The museum held 94,000 catalogued objects and displayed fewer than 700 of them. The rest lived in a records system built in 2004 that only three members of staff could query, and every public image request went through a single curator by email. The board had approved digitisation twice and stalled twice, because nobody could say what the work would cost or when the public would see anything at all.
We started with the records rather than the interface. Six weeks of cleaning reconciled duplicate accession numbers, recovered dates from free-text notes and flagged the 3,100 records too thin to publish. Only then did we design a browse experience around what the collection actually is: a small number of very deep holdings, and a long tail of single objects that rewards wandering rather than searching.
The public catalogue launched one department at a time, so the curatorial team could keep pace with checking. Image requests now resolve through a self-serve licence flow, which returned about a day a week to the curator who used to answer them by hand. The museum has since published two exhibitions built entirely from objects that had never left storage.