This is quite a good report (55 page PDF) and worth reading in full. It came to my attention just today, though it's dated 2018. It's an outcome of the Beyond LMS project run by several universities in Australia. It doesn't mince words, and while it finds that learning analytics can be effective, they are brittle, context-specific, serve the wrong users, prone to misinterpretation and the subject of conflicting specifications. So, business as usual in the learning technology standards space. As the authors state, a more modular solution is necessary, a more flexible data structure is necessary, and a more open codebase is required (note that the report contains links to numerous GitHub repositories containing their own work, so kudos to them).