From automating nearly half of incident-response workflows to creating a trusted view of sourcing data, industry leaders reveal how agentic AI in procurement and supply chain management is revolutionizing operational performance.
DECEMBER 2025
In December 2025, the MIT Center for Transportation and Logistics held its biggest roundtable for supply chain partners yet, bringing together industry leaders, researchers, and practitioners to explore how agentic AI is being applied across procurement operations.
Key information
90+ supply chain practitioners
Real implementations shared by industry leaders
10 breakout discussion groups
Use cases spanning cold chain, sourcing, freight, customs, optimization, and governance
In the implementation experiences shared by leaders from various industries, one message stood out:
Many organizations already have valuable data and systems in place, but still struggle to turn fragmented information into fast, confident action.
Read on for detailed insights.
The cost of inaction:
Manual effort and delayed decisions
Reactive exception management
Missed optimization opportunities Compliance and trade exposure
Examples shared during the session included customs and trade settlements or penalties ranging from $10M to $365M.
Deep Knowledge Lab for Supply Chain and Logistics Director Dr Elenna Dugundji leads a discussion at the Agentic AI Roundtable
The Three Pillars Of Success
Across the roundtable, three themes appeared repeatedly in successful adoption of agentic AI in procurement:
Pillar 1: Data as Foundation
Successful teams are building a consistent data foundation before scaling AI. Core data products included spend, supplier, contracts, and category.
Pillar 2: System Alignment
AI is beginning to act as an orchestration layer across ERP, WMS, and TMS — helping organizations make better decisions across systems, not just within them.
Pillar 3: Human-Centric Governance
Successful deployments are designed around existing user behavior, with clear governance over when humans stay in the loop.
Real-World Outcomes
Participants shared results from their successful implementations of agentic AI:
Cold chain logistics
About 40% fewer notifications requiring response
Under five minutes to diagnose, evaluate, and act in one workflow
Broader automation depended on 85–90% realtime IoT adoption
Incident resolution
1.3M+ incidents annually
44% fully automated
Roughly 2,000 incidents per day handled automatically
About 30–40 team members’ worth of work automated
Procurement intelligence
Historical RFX knowledge became easier to access and reuse
Output was improved through prompt engineering
Confidence increased when answers linked back to source materials
Unlocking Value
Participants described a consistent pattern of value across a wide range of use cases:
Reducing manual work
44% of incidents automated, 2,000 incidents handled per day, and 5% of freight workflow emails processed by agentic AI
Accelerating decisions
Under five minutes to diagnose and act in one coldchain deviation workflow, with faster access to tariff and sourcing intelligence
Improving visibility
More than 98% consistency in sourcing data in one case, and unified views across logistics, quality, and packaging data in another
Strengthening compliance
Participants emphasized human review, access controls, audit trails, and explainability — especially in high-stakes environments
Barriers and Solutions
Legacy
Lock-In
Organizations are not replacing core systems. They are layering AI on top of ERP, WMS, TMS, and procurement platforms to improve decisions.
Data Sensitivity and Governance
Data Quality
Disconnect
In regulated environments, deployment depends on human oversight, traceability, access controls, and auditability.
Participants repeatedly emphasized that AI exposes existing weaknesses in data and process design. Stronger taxonomy, clearer data products, and better consistency came first.
One participant put it simply: AI did not create security or data issues, it magnified problems that were already there.