Guide 04 / Topic

Operations automation

The premise

The best automation shortens the distance between a signal and a responsible action.

A living guide \u00b7 last tended 26 August 2026

Automation is valuable when it removes waiting, repetition and preventable inconsistency. It is dangerous when it hides uncertainty or moves a clinical, ethical or commercial decision away from the person accountable for it.

The operating question is not “Can this be automated?” It is “Which part should be automated, under what control, and what must remain visible to a human?”

Four questions
  • Where does work wait?
  • Which rule is stable?
  • What can go wrong quietly?
  • Who can stop the system?

01

Automate the handoff before the judgement

Many processes do not need an intelligent decision first. They need information to arrive, be checked for completeness, reach the right owner and create a visible follow-up.

Automating those steps can remove substantial friction while keeping the accountable judgement with a person. It also creates cleaner evidence for deciding whether deeper automation is safe.

02

Classify the risk of a wrong action

A reversible reminder and an irreversible clinical or financial action should not share the same control model.

Before choosing technology, describe the worst plausible wrong action, how quickly it would be detected and whether it can be reversed. The answer determines approval, monitoring and human review.

Assist

Prepare or summarise information for a person who makes the decision.

Recommend

Suggest a next action while exposing the reason and uncertainty.

Execute

Complete a bounded action with clear rules, logs and a recovery path.

Escalate

Stop and transfer when confidence, policy or context leaves the safe boundary.

03

Pilot against a real baseline

A demo shows that a happy path is possible. A pilot should show whether the operating system improves.

Choose a bounded workflow, record the current delay or failure pattern, define the acceptable error boundary and name the owner of exceptions. Compare like with like. Do not let novelty become the success measure.

04

Keep the system legible

People need to know when automation acted, what information it used and what to do when the result looks wrong.

A system that saves time but makes responsibility invisible creates a new form of operational debt. Logs, clear status, human override and named exception ownership are part of the product, not later controls.

The operating model

01

Bounded

Start with a narrow workflow and a clear stop condition.

02

Legible

Make system actions, reasons and status visible.

03

Reversible

Prefer actions that can be corrected while confidence is still forming.

04

Owned

Keep a named person accountable for exceptions and outcomes.

Next operating lensCompetitive intelligence