Define the system contract
Record data ownership, format, frequency, acceptable latency, volume, change ordering and acknowledgement. Decide what happens when the receiver is offline or processes a message twice.
Use synchronous APIs for immediate answers
HTTP services and REST/OData suit current-data queries and short operations whose result is needed immediately. Do not keep a long business process inside one request: a timeout cannot prove whether the operation completed.
Use queues to decouple processing
Asynchronous delivery suits high volume, unstable links and work that may finish later. The 8.5.4 Queue mechanism can separate mandatory deferred actions from synchronous document processing. Add idempotency, message identifiers, retries, a failed-message quarantine and delay monitoring.
Keep scheduled exchange where it fits
Batch synchronization is clear and maintainable when delay is acceptable. Control its window, volume, restart behaviour and conflicts. Protect the production database with rate limits, pagination, least-privilege service accounts and monitoring of business outcomes, not transport errors alone.