I use AI across the businesses every week. I also leave it out of far more places than I put it in. The difference is not about the technology. It is about whether there is a real, repeatable judgement for it to support.
Three places it earns its keep
Structuring messy information. Property brochures, maintenance reports and product ingredient lists all arrive in inconsistent formats. Turning them into one comparable structure is tedious for people and natural for a model.
Explaining, in plain language. Once information is structured, a person still needs to understand what it means for them. Generating a clear explanation from verified facts is a strong use, as long as the facts come first.
Triage. Deciding what needs attention first, in a maintenance queue or an inbox, is a repeatable judgement with a clear definition of good.
Where I keep it out
Anywhere the answer needs to be trusted more than it needs to be fast. Pricing. Medical or safety claims. Anything that reads as a promise to a customer.
A model that is confidently wrong once costs more trust than a hundred correct answers earn back.
The practical test
Before adding AI to a process I ask three questions. Is the judgement repeatable? Can the input be verified? Will the person on the other end understand where the answer came from?
If any answer is no, it is not ready, however impressive the demo.
