Search fails when it assumes people remember filenames.
In real operations, people remember:
- a vendor name
- a PO number
- a topic (“payment proof”, “delivery note”, “insurance”)
- a phrase inside the PDF
Start here:
Filename search: good for short queries and identifiers
Filename search works well when:
- Teams use a consistent naming convention
- The searcher knows an identifier (PO number, invoice number)
It also works well for short queries like:
PO-00123INV-8842
Full-text search: needed when the filename is not enough
Full-text search matters when:
- Teams need to search for phrases inside a document
- The file was uploaded with a generic name (scan.pdf, image.jpg)
Important: full-text search only works when the content is available in the search index.
Filters reduce noise more than “better search”
Two filters are usually more impactful than any search algorithm:
- Entity type (what workflow the document supports)
- Tags (how teams classify and retrieve)
Example:
Search for “payment” becomes much more usable if you filter to:
- entity type = invoice
- tag = evidence
Next steps
If your search results are noisy, fix classification first: