Start with the smallest data path
List the system that creates data, the authoritative record and the consumers of each result. Distinguish a one-time import from ongoing synchronisation. Give read-only queries and writes different permissions. An available API is not evidence of business authorisation.
Agree identifiers and failure handling
Define unique identifiers, time formats, null values and duplicate handling. A sync job should be safe to retry without creating duplicate records after a timeout. Provide clear handling for partial success, conflicts and manual correction.
Run a small scope and retain evidence
Compare inputs, outputs and authoritative records using samples in a test environment first. After launch, monitor failed jobs and permission changes and keep only necessary logs. Keep credentials out of source control. The process owner should be able to pause the connection and understand recovery steps.