AI-supported decision support systems promise better organizational decisions, yet in many organizations trustworthiness is still handled through principles, documentation, or post hoc explanation rather than through the system design itself. The result is a persistent gap between responsible AI guidance and enforceable accountability in business information systems. At its core, the paper presents AIJIM, an evidence-first reference model that treats trustworthiness as an architectural property. Using a design science perspective and a focused synthesis of recent literature, we derive five requirements: evidence binding, artifact-mandatory evaluation, audit-safe traceability, effective human oversight, and separation of governance from analytical execution. These requirements are consolidated into a domain-agnostic reference model and a conceptual implementation architecture that demonstrates constructability without prescribing a specific tool stack. The result is a compact reference class for AI-supported decision support systems in which outputs remain reviewable, auditable, and governable across organizational contexts.