Automating business processes: which workflow should come first?

Many business processes can be automated, but not all at once. A useful first candidate happens repeatedly, has recognisable inputs and ends in a clear action. Examples include registering an enquiry, flagging a missing document or passing on an approved order. Making an inconsistent process run faster does not make it reliable.

Collect three candidates from daily work

Ask employees where they retype information, wait for input or repeat the same check. Have them demonstrate a recent example. Record volume, active handling time and exceptions. Separate waiting from work: waiting three days for approval is not three days of labour.

Consider the consequences of mistakes as well. An internal reminder needs different safeguards from a payment or a firm customer commitment. Start with an outcome you can check and mistakes you can recover from.

Compare value with feasibility

• Does the task happen often enough to justify investigation?
• Are inputs and decision rules recognisable?
• Can the systems exchange the necessary information?
• Is someone responsible for exceptions?
• Can you compare the current and proposed results?

This is a decision framework, not a validated scoring model. A low-volume task can still matter if an error has major operational consequences. Discuss that separately rather than reducing everything to time saved.

Decide whether AI is necessary

Fixed rules are often enough for known fields and predictable decisions. AI may help when inputs vary, such as free text or documents. Combine recognition with validation and human review for uncertain cases. Microsoft describes cloud flows as actions started by an event, button or schedule; those workflows do not automatically require AI.

Design a trial that answers a question

Illustrative scenario: an advisory firm collects customer documents. A first flow registers receipt and creates tasks for missing items. An adviser still decides whether the file is substantively complete. Measure both the registration work removed and the corrections introduced.

Agree in advance when to expand, adjust or stop the trial. Explore AI and automation or read about handling workflow failures. Background: Microsoft on cloud flows.