Integration has to start from the actual capabilities of the existing software.
Before promising an integration, APIs, import/export formats, databases, webhooks, permissions and licensing limits need to be verified. Where no reliable integration point exists, that constraint should be explicit.
A fragile automation that simulates clicks may be acceptable for a prototype, but it should not silently become critical infrastructure.
- which system is authoritative for each data type
- which writes need transactional guarantees
- how errors and duplicates are handled
- which integrations are supported by the vendor
- what happens when one system is unavailable