This can happen when there are gaps in instrumentation of the agent. For instance, if there is a host or process downstream that is slowing things down, but there is not a OneAgent on that host, Dynatrace would not be able to correlate the root cause back to the specific issue. Do you have any specific examples you can provide? Or know of any gaps in implementation?
Further, understand that the AI is not always going to be able to identify a root cause for any number of reasons and instead of showing a guess as to a root cause it doesn't display one and instead directs you to the relevant areas so you can investigate further.