Sanctions lists are multilingual and the same Arabic, Cyrillic or Chinese name can be romanised several defensible ways. Exact matching would miss almost everything, so screening systems score similarity and alert above a threshold.
The consequences run both ways. A threshold set for a manageable alert volume will miss deliberate minor variants, and evasion that relies on that is well documented. Conversely, most alerts are false positives, and the labour of clearing them is a real constraint on how tightly a system can be tuned.