Redesign workflows before you automate them

Automation is not a substitute for workflow design

Automation is attractive because it promises speed. But speed inside a poor workflow can make a weak process harder to see and harder to correct.

If a task is unclear, duplicated, or routed badly, automation tends to hide the problem instead of solving it.

Start with the decisions, not the tooling

Before introducing AI or automation, define:

• What outcome the step is supposed to produce
• Who owns the decision
• What information is required
• What happens when the case falls outside the happy path

This creates an operating model that can support tooling choices instead of being shaped by them accidentally.

Where AI belongs

Once the workflow is clear, AI can support several useful functions:

• Drafting or summarising repeat communications
• Classifying incoming work for faster triage
• Suggesting next actions based on defined playbooks
• Pulling together context that humans still review before acting

The key is that the AI step sits inside a designed workflow with visible control points.

What to avoid

Avoid using AI as a patch for:

• Undefined approval rules
• Missing ownership
• Conflicting data sources
• Work that nobody has simplified first

Those conditions create brittle implementations that disappoint teams quickly.