The trouble is that the definition of PII could be interpreted so broadly as to include almost anything useful ever logged on a server, because if a log record references any data that could be associated with a specific individual, including in combination with other data, then it counts. Given what we already know about de-anonymisation of supposedly anonymous data sets, even just based on quite simple patterns and correlations in the data, any approach based on pseudonymisation is likely to be simplistic and open to challenge.
In any case, what is reasonable for a business to want to do with that sort of data? Can we analyse which content has been most popular on a web site over the past week/month/year/decade? Can we analyse which content a particular paying customer has been accessing, in order to promote a new plan as they approach a limit or suggest a more cost-effective one if they aren't using what they're paying for right now? Can we analyse access patterns for an account over a period of time to detect shared use of an account contrary to our terms? There are plenty of legitimate business purposes for which you might want to know the entire history of access for a particular account and which do not violate the privacy of the user in any unjustified or unfairly exploitative way, but again, any method based on log pseudonymisation will immediately fail to satisfy those requirements.
Nothing about this discussion is even remotely as clear and straightforward as some in this discussion are suggesting.