Experiment

Find the Constraint

Automating the greatest number of tasks doesn't necessarily produce the greatest improvement.

Automate any step of a pipeline and watch what happens to the output. The important question is what's currently constraining the outcome.

Systems thinking · Throughput · Automation

The short version.

A pipeline moves only as fast as its slowest step. Speed up any other step and the work just piles up in front of the bottleneck.

So the first question isn't what can be automated. It's what's limiting the outcome right now. Fix that, measure, and the constraint moves somewhere else. Then find the next one.

Demo

Automate a step.

Each click on a step doubles what it can handle. The numbers are illustrative.

A recruiting pipeline

Illustrative numbers
  1. Find candidates
  2. Make contact
  3. Screen
  4. Schedule
  5. Hand off

14candidates reach a recruiter per day

Each click on a step doubles what it can handle.

Find the constraint. Intervene. Measure. Then find the next one.

What it leaves out.

A small model on purpose. Here's what it shows, and what it simplifies.

  • Work piling upEach step shows how much waits in front of it when the step before it is faster.
  • LimitsA step can be doubled three times, up to 120 a day, because no real step scales without limit.
  • What it simplifiesIt ignores variability, rework and the delays between steps. Real pipelines have all three, which usually makes the constraint harder to find, not easier.

Built with plain JavaScript in the browser.

Build note.

A small build to answer one question.

The question
Does automating more steps of a process produce more output?
What it does
A five-step recruiting pipeline with an illustrative daily capacity at each step. Click a step to double its capacity and the output updates, along with a tally of what you've automated.
What it shows
Output only moves when you raise the slowest step. Raise it far enough and a different step becomes the constraint. Automating anything else adds work without adding output.
Where it's used
The same idea drives Dealer Flywheel: find where performance is actually lost before building anything.