Start by observing the current report for a few cycles. Record how long it takes, which steps are repeated, which inputs change, and where someone has to make a judgment.
Automate stable preparation steps first: collecting files, validating schemas, refreshing models, and distributing a consistent output. Keep decisions and exceptions visible to the people responsible for them.
Measure the result after launch. Time saved matters, but so do fewer errors, faster access, and a clearer record of how the numbers were produced.