Even if a company introduces AI and reduces working hours, why do employees often not feel that work has become easier? An analysis said the causes lie not in AI performance itself but in faulty performance measurement, workflow bottlenecks and organisational management that is unclear about where the saved time goes.

On Sept. 3, engineering management head Go Hisamatsu (久松剛) wrote in a contribution posted on Japanese outlet ITmedia that the effect of introducing AI has not disappeared but is being absorbed into other work within organisations. If a company touts only the numbers it can check afterward as performance without measuring working time before adopting AI, it is difficult to judge the actual effect, he said. In cases Hisamatsu encountered, some saved 6 hours a year or shortened a task that took 15 days once a year to 15 minutes, but the perceived benefit may be limited when development costs, time投入 and how often the work occurs are considered.

Where the time saved goes also matters. Even if AI cuts 30 minutes a day, other work fills the gap if there is no decision to use it to reduce overtime or for new business. The field treats how many hours were cut as the result, but a difference in evaluation standards is also a problem when management demands improvements in costs, staffing and profits. If headcount does not fall and overtime stays the same, the saved time does not immediately show up in profit and loss.

A shift in bottlenecks also appears. Even if AI speeds up a frontline worker's pace, output simply piles up faster if review staff and approval procedures remain unchanged. In particular, if people must verify AI-generated output, the burden on senior employees could increase instead. Individual productivity rises but the organisation's overall processing speed does not improve.

Costs also need to be weighed when using AI to replace existing SaaS with in-house tools. Licence fees may disappear, but new work can arise such as security, responses to legal and institutional changes and maintenance. AI usage guidelines and log management, shadow AI detection, responses to information leaks and token-cost management are also tasks that did not exist in the past.

To feel AI's performance gains, companies need to measure workload and goals before adoption and decide where to redeploy saved staff and time. The key is not how much time AI cut but where the time secured that way was used, the analysis said.