Deterministic Enterprise Intelligence: Toward Trustworthy, Reproducible Answers Across Disconnected Data Systems
This working paper argues that in regulated, high-stakes enterprise environments banking, insurance, and healthcare trustworthy data answers require that trust precede intelligence: before an answer can be useful, it must be verifiable, reproducible, and traceable to its source. Modern language-model tools make it easy to ask questions of enterprise data in natural language, but their answers are often plausible rather than correct, and can change between runs unacceptable where a figure must satisfy a regulator, a board, or an auditor. The paper sets out a set of design principles for a deterministic enterprise-intelligence layer that separates the interpretation of a question from the computation of its answer, treats provenance as a first-class property, grades relationships between disconnected data sources by the strength of evidence behind them, refuses to answer rather than answer wrongly, and keeps sensitive data in place. It discusses why prevailing approaches natural-language-to-SQL generation, retrieval-augmented generation, and data-virtualization platforms address access and fluency but not trust, and outlines the implications of a deterministic, provenance-first approach for regulated industries and for the broader goal of auditable enterprise AI.
Authors
- Zahid Zia Siddiqui
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24246368
- Primary Topic
- Explainable Artificial Intelligence (XAI)
- Type
- article
- Field-Weighted Citation Impact
- 0.00