A CRM contains two similar customer records. An automated check proposes a merge, but contracts and contacts differ. In this illustrative case, the reference-data steward must determine whether they represent a duplicate, a branch or another legal entity. A string-similarity score cannot replace that decision.
Separate a signal from authority to change
Matching names, phone numbers or addresses help find a candidate pair. They do not establish identity. Different units may share a phone number, while one organization may change its address.
Define the entity represented by the system: counterparty, unit, contact or contracting party. Then identify authoritative sources and the conditions requiring human review. An automated rule needs an owner and a defined scope.
Assign different decisions to different roles
The data owner approves the business rule. The quality steward assembles evidence and handles exceptions. IT configures checks, logging and technical execution. One person may hold several roles, but that arrangement should be explicit.
The UK Government Data Quality Framework considers quality in terms of fitness for purpose. It is a useful reference for choosing checks, not a legal obligation for Ukrainian businesses. Here, the purpose includes preserving the correct contract relationships, rather than merely reducing the row count.
Check dependencies before merging
List contracts, documents, open requests and analytical datasets referencing either record. Decide which value survives for each field and why. “Most recently changed” does not always mean “correct.”
If evidence is insufficient, leave the pair in a queue with an owner and a next action. Do not delete a record simply to remove an alert. Also establish how to restore relationships if the merge proves incorrect.
| Situation | Action | What to check |
|---|---|---|
| Similar names, different identifiers | Do not merge on name alone | Entity type and verified sources |
| One entity, different contacts | Agree field values | Currency and purpose of each contact |
| Insufficient evidence | Keep the exception open | Owner and next review action |
Leave a decision that can be explained
The log should record original identifiers, evidence, approval, changed relationships and execution time. Post-merge verification must reach the systems consuming the reference data. A successful MDM command does not prove that a report or contract now references the correct record.
E-Data Consulting helps establish these rules before configuring reference data and workflows in Intecracy Group solutions. Logging and recovery capabilities are verified in the selected product. For the first pilot, select different disputed-pair types rather than only obvious duplicates.
Questions and answers
Can IT own data?
Yes, for data belonging to its own process. Technical access alone does not authorize IT to define the business meaning of a customer.
Can obvious duplicates merge automatically?
Only within an agreed rule tested on representative examples, with logging and defined recovery. Ambiguous pairs require separate review.
Source and method
Government Data Quality Framework. The worked example and acceptance procedure above are original editorial guidance, not a quoted standard or a real client case.