Data2Decision: A Prescriptive Analytics Data Agent Bridging Enterprise Information and Optimal Decisions

ICLR 2026 Conference Submission14720 Authors

Keywords: Prescriptive Analytics, NL-to-OPT, Enterprise Decision-making, Data Agent

Abstract: Enterprise business analytics has evolved from simple reporting to sophisticated decision-support systems. Prescriptive analytics, as the most advanced form of business analytics, aims to recommend optimal actions based on data, yet the field lacks standardized benchmarks that reflect real-world complexity where optimization parameters must be extracted from enterprise databases. We present Data2Decision, the first data agent specifically designed for database-grounded prescriptive analytics that produces mathematically optimal decisions, accompanied by Schema2Opt, a comprehensive benchmark simulating enterprise decision environments. Schema2Opt transforms SQL database schemas into realistic optimization problems, providing complete business contexts, operational databases, and verified solutions. Our Data2Decision agent tackles these scenarios through a two-stage pipeline with test-time scaling: first extracting optimization parameters from databases via SQL generation, then formulating and solving optimization models with multi-solver consensus, achieving end-to-end automation without manual preprocessing.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 14720