• qa_deck_100#slide 27 · Design science and mid-range design theory

RAIDT is positioned as a design science research contribution because the papers define the governance problem as one that is simultaneously practical, under-specified in current Information Systems theory, and irreducibly artefactual. Existing responsible-AI principles, model documentation, explainability methods, and audit routines all contribute useful pieces, but they do not provide a bounded proof object for one configured organisational use of generative AI. The design science paper therefore argues that the relevant missing contribution is not another normative checklist. It is a designable governance object that can capture, reconstruct, compare, and challenge a specific use in context.

Within that logic, RAIDT is presented as a conceptual but operationally oriented governance artefact. It theorises run as the unit of governance and links two artefacts: the run-level evidence pack and a score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). This fits design science because the contribution is prescriptive as well as explanatory: it specifies what evidence should be captured, how governance readiness should be represented, and how the artefact can support later evaluation, calibration, and refinement.

The positioning is strengthened by the academic-logic paper, which treats RAIDT not as a one-off tool but as reusable design knowledge expressed through constructs, principles of form and function, mechanisms, propositions, and boundary conditions. That is why the framework is advanced as a mechanism-based mid-range design theory. It allows researchers and organisations to test how different configurations, including influence methods as governance interventions, alter reviewability at run level rather than relying on abstract assurance alone.

In a healthcare note summarisation workflow, a hospital uses generative AI to draft a discharge summary for a complex case. A policy saying that clinicians remain responsible is not enough if the summary is later challenged. Under RAIDT, the organisation captures a run-level evidence pack: prompt template version, model and tool configuration, retrieved guideline excerpts, output text, reviewer comments, sign-off, and relevant timestamps. The same run then receives a score profile using anchors 1=missing / 3=partial / 5=audit-ready.

This is a design science contribution because the artefact changes practice. Managers and reviewers can inspect one configured use rather than relying on memory or generic system documentation. If retrieval snapshots are absent, Auditability and Traceability fall; if escalation and oversight are explicit, Responsibility improves. The example shows why RAIDT is not merely conceptual language. It is a designed governance object that can be evaluated, refined, and reused across organisational settings.

• 12-RAIDT_DSR_Theory_M_v8
• 11-RAIDT_Academic_Logic_M_v11