From Business Events to Auditable Decisions: Ontology-Governed Graph Simulation for Enterprise AI

Hongyin Zhu; Jinming Liang; Mengjun Hou; Ruifan Tang; Xianbin Zhu; Jingyuan Yang; Yuanman Mao; Feng Wu

cs.AIApr 8, 2026

From Business Events to Auditable Decisions: Ontology-Governed Graph Simulation for Enterprise AI

Authors: Hongyin Zhu, Jinming Liang, Mengjun Hou, Ruifan Tang, Xianbin Zhu, Jingyuan Yang, Yuanman Mao, Feng Wu

Organizations: Yonyou AI Lab · Yonyou Network Technology Co., Ltd.

Abstract

Existing LLM-based agent systems share a common architectural failure: they answer from the unrestricted knowledge space without first simulating how active business scenarios reshape that space for the event at hand -- producing decisions that are fluent but ungrounded and carrying no audit trail. We present LOM-action, which equips enterprise AI with event-driven ontology simulation: business events trigger scenario conditions encoded in the enterprise ontology~(EO), which drive deterministic graph mutations in an isolated sandbox, evolving a working copy of the subgraph into the scenario-valid simulation graph $G_{\text{sim}}$ ; all decisions are derived exclusively from this evolved graph. The core pipeline is \emph{event $\to$ simulation $\to$ decision}, realized through a dual-mode architecture -- skill mode and reasoning mode. Every decision produces a fully traceable audit log. LOM-action achieves 93.82% accuracy and 98.74% tool-chain F1 against frontier baselines Doubao-1.8 and DeepSeek-V3.2, which reach only 24--36% F1 despite 80% accuracy -- exposing the illusive accuracy phenomenon. The four-fold F1 advantage confirms that ontology-governed, event-driven simulation, not model scale, is the architectural prerequisite for trustworthy enterprise decision intelligence.

Figures & tables

Panel of Table 1 — Table 1: An illustrative example of the three-phase event $\to$ simulation $\to$ decision pipeline, triggered by a streaming expense-approval event activating an EO-encoded access-scope condition.

Panel of Table 2 — Table 2: An illustrative example of the dual-mode (skill mode + reasoning mode) pipeline, triggered by a streaming organisational-change event that augments the sandbox graph and requires a conflict-free audit-slot assignment.

Task	LOM-action	Doubao-1.8	DeepSeek-V3.2
CONNECTIVITY	1.00	0.98	1.00
NEIGHBOR	1.00	1.00	0.98
PREDECESSOR	1.00	1.00	1.00
EDGE	1.00	1.00	1.00
fc_graph_info	0.88	0.94	0.78
fc_node_info	1.00	0.16	0.86

Table 3: Answer Accuracy (Acc)

Task	LOM-action	Doubao-1.8	DeepSeek-V3.2
CONNECTIVITY	1.00	0.00	0.00
NEIGHBOR	1.00	0.00	0.00
PREDECESSOR	1.00	0.00	0.00
EDGE	1.00	0.00	0.00
fc_graph_info	1.00	0.00	0.92
fc_node_info	1.00	0.00	0.06

Table 4: Tool-Chain F1

Group	LOM Acc	LOM F1	Doubao Acc	Doubao F1	DS-V3.2 Acc	DS-V3.2 F1
Basic Traversal ( $\times$ 4)	1.00 [1.00, 1.00]	1.00	0.995	0.00	0.995	0.00
Information Retrieval ( $\times$ 2)	0.94 [0.89, 0.99]	1.00	0.55	0.00	0.82	0.49
Graph Algorithms ( $\times$ 2)	1.00 [1.00, 1.00]	1.00	1.00	0.01	0.80	0.14
Scenario-Simulation ( $\times$ 2)	0.99 [0.97, 1.00]	0.973	0.82	0.464	0.80	0.511
Hybrid / Mode C ( $\times$ 1)	0.46 [0.36, 0.56]	1.00	0.08	0.22	0.00	0.41

Table 5: Grouped Results (Acc / F1) with 95% Confidence Interval (CI) on LOM-action Acc

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Input	Output	Typical Use Case
Natural Language	Natural Language	Business event or user request $\to$ human-readable decision with evidence narrative
Natural Language	OWL / Triples	Business event or user request $\to$ machine-readable ontology delta for downstream systems
OWL / Triples	Natural Language	Formal scenario or graph state $\to$ explanation, audit report, clarification dialogue
OWL / Triples	OWL / Triples	System-to-system ontology enrichment, automated scenario verification, RAC-driven EO update

Appendix

Table 6: Mapping of Input/Output Modalities and Use Cases

Ontology-Based Reasoning

See all 15 papers

EnterpriseBench: Benchmarking LLM Agents on Enterprise-Level Strategic Reasoning and Decision-Making
Sep 29, 2026Min Yang, Yichen Pan, Jinghua Piao +3AI-Assisted Decision Making LLM Agent Evaluation
Shandong University · Zhongguancun Academy · Beijing Institute of Technology +1
Ontology-Constrained Neural Reasoning in Enterprise Agentic Systems: A Neurosymbolic Architecture for Domain-Grounded AI Agents
Apr 1, 2026Thanh Luong Tuan, Abhijit SanyalOntology-Based Reasoning Enterprise AI Agents
Golden Gate University, San Francisco · Foundation AgenticOS (FAOS) · Associate Director, Data, Digital & IT, Novartis Healthcare Pvt. Ltd., Hyderabad, India
The Era by Eon Benchmark: A Generated Enterprise Estate with Exact Ground Truth for Benchmarking LLM Agents
Sep 9, 2026Benjamin Gruenbaum, Doron Porat, Assaf Natanzon +3Tool-Use Evaluation LLM Agent Evaluation
Eon

Sep 29, 2026J/K move · Enter open · S save

EnterpriseBench: Benchmarking LLM Agents on Enterprise-Level Strategic Reasoning and Decision-Making

Min Yang, Yichen Pan, Jinghua Piao, Dandan Song+2

Shandong University · Zhongguancun Academy · Beijing Institute of Technology+1

AI-Assisted Decision Making LLM Agent Evaluation Long-Horizon Agent Evaluation AI Agent Benchmarks LLM Decision-Making

Open paper AskPDF arXiv

LLM agents are increasingly expected to support enterprise workflows, where tasks often involve missing information, uncertainty, feedback, and long-term trade-offs. However, existing enterprise and financial benchmarks mainly test static capabilities such as information extraction, numerical calculation, domain knowledge, and financial QA, leaving interactive and long-horizon decision-making underexplored. To bridge this gap, we introduce EnterpriseBench, a benchmark that evaluates LLM agents across this spectrum, from static question answering to dynamic decision-making. Specifically, EnterpriseBench reorganizes existing enterprise and financial QA datasets into a unified foundational suite annotated by capability and difficulty, and introduces three professional interactive settings: Consulting, based on management-consulting-style business cases for client problem diagnosis through multi-turn information seeking; the Beer Game, adapted from a classic supply-chain management simulation for inventory control under delayed feedback; and Enterprise Digital Twin, a project-based business simulator for workforce, risk, and project planning. Experiments with nine agent methods under four backbone models show that current agents have not yet achieved stable, comprehensive, and cross-task reliability in enterprise scenarios. These results show that EnterpriseBench provides a practical benchmark for evaluating LLM agents in realistic enterprise strategic reasoning and decision-making.

Abstract

Ontology-Based Reasoning

Explore similar work

EnterpriseBench: Benchmarking LLM Agents on Enterprise-Level Strategic Reasoning and Decision-Making

Ontology-Constrained Neural Reasoning in Enterprise Agentic Systems: A Neurosymbolic Architecture for Domain-Grounded AI Agents

The Era by Eon Benchmark: A Generated Enterprise Estate with Exact Ground Truth for Benchmarking LLM Agents