What I've learned
If a process is confused, automation doesn't fix the confusion — it just moves it faster. I've seen AI initiatives struggle not because the technology was weak, but because it was pointed at a process nobody had actually agreed on yet.
Common mistakes
Automating the exception-handling instead of the repeatable work. Good uses of AI include summarising notes, spotting repeated themes, drafting clear updates, and routing common requests — not replacing the judgement calls that need a person who understands the customer.
Practical approaches
Start with what's repeated: updates, checks, triage, reports, reminders, handovers. Keep people in charge where empathy, context, or trust matters. Then measure honestly — are customers getting clearer help, and are teams actually spending less time on repeat work, or just looking busier?