Both questions would deserve a book each to really answer, but I will try.

How are you defining mass parties? Relatively large participatory base, strategy decided democratically, presence on the local territory and ties with communities. Here though I was more framing them as "parties designed for a mass society", where their strategy relies on the possibility to reduce the individual to mass, as in the case of workers parties. A one-size-fits-all organization, where one strategy, one identity and one theory of change is shared by millions of people.

When did they stop working, and why?

There are at least two big elements: the first is the end of mass society. Once we became all individuals, the mechanism of identification in a collective entity became harder. It got even harder over time, when most young people have no examples or memory of anybody around them ever acting collectively.

The second element is informational: mass parties are incredibly slow. The analysis-synthesis-action-assessment most ML parties are based on is predicated on the assumption that the social and political phenomena you're working with don't change too fast and between the analysis phase and the action phase, the underlying phenomenon is relatively stable. If the analysis is too slow or the phenomenon (i.e. specific industries, specific political landscapes, etc etc) change too fast, your analysis is always late. Correct, but useless. This renders anybody involved in such ecosystems (not just mass parties), very aware of the motivations of their own failure, but completely incapable of escaping them.