Turning Interviews into Confirmable SOPs
A staged AI workflow that preserves source material, identifies gaps, and produces a process map and SOP draft after human confirmation.
Context and challenge
Interviews can contain informal language, omissions, different viewpoints, and unconfirmed exceptions. Generating a document directly from a transcript can turn assumptions into process steps, leaving later readers unable to tell what has actually been confirmed.
Workflow
- Source
- Quality check
- Local transcription
- Human review
- Process model
- Follow-up
- SOP + process map
- Organize source material
- Build an initial model
- Mark information gaps
Process stakeholders answer, correct, and confirm responsibilities
- Draft the SOP
- Draw the process map
- Check consistency
Two-stage structure
The first stage organizes evidence and gaps only. Human confirmation is a required gate before the second stage, not a footnote.
Current artifacts
- Preserve recordings, transcripts, or notes, then identify quality, speaker, and low-confidence issues.
- Build an initial process model from confirmed material while listing roles, exceptions, and questions that still need answers.
- Keep confirmation with the people responsible for the process; produce the SOP and process-map draft only after their responses are incorporated.
Current outcome
The prototype has completed an end-to-end, everyday test case and connects source material, open questions, and document drafts in a clear sequence. It has not yet completed long-form real-interview testing, formal adoption, or impact measurement, so no efficiency or quality improvement is claimed.
Evidence, public scope, and limitations
This case describes only a public-safe method and anonymized prototype material. It contains no real recordings, transcripts, department data, names, system screens, or controlled documents.
- End-to-end everyday test case
- Staged workflow and quality gates
- Editable process-map and SOP-draft structure
Long-form real-interview testing, formal department adoption, process-owner approval, and before-and-after measurement have not yet been completed; this page makes no quantified claim.
What I learned
A reliable AI workflow does not need to decide on behalf of the process owner. Its value is also knowing when evidence is insufficient and making the next confirmation clear.