worth the output of the whole system
  • by stating that : variation accumulates along a chain while gains cannot be stored, and that capacity added anywhere except the constraint produces only inventory