How AI Can Transform the Way Companies Buy What They Need

Introduction

A new paper from NEC Corporation and NEC Laboratories America proposes a framework that addresses both challenges at once, combining automated negotiation with AI-powered demand forecasting to make procurement smarter from start to finish. The paper, “Automated Negotiation and Multimodal Time-Series Forecasting for Efficient Procurement,” was presented at the 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026). It was authored by Yasser Mohammad of NEC Corporation and Haifeng Chen of NEC Laboratories America, Inc. The work comes from NEC Labs America’s Data Science and System Security department.

The Supply Chain Problem

Procurement teams face two distinct but connected challenges. The first is forecasting: knowing how much of something an organization will need, and when. Most existing forecasting tools rely on historical purchase data alone. But real-world demand is shaped by a much wider range of signals, including supplier lead times, market pricing trends, seasonal patterns, news events, and operational changes inside the buying organization. A system that ignores those signals will produce forecasts that are consistently off, leading to either costly overstock or disruptive shortages.

The second challenge is negotiation. Most supplier negotiations are conducted manually, with procurement staff going back and forth on price, volume, and delivery terms through email threads and spreadsheets. This process is time-consuming, hard to scale, and heavily dependent on the skill and bandwidth of individual buyers. In large organizations that manage hundreds or thousands of supplier relationships, the manual approach creates a practical ceiling on how well procurement can perform. The framework the authors propose addresses both problems through two integrated components.

The Multimodal Time-Series Forecasting System Component

The multimodal time-series forecasting system uses historical data sequences to predict future values, much like a weather model that uses temperature readings from past weeks to predict the temperature next Tuesday. The word “multimodal” here means the system draws on multiple types of data simultaneously, not just past purchase volumes but also external signals such as pricing indices, supply chain news, and operational metrics. By fusing these diverse data streams, the forecasting component produces demand predictions that are more accurate and more responsive to real-world conditions than single-source models.

Automated Negotiation Agent Component

Negotiation in this context is treated as a structured problem that software can solve, given a target quantity, a budget, and a set of constraints. The automated negotiation agent engages with suppliers to work toward an optimal contract. Rather than requiring a human buyer to manage every exchange, the agent handles the iterative process of offer and counteroffer, guided by the demand forecasts from the first component. The two systems are designed to work together: better forecasts lead to better negotiating positions, and the negotiation outcomes feed back into the planning process. The authors validated the framework through two case studies using simulations based on real-world procurement data.

The Results

The authors evaluated the framework across five commodity products, measuring inventory reduction relative to three baselines: an oracle condition representing perfect demand knowledge, a negotiation-only approach using mean-based forecasting, and a standard single-modality forecasting model, DLinear.

The framework could reduce inventory by up to 29% by optimizing delivery schedules and quantities through automated supplier negotiations, thereby reducing the carrying costs associated with excess stock.

For directly traded commodities, the multimodal approach came close to or matched oracle performance in every case. For petroleum, inventory reduction reached 24% compared to 10% for negotiation-only. Gas improved from 10% with negotiation-only to 29% with the full multimodal framework, against an oracle ceiling of 34%. Silver reached 20%, matching the oracle exactly, up from 8% with negotiation-only. Gold also matched its oracle figure of 17%, versus 8% for negotiation-only.

The second case study tested a more complex indirect forecasting scenario in which 20 variables, including energy prices, metal prices, technology stocks, cryptocurrency values, and public health data, each influence the procurement plan. The multimodal framework achieved 23% inventory reduction, matching oracle performance and improving on negotiation-only by 15 percentage points. This result is particularly significant because it demonstrates the framework holds up when demand is shaped by a wide web of external factors rather than a single, legible signal.

Real-World Applications

The most direct application is enterprise procurement for large manufacturers. A company sourcing steel, electronics components, or raw materials across a global supplier network could use this framework to automate routine negotiations while simultaneously improving the demand forecasts that drive those negotiations. The result is faster cycle times, reduced reliance on individual buyer judgment, and better alignment between what gets ordered and what actually gets used.

Healthcare supply chain management is another area where this work applies directly. Hospital networks and group purchasing organizations manage contracts for thousands of products across medical supplies, pharmaceuticals, and equipment. Demand in healthcare is particularly difficult to forecast because it is influenced by patient volume, clinical protocols, and regulatory factors that conventional purchasing models do not capture well. A multimodal forecasting system paired with negotiation automation could help these organizations reduce waste and negotiate stronger terms without adding procurement headcount.

Retailers and distributors that manage seasonal inventory and government agencies that handle large-scale infrastructure procurement represent further use cases where the combination of smarter forecasting and automated negotiation would yield measurable operational improvements.

“Procurement is often treated as an operational necessity rather than a source of competitive advantage, but that view underestimates how much value is left on the table. When you give an organization the ability to forecast demand more accurately and negotiate at scale, you change what procurement can deliver for the business. This work shows a concrete path to that outcome.” Haifeng Chen, Data Science & System Security Department Head

Looking Ahead

The authors outline future directions, including expanding the framework to handle more complex, multi-party negotiation scenarios and integrating additional data modalities to improve forecasting accuracy further. As supply chains grow more interconnected and more volatile, the ability to automate both sides of the procurement equation, what to buy and what to pay, will become an increasingly important capability for organizations of all sizes.