Vendor reportedLive since Oct 2023Added Oct 2026

City of Fukuoka

Fukuoka City trials generative AI for document drafting; staff survey reports a 33% cut in work hours

Fukuoka City, Japan, tested a generative AI service called QT-GenAI with city employees from October 2023. The vendor, AndDot, reported in March 2024 that an interim staff survey showed a 33% cut in work hours. The city's final report was more modest.

Key figures

Copied exactly from the sources. Copy any figure with its citation.

33%

Average reduction in work hours reported by city staff in an interim survey (February 2024), as stated by the vendor1

40

minutes

Time to create a document that normally took one hour, with AI assistance (vendor press release)1

96%

Trial participants who wanted to keep using the tool (city final report, trial to 31 March 2024)2

more than half

Participants in the city's final report who felt a 10% to 30% cut in work time2

What it took

CostNot disclosed
TeamNot disclosed
TimelineNot disclosed

Cost, team size and timeline were not disclosed in the sources.

The problem

City halls spend a large share of staff time writing: reports, notices and promotional material. Fukuoka City wanted to know whether generative AI, software that drafts text from a short instruction, could cut that load for its employees.

What they built

The city tested QT-GenAI, a generative AI service developed by the startup AndDot together with QTnet, a subsidiary of Kyushu Electric Power. It has ready-made modes: staff type a few keywords into a form instead of writing a full instruction each time. Staff used it for drafting proposals and promotional material, proofreading, translation, FAQ writing, summarising meeting minutes and generating ideas.

How they did it

The trial ran from October 2023 to the end of March 2024 under a city programme that supports company experiments. The vendor says about 40 employees took part; the city's final report counts 50, drawn from two departments plus staff interested in the technology. AndDot ran two training sessions. Results were gathered through staff surveys and usage logs. The sources do not name the language model.

Results

AndDot reported that an interim survey in February 2024 gave an average 33% reduction in work hours, with a one-hour document taking about 40 minutes. The city's final report was more cautious: more than half of participants felt a cut of 10% to 30%, some felt 50%, all reported better quality of work and 96% wanted to keep using the tool. Logs showed about 19.5 uses a day on average, leaning on a few active users.

Lessons for your project

  1. Start with a small, named group of staff and a narrow set of tasks such as drafting documents and promotional material, so the time saved can be compared with a known baseline.

  2. Treat early survey numbers as a first signal. Here the interim average (33%) was higher than the final picture (more than half of staff saw 10% to 30%), so wait for the end-of-trial figures before quoting a saving.

  3. Watch for uneven use. The final report found use concentrated among a few active users and called for ways to keep others engaged.

  4. Pair the tool with staff training. The vendor ran two training sessions on generative AI.

Caveats

  • The 33% comes from AndDot, the vendor, in a press release dated 11 March 2024. It is an average of answers in an interim staff survey (February 2024), not a measured time study. The press release states both 'more than 33%' and 33.75%.
  • Fukuoka Now, a news summary of the press release, says the city announced the result. The press release itself is from AndDot.
  • The city's final report (trial from 1 October 2023 to 31 March 2024, 50 staff) is more modest: more than half of participants felt a 10% to 30% cut in work time and a few felt 50%. It does not repeat the 33% figure. The report carries no date.
  • Participant counts differ: about 40 in the press release and Fukuoka Now, 50 in the city's final report.
  • The language model behind QT-GenAI is not named in any source.
  • Fukuoka Now says the city is considering full-scale use; the press release and the final report do not say so. No later rollout result was found.

Sources

  1. Fukuoka City trial of generative AI shows average 33% work reduction (press release, in Japanese)

    AndDot Inc. via PR TIMES · 2024-03-11 · Press release · prtimes.jp

  2. Implementation report: Possibilities of generative AI from Fukuoka City (in Japanese)

    City of Fukuoka · Undated · Filing · city.fukuoka.lg.jp

  3. Fukuoka City Cuts Work Hours by 33% with Generative AI Trial

    Fukuoka Now · 2024-03-12 · News · fukuoka-now.com

Summarized in our own words from the sources above. Spotted an error? See corrections.