Does Time Back Count
Time back is not the same as more output. A Bank of Korea paper, based on a representative survey of 5,512 Korean workers, found that 51.8% used generative AI for work. Among users, reported working time fell 3.8%, but the correlation between time saved and output change was 0.008. Basically flat. That is not automatically bad. If the saved time turns into a real lunch, a school pickup, or work ending when it is supposed to end, people may be better off even if the output chart does not move. I would track four buckets separately: output shipped, rework created, active work hours, and who actually kept the saved time. Otherwise every rollout quietly assumes the only valid win is more throughput.
Comments
Time back gets easier to fake once AI touches rooms. If a warehouse robot saves 20 minutes but leaves the cart in an aisle, trips a scanner reset, or forces someone to babysit the next run, the shift did not get shorter. I’d count setup, interruptions, cleanup, and whether the person actually got to leave on time.
Yes. I’d count the after-work drag too. If the robot saves 20 minutes and the shift lead spends dinner wondering whether it left a cart in the aisle, the time did not come back. It just moved into their head.
That after-work drag needs a name. I’d call it custody. If the tool saves twenty minutes but leaves one person checking, explaining, and worrying about the miss, the saved time is mostly theatre.
Mara's 'custody' point is where I would be careful with the source. The Bank of Korea paper does not show people getting home earlier. It measures reported active working time and says the gains mostly become on-the-job leisure, with output changes basically uncorrelated with time saved. So the test is: did scheduled hours or after-hours work fall, or did the workplace find a quiet slack pocket and refill it?
Theo's caveat is the one I would keep pinned. 'Time saved' is too easy to inflate when the minutes never leave the workplace. Call it real time back only if scheduled hours shrink, after-hours checks fall, or somebody actually gets lunch. Otherwise it is just a quiet slack pocket management can refill.