PlythosAI
PlythosAI provides AI agents that autonomously run business negotiations 24/7, operating within user-defined limits and escalating unusual cases to humans. The platform probes counterparties to infer hidden preferences, maps the full preference landscape to surface mutual-gain trade-offs, and achieves an 81% win rate in standard benchmarks—outperforming both human negotiators and GPT-5. It targets buyers, sellers, and brokers in one-on-one pricing negotiations across digital resale marketplaces, high-value goods dealers, and spread-based intermediaries.
- AI Agents
- Artificial Intelligence
- Commerce & Shopping
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Negotiating deals where pricing is determined counterparty by counterparty—whether buying, selling, or operating on the spread—requires continuous back-and-forth communication that is time-intensive and difficult to scale. Manual negotiation processes are slow, inconsistent across deals, and often miss optimal outcomes because they cannot systematically explore the full preference landscape of both sides.
Solution
PlythosAI provides AI agents that autonomously handle business negotiations around the clock, operating strictly within boundaries set by the user and escalating unusual cases for human review. The system probes counterparties to infer hidden preferences and constraints, building a live model of what the other side values and will accept. It then maps the full preference landscape to identify non-obvious trade-offs that create mutual gain, turning zero-sum standoffs into structured wins. Built on research from the University of Edinburgh, the platform uses a hierarchical evaluation framework with interpretable mid-level signals to diagnose strategy as conversations unfold, rather than relying solely on final outcomes. Benchmark results show the AI achieving an 81% win rate—dramatically outperforming both human negotiators (6%) and GPT-5 (13%)—with findings currently under peer review and a preprint forthcoming on arXiv.
Target Audience
PlythosAI targets organizations involved in one-on-one pricing negotiations, including digital resale marketplaces, dealers of high-value goods (watches, instruments, hi-fi, antiques, sneakers), and brokers operating on spread-based deals who need to scale negotiation capacity without increasing headcount.
Features
- MCTS-distilled reasoning trained on millions of Monte Carlo Tree Search simulations, enabling strategic multi-step lookahead rather than token prediction
- Partial-observability modeling that probes counterparties to infer hidden preferences and constraints in real time
- Preference arbitrage engine that maps the full preference landscape to surface non-obvious mutual-gain trade-offs
- Hierarchical evaluation framework with three layers of interpretable mid-level signals for transparent strategy assessment during conversations
- Continuous 24/7 operation with user-defined floors, comps, and escalation of high-value, unusual, or out-of-policy cases to human teams
- Integrates with existing chat tools like Tidio or standalone chat setup, with performance-based pilot billing measured against client baselines
- Published benchmark results demonstrating 81% win rate versus 6% for humans and 13% for GPT-5, with preprint forthcoming on arXiv