I can sympathize with the challenges of running a web scale analytics tool, after helping build and run an analytics stack for an ad network. Even with our relatively small group of customers and lots of planning, the data was abundant, the queries became quite slow, and the pipeline was never "real-time" enough for some.
The web is a noisy place! URLs parameters are tricky to canonicalize into pages when dealing with many properties. There is a constant stream of nonsensical referrers and even referrer spam. Traffic can be exceptionally spiky, and cascading failure in the pipeline is inevitable without careful planning. Data loss is irrecoverable, and any incident needs to be carefully explained to users of the product so they don't deceive themselves into making decisions based on missing data.
One specific point I'm concerned about from the post is that while comparing HackerNews traffic to organic search, their analysis fails to account for relative traffic volume. Its always easy for a small volume traffic source to show outstanding metrics, but miss the broader point: quality traffic sources usually don't scale. A useful analysis should assign weight to overall volume. ORDER BY MAX(CTR) will tell me that some keyword/referrer was the top performer at 80%, but without knowing that was for 3 visitors I could easily get the wrong ideas. I think analytics tools aspire to make it difficult to come to incorrect conclusions, acknowledging that most of their users will have no meaningful background in statistics.
All of those challenges are surmountable, but it definitely gives some credence to the talented engineers who have make analytics stacks look easy. I wish Volument luck in their journey and will keep an eye on their progress, because their pitch remains quite appealing.