Entity Resolution

What Is Entity Resolution?

Entity resolution determines when different data records refer to the same real-world entity. It links records about people, organizations, products, locations, households, accounts, or devices despite duplicates, missing values, conflicts, or different formats.

An entity is the real thing being described, while a record is one system's description of it. One customer may appear as “Robert Lee” in sales, “Bob Lee” in support, and “R. Lee” in billing. Entity resolution weighs the evidence to decide whether these records describe one customer or different people. The result may be linked records, an entity profile, or a governed merged record.

This makes reports, decisions, and automated workflows less likely to count the same thing twice or combine separate things.

A Simple Example of Entity Resolution

A retailer has separate sales, support, billing, and marketing databases. Sales lists “Maya Patel” with [email protected]. Support has “M. Patel” with a phone number and new address. Billing has “Maya Patal” with the same cardholder address but no email. Marketing has subscriptions using a nickname and a shared household phone number.

The sales, support, and billing records may provide strong evidence of one person through similar names, address history, and contact details. The marketing records need more caution because a shared phone number does not prove identity. Automatic merging should occur only when evidence and business risk justify it.

How Entity Resolution Works

Reliable matching is a repeatable process, not a one-time duplicate search. A sound system retains evidence for each decision.

1. Define the entity and decision, such as consolidating service history or detecting linked accounts.
2. Collect source records and retain source names, IDs, update dates, and permissions.
3. Standardize values, including addresses, phone formats, and common name variations.
4. Block or retrieve plausible candidates instead of comparing every record with every other record.
5. Compare useful attributes, such as names, emails, addresses, dates, account IDs, and relationship signals.
6. Apply exact rules, weighted scores, or models to estimate whether pairs match.
7. Assign match, possible match for review, or non-match using defined thresholds.
8. Group records for the same entity while checking that indirect links do not create unsafe match chains.
9. Apply survivorship rules to choose a preferred profile value without deleting source evidence.
10. Publish linked results and monitor new records, corrections, and match quality.

Core Components of an Entity Resolution System

An entity resolution system needs source data and an entity schema defining expected attributes for a customer, supplier, patient, product, or other entity. It also needs identifiers, normalization rules, comparison features, match logic, confidence thresholds, and clustering rules.

Survivorship rules choose the preferred value when records disagree. For example, a verified customer profile may outrank an old marketing list. Strong systems preserve source lineage, timestamps, match reasons, audit logs, and a human review queue. Reviewer feedback should improve rules and models, especially when a CRM workflow depends on a trusted customer profile.

Entity Resolution Methods: Deterministic, Probabilistic, and Machine Learning

Different methods fit different data quality levels and risks. Many systems combine them.

Entity Resolution vs. Deduplication, Identity Verification, and Master Data Management

These practices overlap but solve different problems. The distinction prevents teams from treating a possible match as proof of identity.

Common Entity Resolution Use Cases

Entity resolution answers a practical question: what information belongs to the same real-world subject?

• Customer 360 and CRM consolidation for fuller histories without duplicate contacts.
• Fraud and financial crime detection for connected accounts, devices, or addresses.
• Know-your-customer workflows across onboarding and account systems.
• Healthcare patient matching to reduce fragmented or incorrectly combined histories.
• Supplier and vendor management to identify duplicates and organization-wide exposure.
• Public-sector case management for permitted records about the same person or household.
• Product catalog matching for differently named listings of the same product.
• Data migration to avoid duplicate entities in a new platform or web application.

Benefits of Entity Resolution

When match logic fits the decision and data is governed, entity resolution can improve service and analysis.

• Creates more complete views of customers, suppliers, products, and other entities.
• Reduces duplicate outreach, repeated entry, and inconsistent work.
• Improves reporting by reducing double counting and fragmented histories.
• Helps teams recognize relevant service, risk, and compliance signals sooner.
• Supports safer data sharing through controlled, traceable links.
• Provides a foundation for dashboards and automation when access and quality controls remain in place.

Practical Limits and Common Pitfalls

Matching is an inference problem. Even advanced systems can cause harm when rules, data, or controls are weak.

• False positives combine different entities, while false negatives leave records fragmented.
• Emails, phone numbers, and addresses may be shared, recycled, or entered incorrectly.
• Common names, stale data, and biased historical data can create uneven results across groups.
• Poor thresholds can automate uncertain actions or overload reviewers with weak candidates.
• Transitivity is risky: if A resembles B and B resembles C, A and C may still differ.
• Unsafe merges can overwrite valid information and make correction difficult.
• Missing provenance makes links hard to explain or reverse.
• A match score is model-based likelihood, not a verified fact or permission to use data for any purpose.

How to Evaluate and Improve Entity Resolution Quality

Measure quality against the real consequence of a wrong or missed match, not an abstract accuracy target.

• Start with a specific decision and identify harm from incorrect or missed matches.
• Create labeled test pairs from representative data, including shared addresses and common names.
• Measure precision, how often proposed matches are correct, and recall, how many true matches are found.
• Use separate thresholds for automatic matches, review, and non-matches.
• Sample results regularly and show reviewers the evidence behind each decision.
• Preserve source records and test rule or model changes before release.
• Monitor drift when source systems, formats, behavior, or identifiers change.

Governance, Privacy, and Trustworthy Matching

Entity resolution needs a lawful, defined purpose. Collect and link only necessary data, provide notice or obtain consent where required, and limit access through role-based controls. Retention limits, correction processes, and auditable logs help people and organizations challenge or repair incorrect links.

High-impact uses, including healthcare, financial decisions, benefits, employment, or safety actions, need meaningful human oversight. Responsible systems do not convert uncertain inferences into facts. They preserve source evidence, allow reversal, and treat privacy, fairness, and security as core design requirements.