I've been using Scheduler since very early in its preview. As much as it has saved me time and effort there's no way I'd pay an additional $10/user/month for it. Why not? The NLP still needs a bit more debugging (I think they fixed the confusion between "Jan" as a name and not just an abbreviation for "January" but that was a PITA), and it is far too quick to give up when it sees what it thinks as a conflict between calendars in widespread time zones. I still have a recurring problem with recurring meetings (scheduled instance-by-instance) that it will absolutely NOT turn into a Teams meeting no matter how many times I tell it to do so.
For free, I'd turn it on for every user, no questions asked. For something like $2/user/month, I'd recommend it for my key users. At 50% the monthly cost of an E3? No way. My users will burn more than $10 of time per month trying to figure out how to enable DWIM (do-what-I-mean) mode.