From interview transcripts to a consulting report draft

Situation

The organization's consultants assess a church by interviewing people in it and then writing a report. Each report was written by hand after reviewing all of the transcripts. Past reports had each followed a different outline, so there was no standard to start from.

Solution

Before development began, Jeff Hunt had AI read the organization's past reports and teaching documents. From them it built a framework of themes, recommendation categories and reusable passages. A comparison of the past outlines produced a standard one. The application then works in stages.

1. The consultant uploads the interview transcripts, and the AI identifies themes across all of them.
2. The consultant confirms, rejects or edits each theme and can read the evidence behind it.
3. The application suggests recommendations from the confirmed themes, sorted into time frames, and the consultant chooses among them.
4. The AI drafts each section in the organization's own terms, and any section can be redrafted on its own.

Design decisions

Nothing is drafted until a consultant has approved the themes and the recommendations. Interviews are a small sample, so the AI qualifies its observations, never invents statistics, quotes interviewees exactly and names no one outside a list of key people. A theme must come from at least three people.

Outcome

A consultant can produce a report draft in a standard outline from a set of transcripts, with every theme and recommendation approved before drafting begins. When the tool moved into the organization's dashboard, its finished report on an actual assessment matched the original version's report in every paragraph.