Generative AI for Automated Negotiation (NEC)
Summary: NEC Corporation (2025) demonstrates that generative AI overcomes two key failures of classical automated negotiation — rigid protocols and context-blindness — by adding natural-language interaction, contextual reasoning, and multimodal time-series forecasting. Tested in real procurement scenarios with demonstrated inventory reduction.
Sources: Academia/JOI-Article-XMProFINAL.pdf
Last updated: 2026-05-06
Citation
Mohammad, Y., Ando, T., Chen, H., Higa, R., & Morinaga, S. (2025). Generative AI for automated negotiation. Journal of Innovation (NEC Corporation). November 2025.
The problem with classical automated negotiation
Game-theoretic and decision-theoretic automated negotiation systems suffer from two fundamental weaknesses:
- Rigid protocols: fixed turn structures and bid formats that do not accommodate real-world negotiation flexibility
- Context-blindness: difficulty incorporating market information, qualitative signals, and emergent conditions into preference evaluation — leading to suboptimal actions and “money on the table”
These limitations mean that classical systems can negotiate in theory but fail to capture value in practice.
What generative AI adds
The paper proposes a Generative AI-Powered Negotiation Agent built around three capabilities:
- Interacting with humans: natural-language dialogue enables extraction of preferences, constraints, and context from emails, conversations, and documents — circumventing rigid protocol requirements
- Reasoning during negotiation: LLMs evaluate the negotiation space contextually, assess offers against evolving goals, and generate counter-proposals that go beyond pre-programmed bid functions
- Utility function evaluation: multimodal time-series forecasting (demand prediction) informs what agreement terms are actually worth — correcting the context-blindness of classical systems
Architecture
A cross-boundary multi-agent system. The GenAI layer sits between:
- Human participants (via natural language)
- Automated decision engines (via structured data and forecasts)
- External market signals (via multimodal time-series inputs)
The system does not replace human negotiators for high-stakes decisions but automates the routine negotiation volume that accounts for most of the transactional work.
Use cases
Supply-demand matching (direct procurement): GenAI predicts demand from multimodal signals → generates a target procurement plan → negotiates with suppliers to match the plan. Results showed significant inventory reduction across four product categories; combining negotiation with GenAI-based forecasting outperformed negotiation alone.
Indirect matching: More complex multi-factor business model with 20 input variables. GenAI models the utility function from qualitative and quantitative inputs, enabling negotiation over complex multi-dimensional agreements.
Connections to the AI competency literature
The NEC approach illustrates the Mikalef et al. model (mikalef-ai-competencies-b2b) in action: the value is not in deploying an LLM, but in bundling it with forecasting infrastructure, domain knowledge about procurement, and institutional integration with existing supplier relationships. Without that bundling, the LLM is just a chatbot.
Risks and mitigation
The paper identifies: hallucination in contract-critical reasoning, adversarial negotiation parties probing system weaknesses, and regulatory exposure from automated binding agreements. Mitigations include human-in-the-loop for high-value decisions and auditability of the reasoning trace.