This paper studies whether competition disciplines financial-service pricing when providers learn client types through repeated relationships. We develop a dynamic brokerage-competition model in which brokers set commissions based on their posterior belief that a client is sophisticated. Price dispersion arises from heterogeneous broker beliefs generated by learning noise, rather than from mixed-strategy randomization as in Stahl (1989). Repeated interactions give incumbents a relationship-specific informational advantage: even for clients perceived as sophisticated, incumbents need not offer fully competitive rates because rivals cannot verify their sophistication with the same precision. This non-portability of favorable information creates a loyalty trap and sustains price discrimination. Applying the model to China’s retail brokerage industry, we document that competition lowered average commissions but widened dispersion and the fee gap between sophisticated and naive investors. Model estimates show that increased competition can backfire by amplifying price discrimination and dampening fee reductions.