Every Friday, someone on your team exports three CSVs, pastes them into a spreadsheet, repairs the formulas that broke, and emails a PDF to the enterprise customer whose pilot you want to turn into a contract. It takes the afternoon. When that person is out, the report goes late, and the customer's champion notices before you do.
At Boston Dynamics I built DHL's robot fleet reporting system, working with DHL directly. DHL is a Boston Dynamics customer, and the order of the work on that project is the order I'd use on any recurring customer report.
Start with the reader
Ask the customer which numbers they look at first and what they decide with them. DHL wanted boxes picked and hours operating. A hand-built report collects whatever was easy to export, so cutting what the customer doesn't read makes the report shorter before anything is automated.
Write down where each number comes from
Each number gets one source and one definition. If two of your engineers would calculate uptime differently, settle which way is right before a script calculates it for you, because a scheduled job repeats the disagreement every week without anyone seeing it. Fix the mess first. Then build. Then automate.
Then automate the pull
Once the definitions are agreed, a scheduled job pulls the data and fills the report. Looker Studio, or a dashboard tool you already pay for, often covers it. A custom build earns its cost when the customer needs something those tools can't show, or when the data sits in a Postgres replica or an internal API they can't reach.
Keep the Friday check
Someone still reads the report before it goes out. The check takes five minutes instead of an afternoon, and it catches the week a sensor went offline and the numbers looked too good.