A closely related strategy, which might be useful given your relatively small sample, is mentioned in Bryan Orme's method text. Accordingly, use HB to compute individual utility coefficients. Separately, use Latent Class to generate segments of respondents with similar preferences. Select a Latent Class model (probably 2 or 3 classes) which minimizes BIC and yields a conceptually useful solution. Then use the posterior probability of group membership to determine the most probable segment for each participant.