Vicente Guerola-Navarro

Review controls such as the four-eyes principle rest on a second person catching what a first person missed, for instance a finance manager who approves a supplier invoice after an accounts-payable clerk has checked it. Generative-AI assistants can now pre-check such documents. When both reviewers rely on assistants that read the same evidence, their errors may no longer be independent. Research on review controls credits independence to the separation of people, and research on AI in auditing studies how auditors weigh AI-based evidence. Neither can say how many of an assistant's misses two reviewers inherit. This proposal asks how much independence a four-eyes review keeps when both reviewers rely on generative-AI assistants. A computational study measures how often assistants of several model families miss the same planted errors in supplier invoices. Each assistant reads the system extract, the recognised text or the image of the invoice. A within-person study with eight to ten professionals from accounting, controlling and internal audit measures how strongly reviewers follow a pre-check that is wrong. Combining both gives the joint miss rates of reviewer pairs and triads, compared with unaided reviewers on the same invoices. A document analysis codes whether standards and control frameworks credit independence to persons, systems or evidence sources. We expect a shared evidence source to raise joint misses beyond those of unaided reviewers, and the frameworks to credit independence to persons. The results would show when a second reviewer still adds the assurance that a review control is credited with.

Organisations losing an expert must decide what to capture before the departure, not knowing what the successor will need. In German-speaking practice, Expert Debriefing supports this decision with a knowledge map of the role. Expert Debriefing addresses the future through a judgement of which knowledge will matter, topics for the future, questions about the months after departure and an assessment of the successor's needs. All four look forward from today's conditions, and the work history looks back, so none of the five varies them. This paper adds a scenario dimension to the map's two dimensions, topic and information type, and re-asks its questions under a few stated future states. Future relevance is no longer assigned by judging what will matter but by procedure: knowledge counts as relevant to the scenario under which it was elicited. We argue, from scenario planning, that the map's schema is the special case with one unnamed scenario, the continuation of the present. We derive that schema, its boundary conditions and seven propositions, one of them on when scenarios narrow rather than widen what is captured. We conjecture that scenarios also change what an expert recalls, as experiments found they change strategic judgement. For knowledge retention research, the argument shifts attention from what is lost to the schema that decides what is asked. For practice, three untested steps follow: agree on scenarios and orient the expert to them, select topics under them rather than under the present, and link each element to its scenario.

Critical success factors in IT governance: A quantitative study in the Colombian financial sector