An ERP knows invoices and stock, a CRM knows accounts and interactions, and a commerce platform knows carts and orders. Bringing their tables into one storage account does not create a shared view if they disagree about what a customer or sale is.
Begin with identities and events
Map the business objects before building pipelines. Which CRM account corresponds to a billing entity? How does an order become an invoice? Can one order create multiple shipments or refunds? Write the joins and exceptions down with the business teams. A synthetic “customer 360” row that quietly merges unrelated accounts is worse than an honest gap.
Build a small canonical event vocabulary: order placed, payment captured, shipment dispatched, return received. Retain original source identifiers and timestamps. That makes reconciliation possible when two systems update at different times.
Avoid replacing one silo with another
A new central store can become its own silo if only one team understands its transformations. Publish data contracts, ownership and descriptions for the datasets others will use. Establish tests for missing IDs, impossible state changes and unmatched financial totals. Alert an owner when exceptions grow instead of silently dropping records.
Test a business question end to end
Ask, for example, “How many orders from new customers were delivered late and then generated a support case?” Work through a small sample manually. If the answer depends on assumptions about guest checkout, split orders or case linkage, document them. Integration succeeds when teams can explain the answer and correct it, not merely when three connectors show green.
