Benchmark for evaluating language agents in enterprise software systems
The Era by Eon Benchmark: A Generated Enterprise Estate with Exact Ground Truth for Benchmarking LLM Agents
Artificial Intelligence
Summary
It is hard to test AI agents that work with company software because real company data is private and there is no standard testing setup with exact answers. The authors created a realistic, fake company with all the usual software tools and databases linked consistently, so answers to questions can be calculated exactly. This setup lets people fairly test how well AI agents understand and use enterprise systems. They showed this by testing nine models on many questions and measuring accuracy reliably.
What this means in practice
- •For enterprise software developers: Evaluate AI agents that interact with integrated enterprise tools using a benchmark with exact answers based on a realistic simulated company.
- •For customer support platform builders: Test automated assistants in simulated service environments with consistent data across tools like Zendesk and Slack to improve response accuracy.
Authors
Benjamin Gruenbaum, Doron Porat, Assaf Natanzon, Roy Zavida, Chen Dinachi, Or Itzahary
Abstract
LLM agents for enterprise systems of record cannot be evaluated on customer production data, and no existing substitute provides ground truth. We present the Era by Eon Benchmark for evaluating LLM agents that use enterprise tools. The benchmark is built around a complete fictional company. It includes product simulators, company-specific internal databases, benchmark questions, and computed answer keys. Industry, company size, business model, application portfolio, and a seed define each company. One seeded entity graph supplies shared company data to simulators of Salesforce, Zendesk, Slack, Gong, and other products. A questionconditioned generator creates the schemas and records for internal databases. It takes shared entities, keys, and values from the same graph before generating database-specific facts. Both mechanisms therefore describe one consistent enterprise estate. Every expected answer is computed from the final records, so grading is exact. Design and answer-key checks validate the internal databases. A realism scorecard and adversarial detector validate the entity graph. Across 23 generated companies, the mean realism score rose from 61.8 to 97.0, with zero records flagged as synthetic. In the reported simulator-track comparison, nine models answered the same 33 questions three times each. Accuracy estimates ranged from 42.4% to 76.8%, and three of 36 pairwise differences remained supported after correction.