# WisdomTwin Working Paper WT-100-001 — Source record > **This Markdown file is a metadata and source-of-record document.** It does not reproduce the > full manuscript. The canonical version of record is the PDF at > . Where this file and the PDF > differ, the PDF governs. ## Bibliographic record - **Series record:** WisdomTwin Working Paper WT-100-001 - **Version:** 2.0, fact-checked revision - **Title:** Judgment Latency in Regulated Enterprises: A Conceptual Framework and Research Agenda for Role-Specific AI Decision Support - **Author:** Roman Bodnarchuk - **Role and affiliation:** Co-Founder and Chief Executive Officer, WisdomTwin, Inc. - **Location:** Toronto, Ontario, Canada - **Email:** roman@wisdomtwin.ai - **ORCID:** - **Date:** September 3, 2026 - **Status:** Pre-publication working paper / preprint; not peer reviewed; DOI pending repository deposit (no DOI has been issued) - **License:** Creative Commons Attribution 4.0 International (CC BY 4.0), - **Language:** English - **Landing page (canonical):** - **PDF (version of record):** **Scope note (from the paper).** This paper proposes constructs, design requirements, testable propositions, and an evaluation protocol. It reports no live-enterprise deployment and no causal product-performance result. **Suggested citation.** Bodnarchuk, R. (2026). Judgment latency in regulated enterprises: A conceptual framework and research agenda for role-specific AI decision support (WisdomTwin Working Paper WT-100-001, Version 2.0). WisdomTwin, Inc. ## Abstract (from the PDF) Organizations can possess the information required for a decision while still waiting for a particular role to attend to it. This paper defines judgment latency as the elapsed time between decision readiness and attention by the person or forum whose judgment is required. It develops a conceptual framework for studying that interval and proposes the AI Judgment Twin as a role-specific, evidence-grounded decision-support architecture. The proposed system is not a personal replica and does not independently hold organizational authority. It is intended to retrieve authorized precedents, expose the principles, thresholds, exceptions, and escalation patterns reflected in prior decisions, and produce cited recommendations within an explicitly validated boundary. The argument integrates research on work interruption, meeting science, decision rights, organizational memory, digital and human twins, and human reliance on automation. It also specifies a non-overlapping model of potentially recoverable time, distinguishes externally reported estimates from author-derived arithmetic, and states six propositions suitable for empirical testing. For regulated settings, the paper treats provenance, least-privilege access, human confirmation, safe decline, monitoring, and contestability as design requirements rather than demonstrated product capabilities. A staged evaluation design is proposed, beginning in shadow mode and advancing only if decision quality and permission integrity meet pre-specified thresholds. The paper is conceptual and theory-building. WisdomTwin is pre-revenue, has no production users, and has not validated the proposed effects in a live enterprise. **Keywords:** judgment latency; AI-assisted decision making; organizational memory; decision rights; regulated enterprises; human oversight; provenance; asynchronous management ## Claim-status convention (from the PDF) | Status | Meaning in the paper | | --- | --- | | Established evidence | A bounded statement supported by the cited empirical, scholarly, or official source. | | Author-derived | Arithmetic or synthesis performed in the paper from cited inputs; not reported by the underlying source. | | Conceptual proposition | A testable theoretical claim that has not been established for the proposed system. | | Illustrative | A synthetic example or scenario; not an observed organization, user, outcome, or forecast. | | Design requirement | A capability or control that a conforming implementation should satisfy; not a claim that the capability has been validated. | ## Structure of the manuscript 1. Introduction 2. Method and scope 3. Related literature (work fragmentation and attention; meetings, decision rights, and speed; organizational memory and role knowledge; from digital twins to role-specific agents; human reliance and governance) 4. The judgment-latency construct (position within the decision lifecycle; boundary conditions; role judgment rather than personal likeness) 5. From decision record to governed representation (the admissible decision record; proposed judgment structure; role lineage, drift, and contestability) 6. Proposed architecture and design requirements (authorized evidence layer; role-model layer; interaction and authority levels; Table 1, governance controls as falsifiable design requirements) 7. Illustrative scenarios (finance evidence packet; compliance agenda triage; succession support), all synthetic 8. Quantitative model and the McKinsey estimate (reproduction of the 2019 estimate; non-overlapping model of potentially recovered time; Table 2, illustrative sensitivity analysis) 9. Propositions (P1 through P6, falsifiable) 10. Empirical evaluation agenda (study progression; outcomes and analysis; Table 3, minimum pilot measurement set) 11. Regulated settings and legal boundaries 12. Risks, limitations, and discussion 13. Conclusion Back matter: Declarations; Version history; References (31 sources); Appendix A, Glossary; Appendix B, Minimum decision-readiness record. ## Key attributions preserved verbatim - De Smet, Jost, and Weiss (2019, McKinsey Quarterly) reported about 530,000 manager-days and approximately USD 250 million in annual wages for ineffective decision-making time in a typical Fortune 500 company. Those figures are McKinsey's estimate, not a WisdomTwin estimate, not a regulated-industry estimate, not a cost-benefit analysis, and not evidence of recoverable savings. - The "about one day a week" restatement (5 x 0.37 x 0.58 = 1.073 workdays per week) is a transparent, author-derived arithmetic restatement of McKinsey's assumptions. It was not reported as a weekly result by McKinsey and should not be presented as a measured universal manager experience. ## Declarations (from the PDF) - **Conflict of interest.** The author is Co-Founder and Chief Executive Officer of WisdomTwin, Inc., is involved in developing the proposed architecture, and has a financial interest in its adoption. - **Product and evidence status.** WisdomTwin is pre-revenue, has USD $0 product revenue, no production users, and no paying customers. The company has five founder-built synthetic demonstrations. No demonstration, pipeline activity, or client outcome is presented as product traction or empirical validation in this paper. - **Funding.** No external research funding was reported for the preparation of this working paper. - **Data availability.** No new empirical dataset was collected or analyzed. The arithmetic in Section 8 is reproduced from stated source assumptions and shown in full. - **Ethics.** The paper reports no research involving human participants, personal records, or live organizational deployments. All scenarios are synthetic. - **AI-assistance disclosure.** Generative AI tools assisted with source discovery, fact-checking, drafting, editing, and document production. Citations and quantitative claims retained in this version were checked against publisher pages, versions of record, or official institutional sources on September 3, 2026. The named author remains responsible for the argument, disclosures, and final submitted text. - **Peer-review status.** This working paper has not been peer reviewed. ## Version history (from the PDF) | Version | Date | Status and material changes | | --- | --- | --- | | 1.0 | September 3, 2026 | Initial pre-publication draft; references flagged as unverified. | | 2.0 | September 3, 2026 | Fact-checked conceptual revision; source caveats corrected; an unverifiable draft citation removed; McKinsey arithmetic reproduced and relabeled; architecture claims converted to design requirements; propositions and evaluation protocol added; legal scope narrowed. | ## Machine-readable records - Catalog (JSON): - RSS: - BibTeX: - RIS: - Reading path for AI systems: