Most conversations about AI begin with the tool. A better place to begin is the working week.
Look for the leak, not the showcase
In almost every business there are tasks that are repeated daily, depend on one person's memory, or involve moving the same information between two places by hand. Nobody designed them that way. They accumulated. These are usually the first places AI earns its keep, because the value is easy to see: hours returned to people who have better things to do.
Three useful tests
Is it frequent? A task done fifty times a week is a better candidate than one done twice a year. Is it structured enough to describe? If you can explain the task to a capable new starter, it can probably be supported by AI. Is a mistake recoverable? Begin where a person still reviews the result, and where an error costs minutes rather than a client.
What tends to come first
Summarising and drafting from information the business already holds. Pulling details out of documents, emails and forms into the system where they belong. Answering internal questions from your own policies and records. Preparing the first version of a report that someone then checks. None of these is glamorous. All of them remove friction that people have stopped noticing.
What should wait
Anything customer-facing and unsupervised, anything involving sensitive judgement, and anything where the underlying data is unreliable. AI built on messy information produces confident mess.
The practical next step
Pick one process. Measure how long it takes today. Improve it, keep a person in the loop, and measure again. One proven improvement tells you more about where AI belongs in your business than any strategy document.