The right question is not ‘which AI tool should we buy?’ It is ‘which repetitive task has a clear input, predictable rules and a verifiable outcome?’ If that cannot be described, automation will simply scale the ambiguity.

1. Is the process genuinely repetitive?

A strong candidate occurs frequently, follows similar steps and consumes time without requiring a new strategic decision each time. Examples include classifying incoming requests, transferring data from standardised forms, producing a first report draft, or sending an internal reminder when information is missing.

If the steps constantly change, inputs are unstable or the work relies mainly on experience-based judgement, redesign must come first. We do not automate chaos. We map it.

2. Can we describe the correct outcome?

Every workflow needs a clear success criterion. Which fields must be completed? Which errors are unacceptable? When can an output pass automatically, and when must it be reviewed? How will time saved be measured without sacrificing quality?

The answer need not be complex. It may be a sample of a correct output, a small rule table and three review thresholds. Without these, there is no way to know whether automation improved the process or merely concealed its errors.

3. Where should automation stop?

Exceptions are the critical point. Define who takes responsibility when information is missing, rules conflict, personal or confidential data are involved, or the output may materially affect a person.

Human oversight is not a decorative approval at the end. It belongs where a real decision, risk or need for justification exists. In many cases, semi-automation is the better answer: the system prepares, classifies and recommends, while the responsible person approves or corrects.

A small audit before the first pilot

  • Document the current workflow from input to final output.
  • Measure time, delays and recurring errors across five to ten real cases.
  • Separate stable steps from decisions and exceptions.
  • Define a limited pilot with one process owner and clear success criteria.
  • Compare before and after: time, quality, number of corrections and user experience.

When real value exists

Automation creates value when it reduces pointless repetition, leaves a clear audit trail and gives people more time for judgement, communication and meaningful work. If it merely adds another tool, another subscription and another screen, the result is digital bureaucracy in shinier packaging.

Official and selected sources

Links were reviewed on 27 August 2026. The official text always prevails for authoritative interpretation, together with appropriate professional advice where required.