Real-time decision making for on-the-spot orders is considered using a novel distributed agent architecture. The order agents will have the ability to perform multi-stage negotiation using continuous double auction market-based model representing resources as producers and orders as customers. Qualitative information of resources and order requirements is captured in a Quality Function Definition (QFD) decision table. The QFD decision table also provides the interface for the agent to evaluate objectives, attributes and user preferences relating to resources and order requirements. This directs the negotiation against the most applicable resources at each stage through the market. In particular, the agents can learn from experience and adapt to the market change. The operation of such a system is illustrated within a garment industry application context.