Three Years After ChatGPT Launch: Weekly Work Hours Down by 1.5, But Did Real Output Increase?

Generative AI Use Cuts Average Work Hours by 3.8%
Potential Productivity Gains of About 1.0 Percentage Point
However, First Three Years Show Time Savings Did Not Lead to Output Growth
Need for Work Process Redesign and Performance-Based Incentive Systems

The use of generative artificial intelligence (AI), including ChatGPT, has been found to reduce average work hours by 3.8%. Based on a 40-hour work week, this translates to about 1.5 hours saved per week. When this reduction in work hours is converted to productivity gains, it is estimated to represent a potential productivity improvement of about 1.0 percentage point. However, it was found that such time savings have not actually led to an increase in real output. The analysis suggests that a redesign of work processes and organizational structures, job reassignment, and the establishment of performance-based incentive systems are necessary.

Generative AI Use Cuts Average Work Hours by 3.8%

The Bank of Korea presented this diagnosis in its publication on June 7, "BOK Issue Note – Does AI Adoption Raise Productivity? Analysis of the Initial Three Years' Impact" (Donghyun Seo, Samil Oh, and Jongwon Yoon). The report empirically analyzed whether the use of AI, based on household surveys, leads to potential productivity improvements through reduced work hours, and whether time savings result in actual output growth.

The most direct indicator of productivity improvement is completing the same work in a shorter amount of time. Samil Oh, Head of the Employment Research Team at the Bank of Korea's Economic Research Department, explained, "A quantitative estimate of the change in work hours following the introduction of generative AI shows that the average work hours for employees using generative AI decreased by 3.8%. This corresponds to a reduction of about 1.5 hours per week, based on a 40-hour work week."

If we assume that all of these reduced work hours are directly converted into increased output, the potential productivity gain during the study period is estimated at around 1.0 percentage point. From the fourth quarter of 2022, when ChatGPT was released, to the second quarter of last year, Korea's gross domestic product (GDP) grew by 3.9%. If all the time saved by using generative AI had been used for production growth, GDP could have risen to 4.9%. Oh pointed out, "This is a significant figure compared to existing estimates of the productivity effect of AI, and it shows that even with the current level of AI technology, a substantial productivity increase is possible."

The effects varied among workers. By occupation, the time savings were most pronounced for professionals, office workers, and managers, in that order, while the effects were relatively limited for service, skilled, and simple labor positions. By task, the time-saving effect was notable for cognitive and non-routine work such as developing educational materials, statistical analysis, model design, and software development. On the other hand, for tasks that require high-context judgment or physical collaboration, such as workflow coordination and equipment operation, the effect was limited.

First Three Years: Time Savings Did Not Lead to Output Growth ... 'Productivity Disconnection'

However, these time savings did not actually translate into increased output. When analyzing the relationship between the work hour reduction rate from generative AI use and the rate of increase in work output at the individual employee level, the correlation coefficient between the two variables was zero. Oh interpreted this as, "AI has increased efficiency at the individual task level, but since this has not expanded to workflow improvement, organizational restructuring, or personnel reassignment, a phenomenon known as 'productivity disconnection' is occurring."

This outcome is attributed to the following: ▲ Current AI use is selectively applied to specific tasks rather than to the entire work process; ▲ The introduction of AI has not led to changes in work processes or organizational structures; ▲ Productivity may be determined by the most constrained stage in the production process (the bottleneck), rather than the average efficiency of individual tasks; ▲ Additional performance incentives are weak. Exceptionally, productivity gains were observed among groups with high performance incentives and work autonomy, such as the self-employed, professionals, and intensive AI users. Oh assessed this as "evidence that the effect of AI may be determined more by work structure and incentive systems than by the technology itself."

It was explained that the currently observed productivity disconnection can be interpreted as a typical part of the transition process in the early stages of adopting a general-purpose technology. Oh emphasized, "In the future, the economic impact of AI will depend not only on the level of technology but also on how it is used and how organizational structures are changed. Accordingly, policy should focus not just on spreading the technology, but on supporting the 'transition process' so that AI use actually leads to productivity gains."

Need for Work Process Redesign and Performance-Based Incentive Systems

The analysis first suggests that standardized work and open work should be distinguished. To realize productivity gains, changes in internal work structures and organizational management are more important than the mere adoption of technology. Oh stressed that, "Organizational restructuring can start by dividing work into 'standardized work,' where outputs and evaluation criteria are relatively clearly defined, and 'open work,' where the form and level of the final output are not predetermined, and the experience, judgment, and creativity of the performer are important."

In standardized work areas, it is necessary to redesign the workflow so that AI takes a central role in performing tasks, while human roles should be reorganized to focus on setting work objectives, verifying results, and responding to exceptions. He emphasized, "The important thing is to establish a mechanism at the organizational level to ensure that the time saved by AI is actually redirected into increased output."

For open work areas, where even the same problem can yield different results depending on the performer's experience and value judgments, AI should ideally be used as an augmentation tool to support human thinking and decision-making, such as idea exploration, drafting, information gathering, and presenting alternatives. Oh pointed out, "If AI is introduced in the same way without distinguishing between standardized and open work, it is highly likely that only improvements in the efficiency of individual tasks will accumulate, without fundamental changes to the workflow."

The analysis also suggests that it is important to design routes for new or junior staff to participate in open work early on and to reinvest the time saved by experienced workers through AI into mentoring, coaching, and pair work. Oh emphasized, "What matters in the future is not just the introduction of AI itself, but continuously examining how AI use actually connects to productivity changes. In particular, it is necessary to systematically monitor how the relationships among time savings, work reassignment, and output growth differ by industry, job, and company characteristics."

This content was produced with the assistance of AI translation services.