Abstract This article focuses on languages of the Kwilu-Ngounie subbranch within a branch of the Bantu language family known as West-Coastal Bantu. Within Kwilu-Ngounie, B70 and B80 languages emerge as paraphyletic in the most comprehensive lexicon-based phylogeny of the branch. We assess whether the impossibility to group them into lexicon-based monophyletic subgroups can be bypassed by using the phonological innovation of word-final loss of Proto-Bantu *Ćg as diagnostic of a new subgroup. It is hard to tell whether this new subgroup is a clade by descent or instead a taxon resulting from a contact-induced innovation affecting related varieties. The unconditioned reflexes of *Ćg across varieties signal that both language-internal lexical diffusion and contact-induced crosslinguistic spread of phonological innovation thwart the Neogrammarian axiom of flawlessly regular sound change. Beyond its relevance for low-level Bantu subgrouping, this article contributes to the methodological issue of conflicting lexical and diachronic phonological evidence for internal classification.