In 2025, Walmart deployed an AI chatbot to negotiate with its long-tail suppliers. The bot approached thousands of vendors, closed deals with 68% of them, and delivered approximately 3% average savings. Suppliers rated the bot easy to use 83% of the time. The entire operation ran without a human in the negotiation loop.
This is not a pilot. Walmart, Maersk, Vodafone, Henkel, Linde, and Wesco are all running autonomous AI negotiation agents in production on the Pactum platform. McKinsey reports that AI-guided negotiations at one pharmaceutical company delivered 10 to 15% savings across vendors while cutting the time negotiating teams spent on analysis and email by up to 90%. Agentic AI in procurement has moved past the proof-of-concept stage. The question is no longer whether it works — it is how to deploy it without breaking the operating model.
What agentic AI actually does in procurement
Generative AI produces content. Agentic AI produces outcomes. The distinction matters because the two technologies require fundamentally different governance.
A generative AI tool can draft an RFP or summarize a contract clause. An agentic AI system negotiates the terms, updates the contract, posts the agreed rate into the sourcing system, and triggers the purchase order — all without human intervention at each step. It operates within guardrails set by procurement: target prices, minimum margins, acceptable trade-offs between payment terms and volume discounts, and escalation rules for out-of-bounds proposals.
Pactum's system, for example, conducts thousands of parallel chat-based negotiations with suppliers. Each conversation is autonomous but bounded. The agent can concede within predefined ranges, evaluate trade-offs across cost, service level, and risk, and generate counteroffers automatically. When a negotiation hits the guardrail boundary, the agent escalates to a human buyer. An audit trail logs every decision the agent made and why.
The capacity argument no one is making
The dominant narrative around AI agents focuses on savings. McKinsey's 10-15% figure is impressive. But the bigger story is capacity, not price.
A procurement team managing 500 tail-spend suppliers typically cannot negotiate with all of them. The cost of human effort per negotiation exceeds the expected savings from any single vendor. So the tail goes unmanaged. Enterprise suppliers get attention and long-tail suppliers renew at last year's rates plus inflation. That is not a negotiation failure. It is a capacity ceiling.
AI agents erase that ceiling. Walmart's bot does not replace a buyer who negotiates with top-50 strategic suppliers. It does what no buyer could do: negotiate with thousands of suppliers simultaneously. The same logic applies to logistics spot rates at Maersk, where Pactum's agents handle freight lane negotiations in real time, and to contract compliance enforcement at the pharmaceutical company McKinsey documented, where agents scan every invoice against negotiated terms and flag deviations automatically.
"An AI agent running 1,000 simultaneous negotiations is not a buyer replacement. It is a category of work that procurement has never been able to perform before."
Wharton research shows that weekly generative AI use in procurement jumped 44 percentage points from 2023 to 2024. By early 2025, 94% of procurement executives reported using generative AI at least once a week. Agentic AI adoption is behind that curve — the technology is newer, the governance is more complex, and the integration requirements are steeper — but the adoption trajectory looks similar.
Three deployment patterns that work today
Agentic AI deployments cluster into three patterns. Each has different requirements, risks, and returns.
AI agents negotiate with long-tail suppliers where human effort does not justify the return. Wide latitude on terms. High volume of parallel conversations. This is Walmart's model.
AI agents prepare negotiation fact bases, make real-time suggestions, and generate counteroffers. Humans remain in the loop for final decisions. McKinsey's pharmaceutical example fits here.
AI agents monitor every invoice against negotiated terms, detect deviations, and trigger corrections or escalations. Requires deep P2P integration. Enforces value after contract signature.
Pattern 1 is the easiest entry point. The stakes are low, the volume is high, and supplier relationships are transactional enough that automation creates no strategic friction. Pattern 2 requires integrating AI into the existing negotiation workflow — the agent supports, does not replace, the buyer. Pattern 3 is the hardest because it requires the AI agent to sit inside the procure-to-pay system and act on live transactional data.
The governance question most organizations are not asking
Every CPO considering agentic AI asks about savings. Fewer ask about governance. That is where the failures will happen.
An AI agent negotiating supplier contracts needs explicit guardrails: pricing floors, term boundaries, escalation paths, and exception handling protocols. Setting those guardrails requires category knowledge that most organizations have not systematically documented. If the guardrails are wrong, the agent either concedes too much (leaking value) or rejects too many proposals (wasting the capacity gain).
Three structural governance questions every deployment must answer:
- Escalation criteria. What conditions trigger a handoff to a human buyer? Out-of-range pricing? Novel commercial terms? A supplier requesting a meeting? The answer determines whether the agent is truly autonomous or just an expensive auto-responder.
- Audit trail completeness. Every concession the agent makes, every trade-off it evaluates, and every escalation it triggers must be logged and reviewable. Without this, procurement cannot defend the outcomes to finance or internal audit.
- Guardrail maintenance cadence. Supplier markets change. A price floor set in January may be structurally wrong by July. Who updates the guardrails, and how often? This is a new procurement role that most organizations have not created.
At Maersk, Pactum's agents handle spot trucking rate negotiations — a domain where prices shift weekly and the volume of transactions makes manual negotiation impossible. The guardrails must be updated continuously. The same is true for any category with volatile pricing.
What suppliers think about negotiating with bots
The 83% supplier satisfaction score from Walmart's deployment challenges a common assumption: that suppliers will resist negotiating with AI. The data suggests the opposite for certain categories.
Suppliers dealing with long-tail procurement often face the same capacity problem buyers do. A small supplier cannot afford to spend days negotiating a contract renewal for a few thousand dollars in annual revenue. The AI agent offers a faster, more predictable process. The supplier gets a clear set of parameters, can respond on their own schedule, and receives an immediate decision. There is no waiting for a buyer to review a counteroffer.
This dynamic does not hold for strategic supplier relationships. A $50 million contract with a critical chip supplier requires human judgment, relationship management, and contextual awareness that no current AI agent can replicate. The boundary between transactional and strategic negotiation is the key design decision.
What this means for procurement leaders
Agentic AI in procurement is real, it is deployed in production at enterprise scale, and it will become a standard capability within three years. The opportunity is not just cost savings — it is the ability to manage spend that procurement has never had the capacity to touch. But the governance, guardrail, and operating model questions must be answered before deployment, not after.
- Map your tail spend by negotiation cost. Calculate the cost of a human-led negotiation for each supplier segment. Any segment where negotiation cost exceeds expected savings is a candidate for Pattern 1 deployment.
- Document your negotiation guardrails by category. Before deploying an agent, codify the pricing floors, term boundaries, and escalation rules for each category. Incomplete guardrails will produce worse outcomes than no agent at all.
- Define the escalation workflow. Who handles out-of-bounds proposals from the AI agent? What is the service-level agreement for response time? Without this, the agent becomes a bottleneck creator rather than a capacity multiplier.
- Run a controlled pilot on one non-strategic category. MRO, logistics spot rates, and low-value professional services are good candidates. Measure savings, supplier satisfaction, and escalation rate. Use the data to build the business case for broader deployment.
- Create the guardrail owner role. Assign someone to maintain and update the AI agent's negotiation parameters. This is a new procurement role — category knowledge plus data literacy — and it requires a different skill profile than traditional buying.
What is agentic AI in procurement?
Agentic AI refers to AI systems that can act autonomously within defined guardrails to complete procurement tasks — including negotiating contracts, managing tail spend, and enforcing contract compliance — without human intervention at every step.
Which companies are using AI agents for supplier negotiations?
Walmart, Maersk, Vodafone, Henkel, Linde, and Wesco are among the companies using Pactum's autonomous negotiation agents for supplier contracts, logistics spot rates, and tail-spend management.
What savings do AI agents deliver in procurement negotiations?
McKinsey reports 10-15% incremental savings from AI-guided negotiations. Walmart's AI agent achieved approximately 3% average savings. The primary value is often capacity — AI can run thousands of parallel negotiations humans cannot.
What is the difference between generative AI and agentic AI in procurement?
Generative AI produces content — draft RFP language, supplier summaries, contract clauses. Agentic AI acts: it negotiates terms, updates systems, enforces compliance, and makes decisions within defined guardrails without human step-in.
Sources
- McKinsey — Redefining procurement performance in the era of agentic AI (2025)
- Pactum — Autonomous negotiation platform case studies (Walmart, Maersk, etc.)
- Art of Procurement — State of AI in Procurement in 2026
- IBM — The future of procurement: Moving beyond cost savings to AI-driven value creation
- Kodiak Hub — AI-Powered Procurement: From Strategy to Execution
- Raindrop — Show Me the Money: Hard-Hitting ROI from AI-Driven Procurement