Search Relevance and Taxonomy Debt
A media company's Widen instance has 500,000 assets and inconsistent metadata. Keywords were free-text for years; the same concept appears as "Q3-launch," "q3 launch," "Q3 Launch," and "third quarter launch." Users complain that search "can't find anything" even for assets they know exist. The DAM team proposes adding a controlled vocabulary for keywords and re-tagging everything. Leadership asks why search is broken and whether re-tagging 500,000 assets is worth it. There are 12 power users who could help; the rest of the 300 users will not participate.
1. Explain the root cause of the search failures and assess whether full re-tagging is the right fix.
Read the worked answer
Root cause: free-text keywords with no controlled vocabulary mean the same concept is stored as four different strings. Search matches strings, not meaning, so a user searching "Q3 launch" misses assets tagged "Q3-launch" or "third quarter launch." The assets exist; the metadata is fragmented. This is a taxonomy/data-quality problem, not a search-engine problem - though search relevance settings can help at the margins.
Is full re-tagging right? Not as a first move, and probably not for all 500,000. Full re-tagging is enormous effort, most of it low-value, and 288 of 300 users won't help. Better:
- Normalize the obvious variants programmatically - case and punctuation variants ("Q3-launch" vs "q3 launch" vs "Q3 Launch") can be collapsed by rule. That alone fixes a large share of misses cheaply.
- Introduce a controlled vocabulary going forward so new tagging is consistent, and map the four known variants to one canonical term.
- Prioritize re-tagging by usage - the assets people actually search for and download, not the long tail. Use download/search analytics to find the high-value subset.
- Enlist the 12 power users to curate their areas rather than attempting universal coverage.
So: fix the cheap structural issues first, govern new tagging, and re-tag strategically - not a blanket 500,000-asset project.
2. How would you measure whether the changes are working, and how do you get leadership buy-in for a phased approach?
Read the worked answer
Measurement:
- Zero-result search rate - the clearest signal; it should fall as variants are normalized.
- Search-to-download conversion - are users finding and using assets after searching?
- Top failed search terms - review weekly; they tell you exactly what to normalize or tag next.
- Time-to-find for a sample of known assets (spot-check searches that previously failed).
Baseline all of these before changes so you can prove movement.
Buy-in for a phased approach: leadership is asking "is re-tagging 500,000 assets worth it?" Reframe: the goal is users finding assets, not tagging everything. Present a phased plan with measurable milestones:
- Phase 1 (weeks): rule-based normalization of case/punctuation variants + controlled vocabulary for new tagging. Low cost, fast, measurable drop in zero-result searches.
- Phase 2: power-user curation of high-value areas, prioritized by analytics.
- Phase 3: long-tail cleanup only if metrics justify it.
Show the baseline zero-result rate and the projected improvement from Phase 1 alone. If Phase 1 fixes most misses cheaply, leadership sees ROI without committing to the full project - and you keep the option to expand only where data says it pays.