Whatever indicators they use to select audit recipients may correlate with under-payers, but it's not a guarantee. I'm sure there's some fraction of those they audit that turn out to have paid too much.
They have some data on this (https://www.irs.gov/pub/irs-pdf/p55b.pdf, pages 33-35). Out of 509,917 examinations, 18,988 resulted in a refund. There was $7.0 billion of tax refunded in 2020. For comparison, there was $17.2 billion of additional tax generated with additional recommended and unagreed amounts.
