Which machine learning approach fits this situation?

Unsupervised learning works on unlabeled data and is Google's named route to clustering tasks, customer segmentation among them: no segments exist yet, so the structure has to come out of the basket data itself. Supervised learning is the common miss, because a loyalty member ID looks like a label - it identifies whose row it is, not what outcome the model should predict. Reinforcement learning needs a reward loop, and no campaigns have run to produce one; the segments are meant to come first and shape the campaigns. Semi-supervised learning would apply if part of the membership were already segmented and the rest were not, and the chain has never defined a segment at all.