The first week of an automation engagement contains no code. We sit with the people doing the work and count: how many times a day, how many minutes each time, how often it goes wrong, and what happens downstream when it does.
The result is usually surprising. The task everyone complains about costs eleven hours a month. The task nobody mentions — copying figures between two systems because the integration was never finished — costs ninety.
We rank candidates by hours saved divided by build complexity, then automate top-down. That ordering matters more than the tooling. A crude script against the right process beats an elegant pipeline against the wrong one.
One rule we do not break: every automated step keeps an audit trail and a human override. Automation that cannot be inspected is not a time saving, it is a liability with better latency.