02 · Target state

A governed Azure medallion, delivered in two phases

AWS keeps doing the computation. An Azure VM running the Amazon Athena ODBC driver pulls order data across, and the D365 team publishes Dataverse through Azure Synapse Link - both landing raw in Bronze on ADLS Gen2. Silver and Gold curate; Gold serves Pigment and posts back into the ledger.

Phase 1 protects the record from day one; Phase 2 replaces the manual routesTrainline · target state

100%

Phase 1 · Capture & audit trail

Both sources land raw and immutable, exactly as they arrived. Nothing is reshaped yet. From day one Trainline holds a complete record of AWS and Dynamics 365 data in its own tenant - the thing that makes every later number retraceable.

Phase 2 · Designed with finance

Silver conforms both sources; Gold applies the aggregation grain, measures and batch reference key agreed with finance - so a served figure traces back to raw, and a raw record traces forward to where it landed.

LegendPhase 1 - Raw capture into BronzePhase 2 - Silver, Gold & serving1Numbered step in the primary data flow

A governed medallion in the Microsoft tenant, in two phases

Phase 1 protects the record from day one; Phase 2 replaces the manual routes.

Phase 1 - Capture and audit trail: configure the Azure VM with the Amazon Athena ODBC driver, work with the D365 team to publish Dataverse through Azure Synapse Link, and land both sources raw into Bronze on an ADLS Gen2 storage account with hierarchical namespace enabled. Nothing is reshaped yet - the goal is that from day one Trainline holds a complete raw record of both sides in its own tenant.
Phase 2 - Curate and serve: Silver conforms and deduplicates the two sources; Gold applies the aggregation grain, measures and batch reference key agreed with finance. Gold then serves Pigment and posts back into Dynamics 365, so both sides finally come from the same place.
Because Bronze holds the raw record, every served number can be retraced back to what arrived and forward again - the full audit trail the current estate has never had.
Processing can run on Microsoft Fabric, Azure Synapse, or Daxia - our own React-based Azure Web App, deployed in Trainline's tenant at no licence cost, running the pipelines on cron jobs at a fraction of the usage cost of ADF or PySpark.