Bank of Korea (Issue Note 2026-12): AI adoption cut work time 3.8% but the productivity gain is near zero

The problem

As generative-AI tools spread through workplaces, the open question is not whether they save individual workers time but whether that saved time becomes measurable economic output (source). The Bank of Korea set out to test exactly that, asking in its June 2026 Issue Note “whether AI adoption generates potential productivity gains through reductions in work time and whether these gains are ultimately translated into realized productivity” (source). The distinction matters because a firm or a government that counts “hours saved” as if it were output would be measuring the wrong thing (source).

What was built

This is a research finding, not a product deployment: the analysis is BOK Issue Note 2026-12, “Does AI Adoption Improve Productivity? Effects Over the First Three Years”, published on 9 June 2026 by Bank of Korea researchers Donghyun Suh, Samil Oh and Jongwon Yoon (source). It uses household survey data to examine the first three years of generative-AI adoption and to separate the time workers save from the output they actually produce (source). Independent trade coverage credited the same three Bank of Korea researchers with the work (source).

The outcome

The Bank of Korea found that “AI adoption reduces average work time by 3.8 percent, equivalent to approximately 1.5 hours per week” (source). The Korean daily Kyunghyang Shinmun reported the same figure, noting that “working hours were cut by about 1 hour and 30 minutes based on a 40-hour workweek, representing a 3.8% average decrease” (source). The study also reports that the effect “is particularly pronounced among lower-skilled workers and intensive AI users” (source), a distribution Central Banking summarised as being “particularly true among lower-skilled workers and those who make heavy use of the technology” (source).

The decisive result is what happened next: those hours did not become output. The Bank of Korea states that “these time savings do not translate into realized productivity: the relationship between time savings and actual output growth is essentially zero” (source). Kyunghyang Shinmun put the same finding in blunter terms, reporting that “the correlation between the two variables was only ‘0’. In other words, shorter working hours through AI did not translate into actual productivity gains” (source), and Central Banking’s own headline was that the time saved by AI “does not increase output” (source). Even on the optimistic assumption that every saved hour were reallocated to production, “the implied potential productivity gain is estimated at approximately 1.0 percent” (source), a figure Kyunghyang Shinmun rendered as “around 1% even if all time savings were devoted to production activities” (source). The Bank of Korea’s overall conclusion is that “AI appears to have entered an efficiency stage but has not yet progressed to a productivity stage” (source).

One figure is not fully consistent across sources: the primary Bank of Korea page and Kyunghyang Shinmun both give the work-time reduction as 3.8 percent, while Central Banking reports it as “3.9%” (source); this page uses the 3.8 percent figure stated in the primary document, and the weakest load-bearing point here is that the corroborating outlets are secondary summaries rather than the full Issue Note PDF, so the regression behind the “essentially zero” correlation is quoted from the Bank of Korea’s own summary and has not been read table-by-table (source).

How this was verified

Method: on 27 August 2026 the Bank of Korea’s English Issue Note page was retrieved live and every quoted figure (the 3.8 percent / 1.5 hours work-time reduction, the “essentially zero” output correlation, the ~1.0 percent implied potential gain, the concentration among lower-skilled and intensive users, and the “efficiency stage / productivity stage” conclusion) was copied verbatim from it, then confirmed against an archived Wayback snapshot of the same page (captured 27 July 2026) whose text matches word for word. The two headline figures were independently corroborated against Central Banking and Kyunghyang Shinmun, both fetched live. This is central-bank first-party research, so it is verified against the primary document itself; no author or institution was contacted, in line with TIN’s independent-audit rule.

Related case files

• JPMorgan’s COIN and its famous 360,000-hours figure, a headline hours-saved number that this study’s efficiency-versus-output gap is precisely about, and that was never independently measured
• The UK government banking £45bn a year on AI and digital efficiencies its own auditor says are not yet evidenced, the same leap from claimed savings to realized output that the Bank of Korea finds does not hold
• The UK DWP’s machine-learning fraud model and its estimated £4.4m saving on Universal Credit advances, an estimated-savings figure to read against how hard realized productivity is to establish

Sources

1. Bank of Korea · “Does AI Adoption Improve Productivity? Effects Over the First Three Years [BOK Issue Note 2026-12]” (Donghyun Suh, Samil Oh, Jongwon Yoon) · 9 June 2026 · https://www.bok.or.kr/eng/bbs/B0000354/view.do?nttId=10098400&menuNo=400409&relate=Y&depth=400409&programType=newsDataEng · Tier 1 (primary, the central bank’s own research summary; archived snapshot 27 July 2026)
2. Central Banking · “Time saved by AI does not increase output – BoK study” · June 2026 · https://www.centralbanking.com/economics/7976103/time-saved-by-ai-does-not-increase-output-bok-study · Tier 2 (independent specialist trade press)
3. Kyunghyang Shinmun (경향신문, English edition) · “After adopting generative AI, ‘weekly working hours cut by 1.5 hours, productivity gains still absent’” · 7 June 2026 · https://www.khan.co.kr/en/article/202606071414017/ · Tier 2 (independent national daily)