Matching engines need tolerances to absorb rounding, FX pennies, and timing noise. Those same bands can swallow systematic underpayments or fee misposts if nobody samples the auto-cleared population.
A simple test: take a week of auto-matched items near the top of the tolerance band and re-check them manually. If many would have failed a tighter rule, your tolerance is doing operational work that belongs in exception handling.
Digitalnetops reconciliation automation checks sample forced matches and near-threshold clears so finance leaders know whether automation is helping or hiding.