Skip to content
Ixia Consulting

Master Data

Golden Records: What Survivorship Rules Actually Decide

Matching finds the duplicates. Survivorship decides what the one surviving record says, and that is a business decision.

Dave Welensky, Director · Published November 2026 · 5 min read

Most consolidation effort goes into matching, because matching is tractable. The harder question comes after it: when sources disagree, which value wins. Survivorship has to be agreed by the business, recorded and repeatable, or the golden record is only a guess with a good name.

Matching is the easy half

When an organisation decides to consolidate a customer master, the first conversation is about matching. How do we recognise that three records describe the same customer? Exact keys, fuzzy name and address similarity, tax numbers, phone numbers, scored and clustered. It is real engineering, it is tractable, and it produces satisfying output: a list of clusters, each holding the records that probably belong together.

But a cluster is not a customer. It is a set of candidate records that disagree with each other in small, consequential ways. The name is spelled two ways. The address differs. One record carries a tax number and two do not. The payment terms are not the same. Matching tells you the records belong together. It does not tell you what the single record should say.

Survivorship is where the decisions live

Survivorship is the set of rules that decides which value survives into the consolidated record when sources conflict. The shapes are familiar: the most recent value, the most complete value, the value from the most trusted source, or an explicit business rule per attribute. Different attributes usually need different rules. The delivery address may be best taken from the system that ships goods. The legal name and tax number may be best taken from the finance record.

The risk is not choosing the wrong rule. It is that nobody chose. When survivorship is left to a default in the tool, or to whichever engineer wrote the merge script, the golden record is built on decisions nobody made and nobody can later explain.

Why the business has to own it

Which value wins changes what an invoice says, where a delivery goes and which terms apply. Engineering can make a rule repeatable. Only the business can say that it is right.

We apply three tests to a survivorship rule. It must be agreed, by the people who own the data and live with the consequences. It must be recorded, written down with the reason it was chosen. And it must be repeatable, so that running the consolidation twice gives the same answer. A rule that fails any of the three is a preference, not a control.

On the customer master consolidation described in our case study, the matching and survivorship rules were agreed with the business, and the consolidated record kept lineage back to every contributing source. That lineage matters more than it sounds: when someone asks why a customer's address says what it says, the answer is a source and a rule, not a shrug.

Queue the doubtful ones

Matching at any real scale cannot be checked by eye, so the logic has to be trustworthy enough to merge records automatically. It also has to be honest enough to admit doubt. The workable pattern is a confidence threshold: merge automatically above it, and queue everything in the grey zone for a data steward to decide.

The failure mode is merging on weak evidence because the queue looks long. A wrongly merged customer is harder to detect and more expensive to unpick than an unmerged duplicate. Duplicates are visible. Bad merges hide.

A golden record has to stay golden

Consolidation is a moment, and data quality is a process. Without ongoing checks, the duplicates and the drift return, because the sources that created them are still running. The same rules that decided the consolidation should keep running afterwards: duplicate detection, validity checks and completeness checks, on a schedule, with the trend visible to the people who own the data.

That is the difference between a clean-up and a standard. A clean-up is a project that ends. A standard is a measurement that continues.

Key takeaways

  • Matching finds duplicates. Survivorship decides what the surviving record says.
  • Survivorship rules are business decisions. Engineering makes them repeatable.
  • A rule is only a control if it is agreed, recorded and repeatable.
  • Merge automatically on strong evidence and queue the rest for a steward.
  • Keep lineage from every golden record to its sources, and keep measuring after go-live.

Recognise your programme in this?

Written by the people who deliver the work. The conversation works the same way.

Discuss master data quality