Where to start with AI automation

AI and automation create the most value when they remove reliable, repeatable work — not when they replace judgement. The first question is never which tool to buy. It is which part of the work someone does the same way every single time.

Make a list of the tasks your team repeats weekly. Data entry, report generation, status checking, document drafting. Score each one on three things: how often it happens, how well-defined the rules are, and how costly a mistake would be. The high-frequency, well-defined, low-risk tasks are the place to start.

Automate the boring path first. Connecting two tools so a form moves information without a human retyping it is worth more to most organisations than a clever model that drafts persuasive emails.

Where you do use AI models, keep a human in the loop for anything consequential. An AI draft that a person reviews is a productivity tool. An AI decision that nobody reviews is a liability. Decide explicitly which of the two you are building.

Measure before and after. Track time saved, error counts, or throughput — whatever the task actually produces — using the same method before and after the change. Publish only what you measured, and be honest when the automation did not help.

Pilot something small within two weeks. A real workspace with real feedback will teach you more than any vendor demo.