Forced alignment tools such as the Montreal Forced Aligner (MFA) are increasingly adopted in sociophonetic research, yet their reliability for coding phonological variation remains insufficiently understood. This study evaluated MFA-human agreement across multiple sound change features in Hong Kong Cantonese (HKC), comparing coding decisions between MFA and human coders across two age groups. The results show that agreement is highly feature-dependent, with younger speakers yielding lower agreement rates than older speakers. Focusing on the two features with the lowest agreement, error and acoustic analyses revealed that disagreement does not reflect speaker-related factors but rather systematic differences in how human and MFA process ambiguous phonetic information. These findings suggest MFA’s potential as a complementary tool to human coders in sociophonetics and variationist research.