AI automation for small businesses: begin with the workflow.

Start from the hand-off, not the technology

Map what enters a process, who currently reviews it, which system it ends up in, and where exceptions happen today. An automation project that starts with "we should use AI for this" usually skips this step and builds something nobody asked for.

Automate the repeatable part first

A process that's identical 90% of the time and needs judgment for the remaining 10% is a good candidate: automate the repeatable majority and route the exceptions to a person, rather than trying to automate the judgment call itself.

• Repetitive: sorting incoming enquiries by type
• Repetitive: drafting a first-pass reply from a template
• Judgment-based: pricing an unusual custom request
• Judgment-based: deciding whether to accept a difficult client

Set explicit boundaries before day one

Decide up front which data the system may use, which actions require a person's approval before they happen, and how a human takes over when the automation reaches the edge of what it should decide alone.

Measure the queue, not the demo

A working automation should be judged by whether the backlog it was built for actually shrinks and stays shrunk — not by how convincing it looks in a first demonstration.