Collecting telemetry is essentially doing an impromptu social science study, and it suffers from the same problems - collecting good data and making correct inferences is a difficult task, and it requires specialized skills. Social scientists are trained to do this, and yet they still fail more often than not[0]. A random business trying to run product development off automated metrics? They don't stand a chance.

This is not to say the data is useless - it's just worth remembering that, unless you have strong statistical background, you're probably reading your data wrong, so your metrics should be treated as weak evidence at best.

--

[0] - It's not snark, it's a corollary of the replication crisis. This is a hard job.