The first version of the bankruptcy.ai questionnaire had 23 sections. The current version has 17. The six sections we cut were argued over extensively — each one had a legitimate reason to be there and a legitimate reason to come out. This post is about that process and what it reveals about the tension between completeness and completion rates.
What we cut
Detailed monthly expense breakdown (granular). We had a 22-line expense section modeled directly on Schedule J. We cut it to a 9-line version. The granular version produced more accurate Schedule J data. It also produced significantly lower completion rates — clients abandoned the questionnaire at that section more than any other. We decided that a completed questionnaire with estimated expenses was better than an abandoned one with three expense lines filled in.
Prior address history (10 years). Useful for the Statement of Financial Affairs. Cut because almost no client could provide this accurately from memory, and the section generated more "I don't remember" entries than useful data. Attorneys prefer to gather this in consultation.
Business ownership history. Retained in a simplified form — "have you owned a business in the past 4 years?" with basic follow-up questions. The full business history section is available for cases where the attorney enables it.
What we learned
The right questionnaire is not the most complete questionnaire. It is the most complete questionnaire that clients will actually finish. Every question that reduces completion rates costs more in follow-up staff time than it saves in schedule prep time. The tradeoff is not obvious until you measure it.
Completion rate data from the pilot is what drove most of the cuts. We watch abandonment at every section. When a section shows disproportionate abandonment, we investigate whether simplifying it changes behavior. Usually it does.