Structure a master record of customers and products: identification, deduplication, responsibility and propagation of changes between systems.
Index
Two registrations with similar names can be the same company or different units. Unifying by approximation can bring together histories that should remain separate. A master record project needs to combine identity, assignee, and conflict handling criteria before synchronizing data at scale.
Decision this guide helps you make: Define a consistent identity for records shared across systems.
Choose who decides each information
The official system may vary by field: commercial status in one place, logistical information in another. Document this responsibility and the identifier that links the records. Don't automatically choose the most recent data; it may have been updated by a less reliable source. Relevant conflicts need a rule or human review.
Separate candidates from approved mergers
Algorithms can suggest duplicates, but the decision depends on the context. Define which matches are strong and which require checking. Preserve old identifiers as references when merging. The process must allow investigating the origin and correcting an improper merger without irreversibly erasing the history.
Control the distribution of changes
After patching the master, determine how consuming systems receive the change and commit application. Incomplete updates must be visible. Deletion and inactivation are not equivalent: an out-of-sale product may still be necessary to consult old orders. Policy must protect these relationships.
Assess quality in a limited domain
Start with an entity type and concrete indicators, such as missing required fields and unlinked records. Compare before and after samples with users who rely on the data. An aggregate percentage of quality is not enough if the customers with the greatest impact continue to be wrong.
- Source responsible for each field.
- Identity criteria and conflict review.
- History of mergers and propagation.
A scenario to check out in the demo
Hypothetical example: head office and branch share part of the name, but have their own deliveries and conditions. Text deduplication may improperly join records. The project must determine the identity necessary for each process and preserve corporate links without confusing them with registration equivalence. Human review of difficult candidates enters the deployment effort.
Briefing to request a proposal
- Entities and fields that have an official source defined.
- Matching criteria that authorize or merely suggest unification.
- Procedure for correcting an improper merger or distribution.
Hire governance along with integration
Quantum9 can design the technical flow and controls to keep registration consistent. The company needs to appoint someone who resolves business queries. Without this responsibility, integration tends to spread divergences more quickly instead of correcting them.
Discover the scope of Data engineering and deepen the context in related guide.