I'd contest your claim that MOOCs "don't work".
It's true that only a fraction of those who participate in typical MOOCs complete them (8 - 10%), (https://imgur.com/a/HI4IXbt , from "HarvardX and MITx: Four Years of Open Online Courses", https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2889436), but a significant fraction (~15-20%) also explore these courses (i.e., they watch 50% or more of lectures). So this means that the MOOCs are reaching around 20-30% of the participants in some significant manner.
First of all, is the larger drop out rates really a problem? I'd argue that the option to drop out of a course if you feel that it doesn't serve your needs right now is itself valuable. This means that you're free to spend your time in ways that are more useful to you. Also, a large number of those taking the MOOCs are themselves teachers. This means that the high quality of pedagogy that MOOCs can deliver will have cascading effects through educators who use them to upgrade their own skills.
How to improve the effectiveness of MOOCs? There are plenty of things we can do, too many in fact to go into in one post. A few keys areas that I can see are (a) pedagogy (finding great teaching talent and professionally beefing up their performance), (b) affordability (an issue in developing countries primarily), (c) formal accreditation, and (d) a wider selection of courses that form a genuine alternative to the standard university programs. Models that I find incredibly exciting are that of India's IGNOU (https://en.wikipedia.org/wiki/Indira_Gandhi_National_Open_Un...), NPTEL (https://nptel.ac.in/), and Swayam (https://swayam.gov.in/), all of which issue officially recognized certificates. But unfortunately, as with many things that the Government of India does, the intentions are noble but the execution is mediocre. There's clearly room for players who will "full-ass" what IGNOU set out to do.