Singapore's Digital Records Crisis: Decade of Unchecked Growth Demands Urgent Action

Singapore's public agencies are sitting on a quiet but growing data problem. Across government ministries, statutory boards, and the digital portals that millions of residents use every day, duplicate and degraded images, scanned identity documents, property photographs, urban planning maps, and archival records, have accumulated over years of rushed digitisation, creating bloated databases that slow down services and compromise record integrity.

The issue matters now because Singapore is in the middle of repositioning itself as a regional artificial intelligence hub. Clean, non-redundant data is the foundation that machine-learning systems require to function reliably. Feeding duplicated or corrupted image files into AI-assisted workflows, from HDB flat-condition assessments to the National Library Board's digital archive, produces compounding errors that grow harder and more expensive to correct the longer they sit.

How the Backlog Built Up

The roots go back to the early 2010s, when the Smart Nation initiative began pushing agencies to digitise paper records at scale and speed. The priority at the time was volume, getting physical files off shelves and onto servers. The Urban Redevelopment Authority's digitisation of planning submissions, the National Archives of Singapore's conversion of photographic collections, and the Housing and Development Board's shift to online flat-inspection records all proceeded largely without a unified image-quality protocol.

Each agency built its own repository. Files were uploaded multiple times across departments, a single photograph of a Toa Payoh block, for example, might exist as three separate scans at different resolutions in different folders, none flagged as redundant. A 2024 Government Technology Agency review of cross-agency data quality, cited in Parliament, identified image duplication as among the top-three causes of storage inefficiency in the public sector's central data infrastructure. The GovTech review did not publish a precise duplicate-image count, but the scale of the problem was described as affecting records stretching back to the early 2000s.

The National Library Board's NewspaperSG and BookSG portals, both housed at the library's Victoria Street headquarters, have faced user complaints about misfiled and duplicate digitised pages for years. The Singapore Land Authority's survey maps, accessible via its OneMap platform, similarly carry layered scan histories where boundary photographs were re-uploaded during system migrations in 2016 and again in 2021.

The Cost of Inaction

Storage is not free. Government cloud expenditure in Singapore has grown sharply as agencies moved to the Government Commercial Cloud framework. Duplicate images do not merely consume disk space passively, they inflate search-index sizes, slow query times, and degrade the training datasets that GovTech and agency partners are now using to build AI tools for services like HDB's flat-valuation assistant and the Immigration and Checkpoints Authority's document-verification pipeline at Woodlands and Tuas checkpoints.

Private-sector firms handling government contracts face the same problem from the other direction. Technology companies working out of one-north's Fusionopolis complex on document-processing contracts have found that client datasets routinely contain fifteen to twenty percent duplicate image records, based on figures discussed at industry forums held by the Singapore Computer Society in 2025. Cleaning those datasets before model training can add weeks to project timelines and meaningful cost to fixed-price contracts.

The practical consequence for ordinary residents is slower, occasionally inconsistent service. When HDB's e-services portal retrieves a flat's condition photographs for a resale application, duplicate records in the backend can generate conflicting timestamps, a flat appearing to have been inspected twice on different dates when it was inspected once and the file duplicated on upload.

GovTech has signalled that a whole-of-government image-deduplication and quality-standardisation framework is among the data governance priorities earmarked for implementation before the end of 2026. Agencies are expected to adopt a common hash-based image fingerprinting standard that will flag duplicates automatically at the point of upload rather than requiring manual retrospective audits. For residents and businesses that interact with government digital services daily, the clearest signal of progress will be simpler: faster load times, more consistent records, and fewer cases where the system shows two different photographs of the same flat and asks a clerk to decide which one is real.