Agentic AI and Procurement (Part 2): Why it’s worth it

Agentic AI is sometimes framed as a vision of the future, but its strongest value proposition is rooted in what it is delivering today. Procurement leaders are not waiting for the technology to mature. They are applying it to real problems and capturing measurable value. (Read part 1 – What is Agentic AI?)

However, to understand the value of Agentic AI, it is essential to distinguish between genuine autonomy and enhanced automation. Unlike systems that follow static workflows, agentic frameworks enable LLMs to dynamically determine which tools to use, what steps to take and how to adapt based on evolving goals and context.

To understand this shift, one should step back and consider where and how Agentic AI is driving returns. What makes these returns particularly compelling is their compounding nature. Because agentic systems adapt based on feedback and learn from past executions, performance improves over time, driving progressively higher returns across multiple dimensions. Figure 1 shows the four dimensions that stand out across current deployments.

Figure 1

The table below outlines these dimensions and how Agentic AI contributes to each:

ROI dimension	Agentic AI contribution
Efficiencies	Reduces manual effort by automating data handling, approvals and intake-to-pay flows.
Effectiveness	Improves process outcomes through more consistent, accurate execution.
Innovation	Enables new workflows, such as autonomous sourcing and proactive risk mitigation.
Digital knowledge management	Builds institutional memory and insights through agent learning and adaptation.

Efficiencies

Efficiency that leads to cost reduction remains one of the clearest, most immediate benefits of agentic AI. These improvements do not only come from removing manual steps. They result from the agent’s ability to orchestrate tasks across systems and APIs with its own reasoning. That is a key difference between legacy orchestration capabilities and today’s: automation follows instructions; agents interpret goals and decide how to achieve them.

Earlier this year, IBM reported that companies using AI-driven category intelligence cut procurement costs by 40%–70% in under six months. One company alone prevented $70 million in duplicate or erroneous payments by deploying agents to review agreements.

Effectiveness

Agentic AI improves more than just speed; it transforms process reliability. IBM also reported that its deployment of sourcing agents led to a tenfold improvement in supplier onboarding time and cut pricing analysis from two days to ten minutes.

In a more general study, Deloitte’s 2024 Global CPO GenAI survey found that 50% of early adopters more than doubled their ROI than traditional tech. More than one in five reported plans to allocate over $1 million to agentic and generative AI initiatives.

Innovation

Agentic AI does not stop at streamlining and speeding up old workflows and processes. It opens the door to entirely new ones. Not all implementations will begin at the high end of the autonomy spectrum, though. Agentic capabilities can scale gradually from semi-automated interventions to fully autonomous execution. What matters most is designing systems to evolve with learning, context awareness and increasingly complex reasoning.

Digital knowledge management

The most underappreciated benefit is how agents learn over time. As they handle sourcing, risk and compliance scenarios, they build organizational knowledge that remains intact when someone changes roles or leaves the company. Instead of relying on tribal knowledge or rigid rules, procurement can draw on systems that improve through experience.

Dimension	Organizational knowledge	Tribal knowledge
Definition	Documented and structured knowledge the entire organization can access	Informal, unwritten know-how held by individuals or small groups
Format	Manuals, playbooks, databases, training modules, agents	Conversations, habits, personal notes, experience-based shortcuts
Access	Openly available across teams or functions	Known only by a few people; often hard to find or uncover
Governance	Managed, updated and audited by designated roles or processes	Uncontrolled; evolves organically without oversight
Scalability	Easy to scale and apply consistently across teams or locations	Difficult to scale; may not transfer well to others
Resilience	Survives staff turnover and supports long-term continuity	At risk of disappearing when key people leave
Use in decision-making	Provides a shared reference point for consistent, data-driven decisions	Based on memory or instinct; decisions vary between individuals
Impact on risk	Reduces risk through consistency and transparency	Increases risk due to gaps, inconsistencies and dependence on people
Change readiness	Supports change by providing a baseline for improvement	Hinders change; tacit knowledge is hard to capture and standardize
Example (Procurement)	A documented supplier onboarding checklist in the P2P system	A senior buyer’s unwritten tips on which suppliers are easiest to work with

A strategic investment

Adoption of GenAI and Agentic AI is accelerating. According to the Hackett Group, 64% of procurement leaders believe these technologies will fundamentally reshape workflows by 2030 and nearly half ran pilots in 2024. Hackett also adds that 42% of procurement teams plan to invest in new GenAI technologies in 2025 and that 33% intend to upgrade existing tools. 

Organizations are choosing to invest in agents because of these value drivers and of Procurement’s ever-increasing workload. In the same report, Hackett projects a 10% increase in procurement workload in 2025 but just 1% in budget growth. This represents a 9% efficiency gap that GenAI and agentic systems are expected to help close.

These trends also impact:

• Talent: As agents take on repetitive work, procurement professionals are leaning or will lean further into strategy, policy and governance, further moving the function away from control and compliance toward orchestration and insight. 
• Technology landscapes: Architecturally speaking, agentic AI requires more than just plugging in AI into existing tools. Procurement platforms must evolve to support persistent context, dynamic tool invocation and flexible error handling. Choosing frameworks that expose this control (instead of hardwired/static workflows) is key to scaling safely and effectively. 

While these numbers may sound attractive, it must be noted that Agentic AI does not guarantee maturity or readiness. These systems do not come with a magic wand. Their success depends on data consistency, process discipline, change readiness and clear governance. Many organizations will find that their current infrastructure unready to support persistent state, dynamic tool invocation or autonomous escalation. Without the right foundations, even the most advanced agents will fall short of expectations.

The rest of our Agentic AI mini-series will cover:

• Use cases
• What it takes to make it work
• The roadmap
• The agentic frontier — what’s next?