Ensuring fairness in automated systems is critical in today’s data-driven landscape. Prior work has exposed algorithmic bias in popular name matching systems, particularly affecting names from certain racial backgrounds, but the root causes and effective solutions remain underexplored. This paper introduces FairNM, a novel system that reduces bias through token-based similarity scoring, a Siamese Neural Network-based Short Name Module, and Name Weighting. Using a novel test bench and fairness metric, we show that FairNM achieves accurate fuzzy name matching while significantly improving fairness. Its performance is especially valuable in Web-based contexts requiring high recall and fairness, such as fraud detection, where mismatches can have serious impact.