Plant Walk

Find the constraint — three tells

  1. Find the biggest pile. Work in progress stacks up in front of the constraint. Behind it the line runs starved. You can see this before anyone gives you a number.
  2. Watch the people, not the machines. The bottleneck never waits for work and never gets blocked. Everyone else waits sometimes.
  3. Then time it. Five to ten cycles at your two or three candidates. Fifteen minutes, and a guess becomes a number.
Then ask: "If this station ran 20% faster, could you sell the extra output?" If no, capacity is not the problem and the money is in cost, quality or lead time instead.

Lowest hanging fruit, in order

1Nobody can see the problem

An hour-by-hour board at the line: target, actual, and a "why" column filled in by the operator. Costs a whiteboard and changes behaviour inside one shift, because a gap written down at 09:00 gets asked about at 10:00.

It is first on the list because it is free, and because it produces the data every other improvement needs.

See it in 5 min: no board anywhere, or a board showing last month.

2The bottleneck is not protected

The constraint sets the output of the whole plant, so an hour lost there is an hour lost everywhere. Yet it usually stops for lunch with everyone else, starves when the machine before it hiccups, and sometimes runs parts that are not even due.

Fixes cost nothing: staggered breaks, a small buffer in front of it, and the quality check moved before it so it never processes a part that will be scrapped later.

See it in 5 min: does the line go quiet at lunch? Is there a buffer in front of the slowest station, or does it sit empty and wait?

3Changeovers eat the week

In a small plant with many products this is usually the biggest single number in the building. The first pass needs no money: everything that can be prepared while the machine is still running must happen then, not after it stops. Tools staged, program loaded, material kitted, next job at the line.

Typical first result is 30 to 50 percent off, in weeks.

See it in 5 min: changeovers per week × minutes each. Over 10% of running time means start here. Ask what the operator does in the first ten minutes of a changeover. If the answer is "looks for things", you have it.

4The defect everyone stopped noticing

Every plant has one failure so familiar that a rework bench was built for it and nobody calls it a problem any more. It is permanent cost dressed up as normal work.

See it in 5 min: the red bin, the touch-up table, the one person permanently fixing. Ask: "What do you always have to fix?"

5Too much work in progress

Cutting WIP shortens lead time in direct proportion, and it improves quality as a side effect, because a defect is found minutes after it is made instead of days. It costs nothing, it is a decision.

See it in 5 min: count the piles. The Result tab turns that count into a lead time.

6An operator watching a machine work

If the automatic cycle is long and the hands-on part is short, that person is paid to stand. One operator can often run two machines, or do the inspection or kitting during the auto cycle.

See it in 5 min: somebody standing and watching. Enter cycle and manual time in Capture and the Result tab flags it.

7Motion at the busiest station

Three seconds of reaching, turning or searching, a thousand times a day, is an hour a day. Only worth attacking at the station with the highest repetition.

See it in 5 min: watch one operator for two full cycles and count how often they turn away from the part.

The seven wastes — what they look like

Waiting
Idle people, idle machines, a station with nothing to do
Inventory
Piles between stations, a full warehouse, old material
Transport
Parts carried across the hall and back, forklift trips
Motion
Reaching, bending, walking to fetch, searching for tools
Defects
Red bins, rework benches, retest loops
Overproduction
Making more or earlier than needed, big batches
Overprocessing
Tolerances and checks tighter than the customer asked for
The three you can literally photograph are Inventory, Waiting and Defects. Those are the ones to point at.

Following one part

Imagine a camera taped to one part from raw material to shipping. Almost all of the film is the part lying still. In most plants the working time is one or two percent of the time the part spends in the building.

You can prove that in two minutes without following anything: count the parts on the line, ask the output per shift, and the Result tab does the rest. That ratio is usually the most uncomfortable number of the day, and it moves the conversation away from buying a faster machine towards stopping parts from waiting, which is free.

The five that matter

Flow and process

Data and systems

Commercial

Before any file changes hands: "If you give me data it goes through an AI tool, so it leaves your building. We can replace customer names and part numbers first, the analysis stays the same. Or you say no and we do it differently."
Saved

Visit

Stations

Cycle = seconds between one finished part and the next at that station. Manual = seconds the operator actually has hands on it. WIP = parts waiting in front of the station right now, counted not estimated.

Shift

Notes

What if we relieve the bottleneck

cut by 20%

A snapshot from one walk, not a study. Hand-timed cycles over a few parts scatter, and one shift is not a week. Good enough to point at the constraint and size the prize. Not good enough to buy a machine on. Say that out loud.

Everything as text

Copy this and paste it to Claude, into Telegram, or into a mail to yourself. It contains the inputs and the results.

Everything is stored only on this phone, in this browser. Nothing is sent anywhere. Clearing browser data removes it, so copy the text out before you finish.