What data actually fixed
The rise of analytics in basketball corrected real, specific errors: the overvaluing of raw scoring volume regardless of efficiency, the undervaluing of players whose contributions did not show up as points, and a general tendency to trust vivid memory over a full season's pattern. These were not small corrections, and anyone who dismisses analytics wholesale is arguing against a track record of catching mistakes the eye test reliably made.
That track record is exactly why the pendulum swung as hard as it did, and why an entire generation of front offices built their processes around numbers first. The swing was earned.
What data still misses
But a possession contains information a box score does not capture well — the timing of a cut relative to a teammate's gather, the quality of a closeout, the split-second decision to pass up a good shot for a great one. Some of this is slowly being captured by tracking data, and some of it genuinely resists quantification because it depends on context a spreadsheet does not hold, like what a defense expected to see.
A trained eye, watching for these things specifically rather than simply reacting emotionally to highlights, still adds information that the current state of public data does not fully replace, and pretending otherwise flatters the data more than the data has earned.
Instruments, not tribes
My honest position is that treating this as a rivalry between two camps is the actual mistake, not a disagreement about method. Data is extremely good at correcting for memory and small samples; careful observation is extremely good at capturing context and timing that resists measurement. Anyone doing serious basketball analysis needs both, and choosing a tribe rather than using both instruments for what they are good at is a choice to be worse at the job.
The people I trust most in this argument are the ones who can tell you exactly which question a statistic answers well and exactly where they would still trust their own eyes over it, because that answer requires actually understanding both tools rather than being loyal to one.