How would you build a Customer golden record across fragmented systems?

WHAT THIS TESTS: This question probes whether you can bridge business and technology to ensure uniformity, accuracy, stewardship, semantic consistency, and accountability of shared master data assets. It is not enough to propose a database; the interviewer wants to see systems thinking about distributed truth, conflict ownership, and operational sustainability at scale.

A GOOD ANSWER COVERS: First, identity resolution. Explain how you establish canonical identifiers by mapping foreign keys across CRM, billing, and support systems using deterministic and probabilistic matching. Second, survivorship rules. Describe how you resolve conflicts with business-defined precedence, such as billing overriding CRM on addresses when payment is involved, plus timestamps and source reliability scores. Third, architecture. Sketch a scalable pipeline, perhaps an event-driven merge service or a lambda architecture, that ingests changes, applies rules, and publishes the golden record to a serving layer with clear data lineage. Fourth, governance and feedback loops. Mention stewardship workflows where downstream teams flag bad merges, automated quality monitors track match rates, and a council of business owners revises rules as systems evolve.

COMMON WRONG ANSWERS: A red flag is proposing a naive nightly batch dump into a single table without conflict resolution or ownership. Another is ignoring semantic differences, such as a support ticket email being less authoritative than a billing email. Candidates also err by jumping to a specific technology, like a graph database, before defining the business rules and matching logic. Treating the problem as a one-time data cleanup rather than an ongoing operational process signals immaturity.

LIKELY FOLLOW-UPS: Expect the interviewer to ask how you handle hard deletes versus soft deletes across source systems, how you propagate corrections back to sources, or how you guarantee low latency when a customer updates their profile before a purchase. They may also probe cost tradeoffs between real-time streaming and daily batch, or ask how you measure golden record accuracy.

ONE CONCRETE EXAMPLE: Imagine a customer exists in Salesforce with email A, in Stripe with email B, and in Zendesk with email A plus a new phone number. Your identity graph links all three via email A and fuzzy name matching. The survivorship rule gives Stripe priority for email because it processes payments, Salesforce priority for industry tags, and Zendesk priority for phone if it is newer than ninety days. The merge service emits a versioned golden record to a DynamoDB table, while a data quality dashboard alerts stewards when match confidence drops below ninety-two percent.