Supply Chain Intro and Table of Contents Supply Chain August 2026 | Page 110

Automated Negotiation for Future Supply Chains

A practical design choice in this deployment is an AI – Human negotiation configuration: the AI agent resides on the buyer( procurement) side while the supplier side continues to negotiate through human staff, via a chat interface or email notifications. This design imposes no IT-readiness requirement on the supplier side— no API integration, no agent deployment— which is what makes rollout across the tens to hundreds of heterogeneous suppliers in a typical procurement organization tractable in practice. A fully agent-to-agent configuration remains a longer-term direction; the current design prioritizes deployability. Rather than exchanging single date-quantity offers, the buyer-side AI presents graded concession proposals that include partial-shipment options, broadening the agreement region; cases in which no feasible agreement is identified are escalated to human staff( human-in-the-loop). The AI is integrated with the existing procurement backbone, reading inventory and demand information in real time so that the negotiated outcome is immediately reflected in the procurement plan.

Two real-business evaluations at an NEC group company have been reported. The 2023 proof-of-concept was run with two suppliers across two product types( cables and electronic devices) and two trigger conditions: " unsatisfactory schedule "( the supplier ' s promised dates would lead to negative projected inventory) and " supply – demand matching "( the buyer initiates re-negotiation after a demand change). Median negotiation time fell from a baseline of three hours to two days down to approximately 77 seconds across the four conditions, with an average closure rate of 67.7 %. The PoC also exposed a weakness in the matching condition for devices that motivated a re-tuning of the utility function and the negotiation strategy [ 3 ].

The 2024 real-business deployment was an order of magnitude larger: 10 suppliers and 1,300 product types, with 140 actual negotiations recorded across the same four conditions. Table 5-1 reports the results.

Median time( with AI)

Closure rate

Unsatisfactory schedule

Unsatisfactory schedule

Supply – demand matching

Supply – demand matching

Total / average All 3h – 2d 1 min 17 sec 95 %( 133 / 140)

Table Five Deployment Patterns-1: Results of the 2024 real-business deployment of NEC ' s NAP at an NEC group company( 10 suppliers, 1,300 product types). Adapted from Ando et al. [ 3 ].

106