Source: draft chapter in content/000 Articles/020 !! dezim klar book chapter (DRAFT).md .

• Purpose

• Pilot an AI interviewer (“Harvest Assistant”) for Outcome Harvesting (OH) in the KLAR! programme, focusing on scalability, inclusion, and democratic evaluation value.

• Key method move

• AI-led interviewing + AI post-hoc transcript analysis to draft a structured OH outcome table.
• Strong emphasis on traceability: verbatim short citations + page references; explicit missing-information prompts.

• Human validation checks accuracy and de-duplicates overlapping outcomes across sources (triangulation).

• Results highlights (as reported)

• 39 invited; 19 responded; 38 outcome statements; 6 met SMART criteria; others retained as leads.

• Real-time outcome summaries enable respondent validation and transparency.

• Operational insights

• Prompt simplicity improves adherence; model choice matters; version prompts/models for comparability.

• Scaling shifts bottlenecks to analysis unless the end-to-end workflow is designed.

• Risks / responsible scaling

• GDPR/legal basis, consent, third-party naming pathways; document data flow and model/prompt versions; data sovereignty (EU-hosted inference where possible); attention to equity/digital divides.