Company record matching is rarely a matter of comparing two legal names character by character. A CRM may contain a trading name, a finance system may use a registered entity, and an enrichment provider may identify the same organization through a domain or business identifier. This checklist provides a repeatable workflow for matching company records, separating subsidiaries from true duplicates, and routing uncertain cases to review before a merge changes operational data.
Overview
Business entity matching is the process of deciding whether records from different systems refer to the same real-world company. It is a form of entity resolution and record linkage, but the decision is more consequential than ordinary fuzzy search. A close name match can support investigation; it should not, by itself, authorize a merge.
Company data commonly varies across sources in several ways:
- Legal names include suffixes such as Ltd, LLC, Inc, GmbH, or S.A., while operational systems omit them.
- Trading names, brands, abbreviations, and former names differ from the registered name.
- Domains may belong to a parent company, a regional operation, or a brand rather than the contracting entity.
- Registration numbers and tax identifiers may be missing, reformatted, or stored in separate fields.
- Addresses change, are abbreviated differently, or identify a shared office.
- Subsidiaries, branches, franchises, and parent companies may all appear as separate records.
The safest workflow separates three decisions: whether two records are plausible candidates, whether the evidence supports the same legal entity, and whether the records should be operationally merged. Fuzzy matching is useful for the first step and sometimes the second. The final step should also consider ownership, jurisdiction, identifiers, contracts, invoices, and the purpose of the dataset.
Compare fields by reliability
| Field | Typical value for matching | Important limitation |
|---|---|---|
| Registration or incorporation number | Very strong when verified and tied to the same jurisdiction | Formats differ; a parent and subsidiary have different numbers |
| Tax or business identifier | Strong when the identifier type and country are known | May identify a tax account, branch, or reporting unit rather than the whole group |
| Legal name | Strong when normalized and supported by another field | Common names, translations, and former names create ambiguity |
| Domain | Useful corroborating evidence | One domain can represent several entities; agencies, brands, and parent groups may share it |
| Address | Helpful for disambiguation and location matching | Shared offices, relocations, and formatting differences reduce certainty |
| Phone, email, or contact name | Useful supporting evidence | Contacts move between companies and shared service teams |
| Industry or website description | Helpful for spotting implausible matches | Descriptive fields are not unique identifiers |
Checklist by scenario
Before matching CRM and vendor records
- Define the target entity. Decide whether the output represents a legal contracting party, a corporate group, a customer account, or a physical location. Do not merge records until this scope is explicit.
- Preserve source values. Keep the original name, address, domain, identifiers, and source-system IDs. Create normalized fields alongside them rather than overwriting evidence.
- Build a normalization pipeline. Standardize case, whitespace, punctuation, Unicode variants, common legal suffixes, and safe address abbreviations. Apply country-aware rules, and retain the original tokens for auditability.
- Separate identifier types. Store the identifier value, issuing country or jurisdiction, identifier type, and verification status. Never compare bare numbers across countries as though they used one global namespace.
- Create candidate blocks. Use exact or near-exact identifiers first, then combinations such as normalized name plus country, domain plus country, or name plus postal area. Blocking reduces unnecessary comparisons and limits unrelated fuzzy matches.
- Score evidence by field. A normalized legal name match should not automatically outweigh a conflicting registration number. Use explicit rules for agreements and contradictions.
- Route borderline cases to review. Store the matched fields, score or rule outcome, candidate alternatives, and reason for the decision.
When names differ but the domain is the same
First determine what the domain represents. It may be the public website of a legal entity, a parent group, a brand, or a shared technology platform. Compare the domain with the legal name, address, contact details, contract entity, and jurisdiction. Treat domain equality as corroboration rather than proof.
A useful rule is to accept a domain as strong evidence only when it agrees with the intended entity level. For example, a CRM account for a regional subsidiary should not be merged with a parent-company record solely because both use the parent domain.
When a parent and subsidiary appear in the same dataset
Represent the relationship instead of forcing a duplicate decision. Maintain separate entity records when the legal names or registration numbers differ, and add a parent-child relationship where that information is known. Also distinguish branches and trading locations from independently incorporated subsidiaries.
For vendor master deduplication, ask whether invoices, purchase orders, tax reporting, payment instructions, and contracts must be associated with the same legal party. For CRM deduplication, ask whether sales ownership and engagement should roll up to a group or remain at subsidiary level. The correct answer depends on the operational purpose.
When importing enrichment data
Do not treat a provider's match key as interchangeable with your own master ID. Record the provider, retrieval date, source identifier, confidence or match status if supplied, and the fields used to connect the enrichment record. Compare new values with existing values before replacing them, particularly for legal names, domains, and addresses.
For large imports, use a staged table and an exception queue. This makes it possible to review conflicts without partially changing the production master.
What to double-check
Use an evidence-weighting template
Keep the decision explainable. A simple template can contain:
- Candidate pair: source record IDs and proposed master ID.
- Entity scope: legal entity, group, account, or location.
- Positive evidence: exact identifier, name similarity, domain agreement, address agreement, and other supporting fields.
- Negative evidence: conflicting identifier, different jurisdiction, incompatible address, or different contracting party.
- Decision band: auto-match, manual review, or non-match.
- Action: link, merge, retain separately, or create a parent-child relationship.
- Reviewer and timestamp: enough information to revisit the decision.
Similarity methods can support the evidence layer. Levenshtein distance can identify small edits, Jaro-Winkler can help with similar names, and trigram similarity can retrieve records with overlapping character sequences. These techniques are useful for candidate generation, but their thresholds should be tested against representative company names. More detail is available in the guide to fuzzy matching algorithms.
Check normalization side effects
- Confirm that removing legal suffixes does not erase meaningful distinctions.
- Keep diacritics and transliteration behavior consistent across all sources.
- Do not remove numbers that distinguish legal entities, branches, or locations.
- Review abbreviations that have different meanings by country or industry.
- Test punctuation and token handling on names containing ampersands, hyphens, and parenthetical brands.
Address matching deserves its own review because a similar address may indicate a shared office rather than the same company. Use a location match as supporting evidence, not an automatic merge rule. See the practical guide to fuzzy address matching for a safer field-level approach.
Measure the workflow, not just the algorithm
Sample accepted matches, rejected matches, and manual reviews. Label whether each decision was correct, then inspect false positives and false negatives by country, source, entity type, and field completeness. A threshold that works for domestic legal names may perform poorly for multilingual or highly abbreviated data. Keep a small regression set so changes to normalization, blocking, or scoring can be tested before release.
Common mistakes
- Merging on name similarity alone. Common names and trading names can produce convincing but incorrect matches.
- Treating a domain as a legal identifier. Domains express web ownership or use, not necessarily incorporation or contracting status.
- Collapsing an entire corporate group. Parent companies and subsidiaries may need separate records for finance, compliance, contracts, and account ownership.
- Discarding source evidence. Overwriting raw values makes later investigation and rollback difficult.
- Using one global normalization rule. Legal suffixes, identifier formats, address conventions, and naming practices vary by jurisdiction.
- Making every uncertain case automatic. A review queue is a control, not a failure of the matching system.
- Ignoring negative evidence. A high similarity score should not override a verified conflicting registration number or incompatible entity scope.
- Failing to make merges reversible. Store merge history, source IDs, field provenance, and the previous record state.
For a broader match-review-merge pattern, use the entity resolution workflow. If the input arrives as spreadsheets or recurring exports, the CSV matching checklist can help separate schema mapping from entity decisions.
When to revisit
Company matching rules should be reviewed whenever the data or the business decision changes. Schedule a check before seasonal planning cycles, recurring vendor-master reviews, CRM merge campaigns, acquisitions, and major enrichment imports. Revisit the workflow when a new country, identifier type, CRM, finance platform, or data provider is introduced.
Also review it after any change to normalization, blocking, scoring, or automation thresholds. Compare the new results with your labeled regression set and inspect borderline cases before enabling automatic merges. If users report duplicate accounts, incorrect parent-child links, or invoices attached to the wrong entity, treat those reports as evidence for improving rules and training examples.
A practical recurring routine is:
- Export unresolved and recently merged pairs.
- Review a balanced sample of automatic matches and non-matches.
- Check whether identifier formats, domains, or provider fields have changed.
- Update jurisdiction-specific normalization and blocking rules where needed.
- Test changes against known examples, then release them with a documented version.
- Keep uncertain subsidiaries and shared-domain cases in a review queue rather than forcing a duplicate decision.
The goal is not to make every company record look identical. It is to create an auditable relationship between records while preserving the distinctions that matter. A conservative matching workflow usually produces cleaner data, safer merges, and better decisions than a single aggressive fuzzy-match threshold.