Agentic AI is turning long-tail purchase orders into true cost savings

Agentic artificial intelligence is beginning to shift procurement from a function defined by insight and analysis to one increasingly driven by execution, particularly across the high-volume, low-value transactions that have historically gone unmanaged.

That was the message from Kaspar Korjus, co-founder and CEO of Pactum, who described how AI agents are already being deployed to autonomously manage portions of the procurement process, including supplier negotiations.

“Now the agents are covering end-to-end procurement,” Korjus told Supply Chain Management Review. “Some things are helping buyers, and in some instances they are doing everything automatically.”

The shift reflects a longstanding constraint in procurement operations. While organizations have invested heavily in analytics, visibility platforms, and sourcing tools, the ability to act on that insight, particularly at scale, has remained limited by human capacity.

From visibility to execution

Procurement teams have traditionally focused their efforts on high-value, strategic suppliers, leaving a significant portion of spend, often referred to as the “long tail,” largely unmanaged. The result is missed opportunities for cost savings, inconsistent contract compliance, and inefficiencies in requisition handling.

Agentic AI is designed to address that gap by executing routine procurement tasks autonomously. In practice, that includes validating requisitions, ensuring compliance with contract terms, and negotiating within predefined parameters.

The approach has already been tested at scale. In a widely cited example, Walmart partnered with Pactum to automate supplier negotiations. As reported by Harvard Business Review, Walmart identified that approximately 80% of its suppliers were not actively engaged in negotiations due to bandwidth constraints.

By deploying AI agents to manage those interactions, the company was able to extend its procurement reach without increasing headcount. That use case highlights a broader shift in how organizations are approaching procurement performance by enabling more consistent execution.

A phased approach to adoption

Despite the growing capabilities of agentic AI, adoption within procurement remains measured and incremental. Korjus noted that most organizations begin by applying AI agents to low-value, low-risk transactions, typically under $50,000. These transactions represent a high volume of activity but relatively limited downside risk, making them a practical starting point for automation.

“Usually, we kick off without humans in the lowest spend, lowest value negotiations,” he said. “Once the value is proven and trust is built, we scale.”

As confidence in the system increases, organizations expand the role of AI agents into more complex scenarios, while maintaining human oversight for high-value or strategic decisions. In many cases, procurement teams define thresholds within existing platforms that determine when AI can act independently and when human intervention is required.

Korjus emphasized that AI agents require onboarding similar to human employees, though.

“They are not magical,” he said. “You need to teach them where the data is, how decisions are made, and what the company values.”

This structured rollout reflects both the opportunity and the caution surrounding AI adoption in procurement, where risk management and supplier relationships remain critical.

Scaling small decisions for measurable impact

While much of the attention around AI in procurement has focused on strategic sourcing and large-scale negotiations, the primary value of agentic AI is emerging in its ability to manage high volumes of smaller transactions.

Organizations often process tens or hundreds of thousands of requisitions annually, many of which receive limited scrutiny due to resource constraints. AI agents can systematically evaluate and act on those transactions, improving pricing alignment, enforcing contract terms, and identifying opportunities for incremental savings.

“If you need to handle 200,000 requisitions coming in and someone has to clean that, the AI agent is making requisitions clean and according to compliance recommendations,” Korjus said.

The financial impact of those improvements is cumulative. While savings on individual transactions may be modest, applying them across a large volume of activity can produce significant results in aggregate, including cost reductions, improved working capital, and better payment terms. Even a $100 savings on a low-value contract, multiplied across thousands of contracts yearly, can result in significant savings.

Redefining the role of procurement

The introduction of agentic AI is not eliminating the need for procurement professionals, but it is changing how their time is allocated. As routine transactional work is increasingly automated, procurement teams are able to focus more on strategic initiatives, supplier relationships, and complex negotiations that require human judgment.

At the same time, the technology is prompting organizations to reconsider how procurement functions operate at scale. Rather than limiting engagement to a subset of suppliers, AI enables broader coverage across the entire supplier base.

That shift aligns with earlier discussions in Supply Chain Management Review around the evolution of AI-driven procurement. In The Invisible Handshake: AI-to-AI Procurement Negotiations, it was noted that automated negotiation capabilities is expanding to interactions between AI systems on both sides of a transaction, further increasing efficiency and reducing friction.

While that scenario is still developing, the current trajectory suggests that AI will continue to expand its role in procurement execution, particularly in areas where processes are structured, repeatable, and data-driven.

Balancing automation and control

Despite the progress, organizations remain cautious about extending AI into high-stakes procurement decisions. Strategic sourcing events, critical supplier relationships, and large contract negotiations continue to require human oversight.

Korjus acknowledged that adoption timelines vary depending on the complexity and risk of the application. Simpler use cases, such as data validation or compliance checks, can be implemented quickly, while more advanced applications may take months or longer to gain organizational acceptance.

“Usually things are in the extremes,” he said. “Some agents can give value within weeks. For more critical parts, adoption is slower because trust needs to be built.”

That balance between automation and control is likely to define the next phase of AI adoption in procurement as organizations seek to capture efficiency gains without introducing unnecessary risk.

A shift already underway

While agentic AI is often framed as an emerging technology, its application in procurement is already moving beyond pilot programs and into production environments.

The key distinction is not the presence of AI itself, but its role in the process. Rather than simply providing recommendations, agentic systems are increasingly responsible for executing decisions within defined boundaries.

For procurement leaders, that shift represents both an opportunity and a challenge: the opportunity to extend operational reach and improve consistency, and the challenge of redefining how decisions are made, governed, and measured in an increasingly automated environment.

As organizations continue to test and scale these capabilities, the focus is likely to remain on a familiar constraint, how to move from knowing what to do to actually doing it.

SC
MR