What it looks like over a year

Months one to three: a company AI policy, staff training and one or two high-volume processes automated, such as enquiry replies and bookkeeping capture. Months four to nine: more functions follow, systems get connected so data flows without retyping, and reporting becomes automatic. Later: AI becomes part of how new processes are designed from the start.

The order differs by company, but the pattern of small steps with measured results is what makes it stick.

Where it goes wrong

Treating it as a technology purchase rather than a change in how people work. Rolling out tools without training. Automating broken processes. And measuring activity, such as the number of tools adopted, instead of outcomes such as hours saved, faster replies or more revenue per employee.