But it's not just anticipating demands by learning past behaviour. There's also the problem in any multi-battery system (which is what they cover), let alone with multiple types, optimizing what should take charge or load at a given time. Most people don't realize that many battery chemistries actually don't have a flat efficiency rating with charge: often, the closer to 100% charged you get, the less efficient (more loss) you have compared to what you'll be able to extract later.
For example, charging from 0 to 100% might be 85% efficient for a given battery type. But charging it from 80% to 100% full might only be 40% efficient.
And this variable efficiency also applies to discharge as well. And the variables all change with what recent battery demands have been, let alone the current loads - but also cell voltages, temperature, age and cycle count.
Even planning to cope with self-discharge over days, weeks or months might benefit from smarter BMS. Some battery management systems even take into consideration thermal management (taking into account the cost of ramping up active cooling or throttling charge rates to keep batteries at a temperature efficient for taking charge).
With that said I don't think the concept itself is bad, its just that the execution will inevitably fail except under very regular circumstances such as carrying your laptop back and forth to the office each day. Volvo trucks has a similar thing to save fuel for their gearboxes that automatically chooses gear based on the performance you got the last time you drove that very route.