Hong Kong's Digital Policy Office and the HKGAI research centre built HKPilot (document drafting/translation) and HKMeeting (AI minute-taking) for civil servants: participation grew from 800 officers in 20 bureaux (Nov 2024) to roughly 50,000 across 100 departments by 2026, thoug
800 officers across 20+ bureaux/departments
Civil servants participating in the pilot (November 2024)
~50,000 officers across ~100 departments
Civil servants participating in the pilot (March 2026 (as reported))
235 HK$ million
Disclosed expenditure on HKGAI centre (cumulative; covers HKGAI centre incl. HKChat and infrastructure, not isolated to HKPilot/HKMeeting)
Details
Maturity
Scaling
Promoter
Digital Policy Office, Government of the Hong Kong SAR, with the Hong Kong Generative AI Research and Development Center (HKGAI)
Period
2024–2026 (pilot, ongoing)
Keywords
generative AI, civil service productivity, document drafting, meeting transcription, digital government
Context
Hong Kong's Digital Policy Office, formed in 2025 from the merger of the Office of the Government Chief Information Officer and the Efficiency Office, has piloted two internal generative-AI tools for civil servants since 2024: HKPilot, for drafting, translating and summarising official documents in Hong Kong's bilingual Chinese/English legal-administrative register, and HKMeeting, for AI-generated meeting minutes. Both were developed with the Hong Kong Generative AI Research and Development Center (HKGAI), an InnoHK-funded lab established in October 2023 by five Hong Kong universities led by HKUST. Neither tool is citizen-facing; they are separate from the government's public chatbot, HKChat.
Objectives
To reduce the administrative burden of document drafting, translation, summarisation and minute-taking for civil servants using a large language model tuned for Hong Kong's bilingual legal-administrative register.
Activities
The pilot began with around 800 officers across more than 20 bureaux and departments in November 2024, expanding to over 70 departments by April 2025, according to written replies to the Legislative Council.
Results
Government-cited figures reported by news outlets in March 2026 put participation at close to 50,000 civil servants across roughly 100 departments. Total disclosed expenditure on the HKGAI centre — covering model development, HKChat and infrastructure, not isolated to HKPilot/HKMeeting alone — is about HK$235 million. However, officials have told the Legislative Council directly that no data exists yet on time savings, cost reduction or efficiency gains from the pilot: adoption has scaled quickly, but outcome measurement has not kept pace.
Conclusions
The case illustrates rapid internal-tool adoption within a civil service outrunning the government's own ability to measure whether that adoption is delivering efficiency gains.
Implementation
Indicative cost
High (€500k–€5M) — Disclosed cumulative expenditure on the HKGAI centre (model development, HKChat and infrastructure) is about HK$235 million; this figure is not isolated to HKPilot/HKMeeting specifically, so 'high' is a conservative reading pending a tool-specific cost breakdown.
Time to results
Medium (1–3 years) — Pilot running since November 2024 and still expanding as of the most recent 2026 reporting — over a year of active rollout with no stated end date.
Staffing & skills
Digital Policy Office, Government of the Hong Kong SAR (formed 2025 from merger of GCIO and Efficiency Office), Hong Kong Generative AI Research and Development Center (HKGAI), an InnoHK-funded lab run by a five-university consortium led by HKUST
Conditions for success
A dedicated, purpose-built LLM tuned for Hong Kong's bilingual (Chinese/English) legal-administrative register, rather than a generic off-the-shelf model
University-government research partnership (HKGAI) supplying the underlying model
Rapid cross-departmental rollout backed by Legislative Council reporting and political visibility
Common failure modes
Officials have confirmed to the Legislative Council that no time-savings, cost-reduction or efficiency data exists yet, despite scaling to tens of thousands of users
Total disclosed spend (HK$235m) is bundled across the whole HKGAI centre, so tool-specific cost-effectiveness cannot be assessed
Adoption scale has outpaced outcome measurement, creating a risk of continued expansion without evidence of impact
Where it fits
Governance type
territorial/special administrative region civil service
Scale
territory-wide civil service (multi-department)
Income level
high income
Data sources
Where this practice's information was retrieved from, and when.
