Start with the work, not the tool
In service businesses, automation usually starts with the question "what can we automate?". That is the wrong question. It produces a list of thirty ideas, twenty-five of which have no owner, and five of which touch processes nobody needs.
A better question: where did the team lose the most hours this week on work nobody should be doing? The answer is usually boring: retyping data, sending the same information, hunting for a file, reconciling two versions of a report.
Three mistakes I see most often
1. Automating a mess. If a process has three exceptions and nobody knows which one applies, a script will simply produce wrong data faster. Simplify first, automate second.
2. No single number. A project without a measure cannot be defended after three months. Pick one number before you start: hours, error count, response time, cost per transaction.
3. No owner. An automation nobody watches breaks quietly. You need a person who checks once a week that the output still matches reality.
How I start
Always with one process and one week of observation. I write down the steps, measure the time and ask the team what gets in their way. Only then do I choose a tool. Sometimes the answer is a better form, not an AI agent.
That is less impressive than a demo, but it works. And in operations, only what works on Monday at seven in the morning counts.
