The unit of migration is the flow, one business movement of data, end to end, not the BizTalk application and not the server. For every flow, record: source and destination systems with their adapters, the schemas and maps it touches, any orchestrations (sorted into the four triage piles), BRE policies, custom components, and the bindings that differ per environment. BizTalk admins can pull most of this from the admin console and bindings exports; the orchestration triage is the only part that needs an architect’s eyes.

Then color the list. Green: messaging-only flows, piles 1 and 2, no orchestration state. Yellow: serial loops and debatching, pile 3, shipping patterns cover them. Red: correlation-dependent, pile 4, sequenced last. Rank within each color by business risk, lowest first: your first migrated flow should be one whose failure nobody would notice for an hour, not your order intake.

Deploy Art2link ESB from the Microsoft Marketplace into your tenant. The two platforms share nothing and cannot interfere with each other, BizTalk keeps running every flow it runs today, and the new platform sits empty, costing only its Azure footprint. Stand up your environments the same way you run them now (the snapshot seam carries an application across them), connect Authentications to the same systems BizTalk talks to, and prove connectivity with a throwaway flow before any real migration starts.

Per flow, the loop is always the same. Rebuild it as configuration: an Application, message types with the promotions you route on, maps (the AI Accelerator drafts XSLT from samples of both sides), ports with the right adapters. Test with real recorded payloads, watching every hop in Tracking with the level turned up, and wire the exception route before go-live, fail loud, never drop. Cut over by repointing the partner or source system to the new endpoint, start send ports before receive ports, and stop the BizTalk version without deleting anything: a stopped BizTalk flow is your rollback, one repoint away. Promote the finished flow across environments with a snapshot, re-valuing Constants at import.

Run the migrated flow under watch, Activity Notifications on error, tracking at Only on Error once stable, until it has survived a full business cycle including the month-end or seasonal spike that breaks things. Then take the next flow off the list. Red flows close the list: by the time you reach them, either Correlation has shipped or the redesign option has had months of your team’s new platform fluency applied to it, both inside a support window that runs to 2028 mainstream and 2030 extended.