← Thinking

Principle 03

Automate the mess and you get the mess at scale.

Speed applied to a broken process is not efficiency. It is a faster way to be wrong.

01What I mean

Every process encodes decisions, some deliberate, most accidental. Automation freezes those decisions and runs them thousands of times a day. Anything left ambiguous becomes ambiguity at industrial volume, and the people downstream absorb it.

02Why it matters

The damage is delayed and diffuse. Nobody notices the automated mess for a quarter, and by then it has generated data, downstream dependencies, and a constituency. Cleaning up before automating costs weeks; cleaning up after costs quarters.

03What it looks like in practice
  • 01Map the process as it truly runs, including the workarounds, before designing the automated version.
  • 02Fix or explicitly accept every ambiguity you find. Undocumented judgment does not survive automation.
  • 03Simplify first, automate second. The simplification is often the entire win.
  • 04Instrument the source, not just the output, so a rising error rate points at a cause.
04From my work

Help content had drifted from the product. Automating answers on top of drifted content would have produced confidently wrong answers at global scale, worse than a queue. The content audit came before the agent.

Turning Consumer Support Into a Product

Contact

Let's get into it.

Ambiguous problem, AI adoption that stalled, an operating model that stopped scaling, a support function that should be a product. That is the conversation I want.