What Should a Digital Twin Learn First?

A personal digital twin does not need to learn everything about someone to be useful. It needs to understand one recurring task well enough to make that task easier—without making personalization feel like surveillance.

Imagine a tool that helps independent consultants prepare a weekly client update. Before building a broad memory system, what would you want to learn from a small prototype: which details people choose to reuse, what they correct, or whether they come back to it?

What is the first useful signal that a digital twin has earned more context—and what would make you stop the experiment?