Automation catches the obvious errors; it takes a second set of trained eyes to catch the assumption that's quietly wrong.
Modern tools are very good at catching what they're built to catch: a broken formula, a reconciliation that doesn't tie out, a duplicate entry. What they're not good at is catching an assumption that's internally consistent but wrong — a growth rate that's plausible on paper but doesn't match how the business actually sells.
That's the gap a second reviewer closes. Not by re-checking arithmetic a tool already checked, but by asking whether the story the numbers tell actually matches the business being described.
It's the reason every deliverable that leaves this practice gets reviewed twice, by two people with different backgrounds. Not because the tools aren't good — because judgment is still the part that doesn't automate.