A composable stack has work that belongs to none of its platforms. Data that needs consolidating, integrations that need somewhere to run, jobs that need to happen on a schedule and be visible when they fail.
We use Google Cloud for that layer. BigQuery when reporting needs data from several systems in one place, managed services for the integrations that keep commerce and the ERP in step, and monitoring so a failure surfaces before a customer finds it. This is plumbing rather than a product decision, and it should be sized to the problem.