A few weeks ago, I sat with a team at a private members’ club and looked at one everyday job. They wanted to use member information to send better messages, make joining easier and give staff clearer reports. They weren’t looking for a clever chatbot. They wanted the work to run more smoothly.
It’s tempting to choose a tool first. I’ve learned to start with the job instead. Watch what happens from the first request to the finished result.
Write down each step
First, we wrote down what happened. Someone collected member information. Someone decided what to say. A message went out. Member preferences were recorded. The results went into a report.
At each step, we noted who did the work and what they needed next. That showed us where work could wait, get repeated or disappear.
The AI request became one job with a beginning and an end. We could see which information mattered, where someone had to make a decision and where a small tool might help. That was better than choosing an app and forcing the club to fit it.
Write down one real task
Most owners carry the process in their heads. You know who checks the inbox, which spreadsheet gets updated, what needs approval and what happens when someone is away. The business depends on that knowledge, but nobody else sees the whole job.
Write down one real task. Start when the request arrives. Finish when the customer, colleague or owner has what they need. Include the awkward bits: missing information, an approval, an exception and the last-minute change.
Leave those out and automation will look neat but work badly. Then decide whether AI has a useful role. It might draft a message, sort information or prepare a first report. Decide where it may act, where it must stop and who checks the result.
Then you can see whether it saves time or moves confusion into a new screen. You do not need a perfect diagram. You need enough detail to test one part of the job.
This week, choose a task that happens often. Write the steps in order. Mark where someone waits, checks something or makes a decision. That is a better place to start than another hour comparing tools.
Also in this issue
The news I picked out on 18 August 2026. Some of it may have moved on since.
- Worth being careful aboutNew Claude models now carry a hidden watermark
Anthropic says Claude models launched in the EU on or after 2 August 2026 get an invisible watermark in generated text. It can survive copying and some edits. Read the guide for files and the limits.
- Tools worth knowingChatGPT gives free users more thinking time
OpenAI is giving free ChatGPT users unlimited text chats and a Think button for harder questions. Try it on a real brief beside a normal answer. Keep the one that helps you decide.
- Running the businessZapier can save you from copying data into Google Sheets
Zapier connects Google Sheets to other apps, so new lead details can become rows automatically. The guide shows examples for leads, alerts and projects. Start with one simple handoff and check the rows before expanding it.