The latest BCG report I saw on this was the one from this summer that found that employees are saying they're saving time, but businesses still haven't figured out how to turn that into dollars.
And employees self-reporting that they saved time doesn't necessarily mean they actually save time. Humans are famously crap at estimating that sort of thing.
The same report said 67% of regular AI users have increased job satisfaction, so that has value. Even from an entirely capitalist mindset you cannot argue that increased satisfaction = increased wellbeing = less time on sick leave = increased productivity.
I'd want to know a lot more about how they get that number before trying to extrapolate it to a general business trend. There could be a huge self-selection bias there. Perhaps people who enjoy using AI are significantly more likely to become regular AI users. In which case, what that statistic really says might only be, "People who like a thing report that they enjoy using it."
You also have to be careful about the Hawthorne Effect. It's quite common for workplace interventions to produce improved productivity measures that are illusory or fleeting because what workers are really responding to is the novelty of the change or the simple fact that they know they're being observed.
Which doesn't necessarily mean that the benefits people are seeing right now aren't real. Just that, again, it may not translate to a sustained financial benefit that would justify a meaningful contribution to a global $6T/year rate of AI spending.
And employees self-reporting that they saved time doesn't necessarily mean they actually save time. Humans are famously crap at estimating that sort of thing.