ACBC's designer has a specific way of selecting the composition of product concepts, described in good detail in the manual. Although we use d-efficiency to help guide the designer, d-efficiency is just one of the multiple goals that are traded off. One of the overriding goals for the designer is to over-sample the levels picked in the BYO section of the survey; which obviously is at odds with traditional d-efficiency metrics.
Regarding the modeling of the data, we use MNL with its additive compensatory assumption & RUT (random utility theory). However, the way that we augment the data matrix when respondents state that a levels are unacceptable tends to push those unacceptable parameters a bit lower in utility than a standard CBC (without the data augmentation) would do.
As an example, if in the middle of the survey a respondent states that the color "red" is unacceptable (and we only allow the respondent to do so after we observe a pattern of that respondent consistently rejecting any product concept containing the red level), then we look forward to the concepts not yet evaluated by the respondent (but that are planned to be asked) and automatically reject any concepts containing the color red. We replace the rejected concepts with new concepts that don't include red.