Home
Adaptive Evolution Technology
A core "self-evolving" technology that estimates the interests and strategies of counterparts even under imperfect information. Agents progressively redesign prompts and workflows based on offline analysis of operational logs. While quantitatively evaluating fairness and efficiency grounded in game theory, it autonomously converges toward consensus proposals highly suited to the field. Negotiation is a representative application example, and this technology can be extended to coordination, allocation, and planning in other domains.
Challenges in Retail Business Negotiations
In the retail merchandising (MD) cycle, midstream processes such as demand–supply planning and procurement often become bottlenecks. Transaction terms like rebates tend to become personalized and complex, undermining fairness and transparency and causing delays in decision-making. This technology visualizes the negotiation process and presents decision criteria and process guidelines that help steer discussions toward well-grounded consensus.
With Adaptive Evolution Technology
It updates the other party's values (belief space) Bayesically from conversation and proposal histories, visualizing both parties' satisfaction in a utility space (= a space where each party's satisfaction with each proposal is quantified and made comparable as coordinates). It evaluates proposals using metrics like envy-free and Pareto efficiency while offline verifying and updating prompts/workflows based on negotiation logs. With each planned release, it enhances estimation accuracy and field applicability, reaching viable solutions with fewer iterations.
The benefits of Adaptive Evolution Technology
- Fair and Satisfactory Agreement Forms
- Achieves both fairness and efficiency based on game theory metrics like envy-free and Pareto efficiency, adding transparency to condition negotiations prone to personalization
- Accelerated Agreement Formation and Quality Stabilization
- Objectively compares candidate proposals in the Belief Space × Utility Space, quickly narrowing down acceptable options
- Planned Updates via Log Utilization
- Analyzes collected history to improve field suitability and success rates through periodic releases
- Ensuring Explainability
- Combines LLM outputs with mathematical evaluation (efficiency/fairness) to visualize and explain proposal rationale
|
|
|
|
|
|
|
|
|
|
|
|
Demo
App
The demo app is currently undergoing maintenance
API The Web API is currently undergoing maintenance
Technical Overview
Target Industry/Users
Purchasing and negotiation personnel in retail (supermarkets, department stores, convenience stores, drugstores, etc.)
※ Same technology can be horizontally deployed to negotiation/coordination tasks like lease renewals, supply-demand adjustments, internal allocation, and vendor selection
Challenges in Target Industry and Operations
- Retail MD cycles broadly fall into upstream (product development/sales prep), midstream (supply-demand planning/procurement), and downstream (sales/customer touchpoints)
- While companies want to focus on upstream/downstream to differentiate themselves, the midstream becomes a bottleneck, preventing this—a common industry-wide challenge
- Particularly, the personalization and complexity of transaction terms make fair and transparent negotiations difficult
Technical Challenges
- Estimation accuracy of counterparties' values and priorities under incomplete information
- Lack of fairness and explainability when relying solely on LLM
The benefits of Adaptive Evolution Technology (Details)
Fairness Assurance: Utilizing game theory-based metrics like envy-free and Pareto efficiency enables consensus formation without dissatisfaction. This enhances transparency across condition negotiations—which often fall into individual optimization—while mitigating the impact of personalization.
Fujitsu's Technological Advantages
- Self-evolving mechanism that improves prompts and workflows by leveraging negotiation logs
- Transparency and explainability through combining LLMs with mathematical reasoning
Use Cases
- End Users
- Buyers and sales representatives conduct fair and efficient negotiations via AI agents
- Application Developers
- Integrate into negotiation support apps for the retail sector, customizing for specific business operations
Case Studies
- Proposal in preparation
Technical Trial
- Demo App: The demo app is currently undergoing maintenance
- API: The Web API is currently undergoing maintenance
- A Proof of Concept (PoC) is possible