The process body of knowledge, admin-first

Before you automate anything, you need the older discipline that automation forgets: process improvement. Decades of it (Lean, the Theory of Constraints, Six Sigma, business process re-engineering, total quality management, and business process management) already explain how work flows, where it clogs, and what a fix is worth. The catch is that this canon was written for the factory floor, and most businesses today run on documents, emails, and approvals. Read admin-first, it still tells you exactly where to look before an agent is ever considered.

Why does pre-AI process knowledge still matter?

An agent applied to a broken process simply makes the mess faster. The questions that decide whether work is worth automating are older than AI and unchanged by it: where is the constraint, what does a unit of work cost fully loaded, how much of the effort is rework, and what breaks if a step is removed. Answer those first and the automation decision becomes obvious. Skip them and you are guessing.

What does “admin-first” mean?

The classic tools were built to watch a production line. In an office the line is invisible: a document waits in an inbox, an approval sits with one person, a number is re-keyed between two systems that do not talk. The same ideas translate directly. Takt time becomes the pace a service must keep. Overall equipment effectiveness becomes the real utilisation of a team. Value-stream mapping becomes a map of how information moves. The waste called muda becomes the office wastes: waiting, rework, over-processing, and the swivel-chair shuffle between screens.

Where is the constraint?

Throughput is governed by one bottleneck, and in an admin business it is almost always a person or an approval, not a machine. The partner who must review everything, the single licensed signatory, the one inbox that everything routes through. Relief at the constraint multiplies across the whole business. Relief anywhere else produces no gain at all, which is why the first thing a diagnostic does is find it.

What comes before automation?

Automation is the last rung of the ladder, not the first. Eliminate the step if you can. Simplify and standardise it. Combine or re-sequence. Error-proof it. Only then automate, and only then re-engineer. A credible plan reaches for an agent fifth, not first, and says so.