DeciSyn-SaaSBench: A Synthetic Benchmark Dataset of B2B SaaS Pricing Decision Dialogues

DeciSyn-SaaSBench is a synthetic benchmark dataset of 12,000 organizational decision dialogues in the domain of B2B SaaS deal pricing approval, generated using the DeciSyn framework, a domain-adaptable pipeline for producing LLM-generated dialogues grounded in domain-specific decision contexts, role-based authority structures, and organizational governance constraints. Each dialogue depicts a two-party exchange between organizational personas (Salesperson, Sales Manager, Country Sales Leader, Deal Desk Analyst, and Deal Desk Director) negotiating a pricing decision under a defined set of scenario attributes, request types, and governance rules, concluding in one of four decision outcomes: Approve as Submitted, Adjust/Counter-offer, Escalate Up, or Reject. The corpus spans 300 unique scenarios stratified across seven complexity tiers, 10 persona pairings per scenario, and 4 independent generation runs per pairing. The dataset was validated through script-based structural checks, manual narrative review, stratified human annotation across two phases, and automatic linguistic quality metrics, as reported in the associated paper. The dataset is provided as a single JSON Lines (JSONL) file. Field definitions and further detail are provided in the accompanying README.