A zero-result rate is useful only after you understand why it happened. A product may genuinely be unavailable, the user may use a local term your catalogue does not recognise, or a tracking defect may have hidden otherwise valid results.
Start by grouping repeated queries by intent and language. For Malaysian audiences, review English, Bahasa Malaysia, abbreviations, spelling variants, and mixed-language phrases together. Add device type, location at a broad level, and the screen where search began.
Then sample real sessions. Prioritise queries with commercial or task value, not simply those with the largest count. The most useful response might be a synonym, improved content, a new inventory decision, or clearer navigation—not necessarily an algorithm change.