Source note1 min read
EnterpriseVal: Quantifying the Efficacy, Reliability and Value of Generative AI in the Enterprise
Source
arXiv
arxiv.org/abs/2609.21841v1
Source facts
EnterpriseVal specifies the tested workflow and frozen socio-technical configuration, including model, prompts, retrieval, tools, guardrails, human oversight, autonomy level, and consequence tier. It defines metrics across fidelity, utility, efficiency, reliability, assurance, and oversight; uses calibrated LLM-as-judge scoring with prediction-powered inference; and maps confidence-bounded metric vectors to REJECT, CONDITIONAL, or SCALE gates. The authors report a pilot across three workflows in a global bank.
Analysis
The framework makes deployment decisions an explicit measurement and governance problem rather than a generic benchmark comparison. It is especially relevant to enterprise adoption because it connects workflow fit, reviewer catch rate, operational risk, and value to a staged decision gate.
Uncertainty/Doubt
- This is an arXiv preprint and includes author-reported pilot evidence; the bank pilot, thresholds, and value estimates were not independently verified here.
