Clear data ownership across ERP and commerce
The challenge
An ERP-led commerce architecture with separate responsibilities for catalogue, stock and pricing.
An anonymised engineering account. Customer identities, operating figures and engagement dates are withheld. The focus is the problem, design decisions and validation, rather than a claim of commercial improvement.
A storefront replacement can expose responsibilities that an older integration left implicit. Finance needs the ERP to remain authoritative, while the commerce platform needs a product model that supports merchandising and customer journeys. Sending every source field to every destination makes ordinary catalogue changes difficult to manage.
The engineering problem is to separate the records that need different ownership and recovery rules. Product descriptions, saleable availability, images and prices do not necessarily change together. A failed image should not prevent an approved price update, and a catalogue repair should not overwrite inventory.
The approach
Agree ownership before mapping
Define the owner of each field and the circumstances in which another system can change it. Preserve source identifiers and map them to destination records. Record the treatment of variants, inactive products, units of measure and prices before building transformations.
Separate changes by responsibility
Keep catalogue, availability, media and pricing responsibilities independently recoverable. Shared identifiers connect these processes, but their completion and failure states remain visible separately. The aim is to repair the affected responsibility without replaying unrelated changes.
Make mappings reviewable
Store field rules and transformations in a form the operating team can inspect. Validate mandatory values and reference data before submitting a destination request. A rejected product needs an understandable reason and an owner, rather than a generic failed-run message.
Plan partial failure and reconciliation
Record which destination accepted a change and which did not. Reconcile source and destination by stable identifiers and relevant field values. Check the current source state before replaying an older update so a recovery does not restore superseded data.
Document the trade-off
Separating responsibilities makes failures easier to isolate, but introduces dependencies to manage. Agree when an item is ready for publication, how incomplete data is held back and who can approve exceptions. A more modular design still needs an explicit operating model.
Design outcome and validation
The design makes the boundary between ERP ownership and storefront presentation explicit. Configuration, failure handling and handover can be reviewed by responsibility rather than treated as one opaque synchronisation.
Validation evidence to retain
Use synthetic or approved test records to demonstrate a new product, a variant change, a withdrawal, a price change and a stock correction. Record expected and observed destination states. Include an unavailable endpoint, a missing mapping and a repeated notification. Confirm that repairing one responsibility does not overwrite another.
Keep the mapping register, acceptance results, unresolved limitations and recovery instructions with the delivery record. This account does not publish measured performance, transaction volumes or customer results.