Collect real queries
Ask users for examples of what they remember when looking for a record. They may know an exact reference, part of a name, a date range, or a phrase from a note. Each clue suggests a different search behavior.
Start with the queries that occur frequently or block important work. Exact identifier lookup may matter more than sophisticated text ranking. Document what should happen when punctuation, capitalization, or a common abbreviation differs from the stored value.
Make results distinguishable
Show enough information to identify the correct record without opening every result. A name alone may be ambiguous; a location, owner, status, or date can provide useful context.
Apply access rules to the results and to the details that appear in them. Do not let a search preview reveal content that the user could not otherwise read. Explain how filters affect the result set and keep the current query visible.
Evaluate with a small query set
Build a collection of representative queries and expected useful results. Include misspellings, absent records, duplicate names, and records the current user cannot access. Use it to evaluate changes rather than relying on one impressive demonstration.
Observe what people do after an unsuccessful search. Repeatedly changing a query may indicate a vocabulary mismatch; repeatedly opening the wrong result may indicate poor result summaries.
Search quality includes the experience of finding nothing. A clear explanation and a way to adjust filters can be more useful than an unexplained empty list. Improve the matching behavior that the actual workload needs before adding complexity that is difficult to evaluate.
A practical example.
An operator looking for request “SR-1042” has a different need from someone browsing every request about a broken door. Exact identifiers should be easy to find, while descriptive searches may need to match several fields and common terminology.
Create a small evaluation set from real, suitably sanitized search tasks. Include an identifier, a misspelling, a person’s name, and a term that appears only in a description. Decide what a useful first result looks like for each. Then check permissions: a result count or snippet should not expose records the user cannot open. Keep filters visible and removable, because a correct search can still appear broken when an old filter silently excludes the expected result.
Put it into practice.
- Test known-item searches separately from exploratory searches.
- Explain which filters are active and how to remove them.
- Check result titles, snippets, and counts against the same access rules as records.