A commercial gallery sold original works and limited editions online and in-room. Inventory lived in spreadsheets and image folders named by artist or acquisition date — not by how buyers search.
Staff spent time answering repeat enquiries ("Do you have anything in this style?") because the public site only supported basic filters. The gallery needed:
- Consistent metadata across artists, mediums, and periods
- Search that understood style, theme, mood, and technique — not exact titles
- Fast publishing when new works arrived
- A browse experience that felt curated, not like a file dump
We mapped how curators and visitors already talked about the collection, then designed a lightweight catalogue model and semantic layer:
Phase 1: Catalogue discipline
- Standardised fields for artist, medium, dimensions, price band, and availability
- Bulk import from existing folders with human review on edge cases
- Image pipeline with consistent crops and alt text for accessibility
Phase 2: Semantic categories
- Tagging schema aligned to style, theme, technique, and collection series
- Assisted labelling from titles and descriptions, with curator approval
- Related-work suggestions on each artwork page
Phase 3: NLP search and discovery
- Natural-language search over approved metadata and descriptions
- Featured collections driven by tags instead of manual HTML updates
- Enquiry forms pre-filled with the artwork context the visitor was viewing
Rollout stayed incremental — the in-room experience unchanged while the online catalogue improved week by week.
Faster discovery
Average time-to-relevant artwork on the site dropped sharply; fewer "nothing matched" search sessions.
Staff capacity
Repeat style-and-theme enquiries reduced; curators spend more time on sales conversations and hang planning.
Publishing speed
New works go live in one workflow — metadata, images, and related links — instead of ad hoc page edits.
Visitor engagement
More pages per session and longer on-site time on collection and artwork routes after semantic browse launched.
Structured catalogue database, semantic tagging layer, NLP-backed search index, image CDN, CMS integration, analytics on search queries and zero-result terms.
Lessons learned
Creative retail benefits when search vocabulary matches how people talk, not how files were stored. Curator sign-off on tags kept quality high; automation accelerated labelling but did not replace judgment. Measuring zero-result searches surfaced gaps in metadata early.
Related resources
All case studies