Legal Services Offshore research · legal-support-governance

How consistent are conflict search results across different name spellings?

A reproducibility study for conflict searches that compares result sets across name variants and separates a database gap from a judgment call.

Published · 4 sources

How consistent are conflict search results across different name spellings? research illustration

Research context

This research brief is designed for legal operations planning. Use the findings to ask better workflow questions. Test any proposed process against the firm’s jurisdiction, professional duties, systems, and client commitments. Treat every local result as a starting point for review rather than a universal benchmark.

Question and scope

This study asks how much the result set of a conflict search changes when the same party is searched under different name spellings. It treats the result set as an administrative output and the name variant as the input. The scope is one firm, one database set, and a fixed list of test names. The result describes that setting and should not be read as a measure of whether a conflict exists.

The unit is one party name and one variant. A variant is eligible when it is a plausible spelling of the same party, such as a maiden name, a trade name, or a common transliteration. Variants that refer to a different person are excluded. Record the rule for deciding that two names refer to the same party before running the searches.

The study also records whether the search used a fuzzy-match setting or an exact-match setting. A fuzzy search can return more records but may also add noise that a reviewer has to clear. Report the setting with the query. A difference between two searches may be a setting difference rather than a name difference.

Definitions and evidence standard

Define a result set as the list of records returned by the database for a given query, after the same filters are applied. Record the database, the query string, the date, and the filters. A result set that cannot be reproduced from those fields is not evidence. Do not compare a broad search with a narrow one without recording the filter difference.

Define overlap as the share of records that appear in both result sets after matching on a stable identifier. If the database does not provide an identifier, use a combination of name, address, and record type. Record the matching rule and a sample of the matches. A name-only match is too weak to support an overlap claim.

Sampling and measurement

Choose test names from closed intake records, not from live matters. Remove client identifiers and use a code for each party. Include names with hyphens, middle names, suffixes, and non-English characters. Freeze the variant list before searching so the study does not chase whichever variant produces an interesting result.

Measure the number of records per variant, the overlap between variants, and the number of records that appear only under one spelling. Report the database and the date for each query. Add the number of queries that returned an error or a timeout. A result set that is missing because of a technical failure is not the same as an empty result.

Include at least one variant that is a known alias from a public business registry when one is available. That gives the study a realistic test of how the database handles an official alternate name. If no registry alias exists, record that the variant is a constructed spelling. Do not present a constructed variant as an official name.

Findings and interpretation

Present the overlap before suggesting a cause. A low overlap between a legal name and a trade name may reflect how the database indexes companies. A low overlap between two transliterations may reflect a normalization rule. A high overlap with one variant and a low overlap with another may point to a data-entry habit rather than a search rule.

Read a sample of the unique records under each variant. Confirm whether they refer to the same party or to a different one. Do not assume that a unique record is a missed conflict. The study measures result-set variation, not the presence of a conflict. The finding should suggest a search rule or a data-quality check to test.

Look at the records that appear under only one variant and classify them as same party, different party, or unclear. A unique record may be a genuine additional match or an unrelated person with a similar name. Report the classification counts with the overlap rate. The classification is what makes the overlap meaningful.

Limitations

The study cannot see records that are not in the database, and it cannot decide whether a name variant is legally the same party. It also cannot measure whether a reviewer would have found the record by another route. Those questions need a different method and a lawyer's judgment. State that boundary in the methods note so the result is not stretched.

The database may change during the window, and a query repeated later may return a different set. The test names may not represent the firm's full intake mix. If the firm uses more than one database, the results may not be comparable across systems. Record these limits with the result and keep the query log with the sample.

Practical next step

Choose one control that matches the most common variation, such as a required variant list, a normalization rule, or a second database check for company names. Write the change, the owner, and the date it starts. Define the measure that will show whether it helped, using the same overlap definition. Test it on the next set of test names.

Keep the original and revised query logs separate so the comparison is honest. Record the names that were excluded and why. Review the result with the intake team and the lawyer who decides conflicts. If the change does not help, record that result too. The purpose is to learn which search step makes the result set more reproducible.

Pilot the variant list on a small set of names before the full run. Check whether the list produces queries that the database cannot handle, such as names with punctuation or very short strings. Revise the list and record the revision. A variant that fails technically is not evidence about name variation.

Comparing entity and individual names

Company names and individual names behave differently in search. A company may appear under a trade name, a former name, or a parent entity, while an individual may appear under a maiden name or a suffix. Report the results separately so one type does not hide the other. A rule that works for a person may be wrong for a company.

If company searches show more variation, check whether the database indexes the registered name and the trade name in the same field. If individual searches show more variation, check the transliteration and suffix rules. Compare the process, not the people. The next study can test a type-specific variant list before applying it to all intake.

What the evidence does not show

The study does not establish that a conflict exists or that the firm should decline a matter. It also does not measure whether a name variant is legally the same party. Those are decisions for the authorized lawyer. A result-set difference is a search observation, not a conflict conclusion.

Do not use the result to promise a client that every conflict will be found or that no name variant will ever be missed. Do not publish party names or matter details. Keep the sample size, the matching rule, and the query log with the result. If the firm needs a legal conclusion, it should ask counsel rather than rely on a search report.

Keep the query log in a restricted file because it may contain party names and matter references. Publish only the method and the aggregate overlap. If a client asks for the result, describe the search process rather than disclosing another party's information. The study is an internal quality tool, not a client-facing conflict report.

Using the result in a client conversation

A client may ask how the firm screens for conflicts, and the study supports a careful answer. Describe the name variants tested, the databases used, the overlap observed, and the check that was added. Do not say that the firm will find every conflict or that no name variant will ever be missed. Explain that the screen is a documented search and that the lawyer makes the decision.

If the client asks for a commitment, offer a process commitment such as a variant list, a query log, and a lawyer review. Record the commitment and the measure that will show whether it was met. Keep client-specific details out of any public write-up. The study is an internal search-quality tool, not a guarantee about any particular matter.

Sources and method

Open each source to confirm its scope, publication date, and applicability before relying on it for a legal workflow decision.

  1. U.S. Courts, Federal Rules of Civil Procedure
  2. American Bar Association, Model Rules of Professional Conduct
  3. GAO, Standards for Internal Control in the Federal Government
  4. NIST Cybersecurity Framework 2.0

Related research

Bring the messy task list

We will help turn it into one clear legal support role.

Share the work, tools, hours, and review needs. The next step is a practical staffing scope, not a public rate card.