Argo-bench tests data agents on large enterprise workflows
Argo-Bench: Evaluating Data Agents on Enterprise-Scale Workflows
Computation and LanguageArtificial IntelligenceDatabases
Summary
Many real-world business tasks use huge amounts of data spread across many tables and require complex reasoning and actions. Existing tests for AI that generate queries only check part of this work and often rely on small public datasets. The authors created Argo-Bench, a large testing setup simulating a real food delivery company with billions of data points and many tables. Their tests ask AI agents not just to get answers but to perform actions like stopping fraud, and score them on the actual results. They find current top models can only fully solve about a third of these tasks, showing the challenge ahead.
What this means in practice
- •For enterprise data teams: Evaluate AI agents on realistic large-scale workflows that require navigating complex data warehouses and taking business actions.
- •For fraud detection teams: Test AI tools for identifying and acting on fraudulent accounts within a simulated enterprise database representing real-world complexities.
Authors
Gabriel Tomitsuka, Arman Raayatsanati, Emma Xing, Duke Gand, Joseph J Ma
Abstract
Real-world enterprise data science and analytics workflows require reasoning across dozens of tables, performing statistical analyses, and acting on the results. Established text-to-SQL benchmarks evaluate query generation alone, and audits have found their answer keys frequently wrong. Because real enterprise warehouses are too sensitive to release, these benchmarks are built on public datasets where a business event fits in a single table. We introduce Argo-Bench, an evaluation framework comprising 210 data science and analytics tasks. Drawing on public data, peer-reviewed industry literature, and regulatory filings, we simulate a food delivery platform in New York City at true scale, with 81 million orders in 2024, grounded economics, fraud patterns, and marketplace incentives. We export this world to an ERP warehouse of 235 tables and 7.5 billion rows, modeled on the Oracle E-Business Suite schema. The simulator's ground-truth state is withheld from the warehouse the agent sees, so tasks require reconstructing facts by navigating the warehouse before acting on them. Argo-Bench goes beyond text-to-SQL: the agent files actions such as banning fraudulent accounts, allocating courier incentive budgets, or issuing back pay, and the grader scores each by its consequences in the simulator. Every task has an executable reference solution that demonstrates solvability using only the warehouse. The strongest of 14 frontier and open-weight models scores 95 or higher on only 34.8% of tasks and averages 59.5 points. We hope Argo-Bench drives progress toward agents that understand, navigate, and act within real data environments.