Chief procurement officers (CPOs) today face two seemingly incompatible pressures. On the one hand, they’re being asked to reduce business spending. On the other, they increasingly need to be ready to change quickly in the face of sudden global trade changes that affect sources of goods.

To cut costs, they’ll need to spend more time on strategic activities like negotiating contracts, exploring and vetting lower cost suppliers, and ensuring supplier adherence with contracted performance terms. But the increased workload won’t come with more resources. A 2025 Hackett Group survey of procurement leaders found they expect their workload would rise by 9.8% this year, but with only a marginal increase in staffing.

Given these ongoing pressures, automation is a natural area for CPOs to seek support. And artificial intelligence agents have something to offer. Agentic AI systems incorporate intelligent, often autonomous AI agents that can understand natural language, bridge information gaps, integrate across systems, and even take actions.

Agentic AI is one of the three advancements that will shape the future of procurement, according to analyst firm Gartner Inc. The other two are the use of agentic reasoning to provide suggestions to humans and the ability of generative AI (a precursor to agentic AI) to integrate and process multiple forms of data to answer questions.

The rollout of agentic AI in procurement is still in its early days, says Amy Hillcox, senior research director of procurement and purchase to pay advisory at the Hackett Group. The highest rate of adoption is in payables management, according to Hackett research, with 21% of companies using agentic AI in production environments. Hillcox says she expects a big increase in the number of agentic AI pilots this year, driven by its potential to streamline procurement and an increasing number of off-the-shelf agentic AI capabilities. These include prebuilt and trained large language models (LLMs), prebuilt procurement agents, and easy-to-use agent creation tools.

“AI agents offer a new, very flexible way of patching any problems, either by automating steps, or by handling exceptions more flexibly,” says Timo Elliott, a global innovation evangelist for SAP.

Agentic systems at work in procurement

One example of how agentic AI can streamline procurement is payments. Suppose a supplier e-mails a procurement team about a late invoice. An AI agent could read this e-mail, understand its contents, check with sub-agents to verify the claims in the e-mail that (a) the goods were delivered, (b) the invoice is eligible for payment, and (c) payment has not yet occurred. Then the agent comes up with a selection of possible actions for a procurement professional to choose from.

Agentic AI applications can also reduce procurement costs by whittling down the list of suppliers the company uses to identify the lowest-cost partner. To do so, they would scan supplier information in company e-mails, PDF files, databases, and websites, compare and choose among various suppliers and payment terms, recommend a supplier, and generate a detailed written report on their suggestion. Agents could also track supply levels and automatically order new stock, as well as optimize delivery routes to cut costs and environmental effects, triggering workflows for human review where needed.

While such examples drive short-term cost savings and reduced rework, agentic AI can also deliver more strategic benefits. These include quickly finding new suppliers, speeding the replacement of suppliers that have become too expensive or unreliable, and providing greater insights on spending, supplier performance, and alternative suppliers.

Some companies are already allowing fully autonomous agents to negotiate and agree on terms for “long-tail spend”: routine, recurring small purchases such as for office supplies or landscaping that currently lack any oversight. “We’re seeing customers in Europe who are starting to leave the sourcing and procurement process for low-value items and services completely to agentic AI,” says Christoph Menne, a partner with Germany-based management consultancy apsolut Group.

Such examples point to the potential for using agentic AI to orchestrate the automation of business processes.

Orchestration—the simplification and coordination of complex procurement processes—is an ideal use for agentic AI when used to manage suppliers, says Pierre Mitchell, chief research officer and managing director at consultancy Spend Matters. Mitchell, during a 2025 SAP webinar, said he encouraged procurement leaders to “automate as much as you can” with agentic AI, “so you can spend not 2 hours, but 40 to 50 hours with your most strategic suppliers.”

Where to start with agentic AI: How procurement people work

Realizing the promise of AI agents requires a careful analysis of business processes, in particular the human factor, says Hillcox. To identify a good use case to begin with, she recommends examining the experiences of procurement professionals. Ask questions like, “How well is your procurement team working with stakeholders? How well are you automating front-end functions—such as purchase order processing—and back-end payment functions?”

She also advises going beyond automating existing processes to imagining “what really cool things we can automate better,” such as using agentic AI to suggest negotiating strategies with major suppliers—an effective approach to controlling costs.

Such uses of agentic AI will lead to changes in the work procurement teams do, as well as the skills required. For example, as agentic AI automates routine data entry and forms processing, procurement specialists “are going to start spending more time on negotiations with suppliers or category management,” Hillcox says, referring to the practice of grouping similar goods and services into categories to better understand and control costs for each category. That will require soft skills to work more effectively with other groups within the organization and with suppliers, she says.

Implementation challenges for agentic AI in procurement

Successful implementations of agentic AI will require the right talent and strong business processes. Menne of apsolut Group notes that he has seen CPOs adding AI experts and data analysts to their teams.

But without well-managed, high-quality, and secure data, CPOs are likely to encounter mixed results. As with every application of artificial intelligence, the details are in the data:

Data management: Make it discoverable. A modern data management infrastructure aids discovery, making it possible to access data needed to train LLMs that will deliver accurate insights. Gartner Inc. recommends CPOs double down on data governance that collects procurement data from internal and external sources.

Mitchell, of Spend Matters, recommends including unstructured data—such as from customer support chats and contracts—in the mix. He also suggests using a technique called GraphRAG (a form of retrieval-augmented generation), which integrates AI models with external data sources to help procurement staff map the relationships among data sources on a graph.

Managing this data well will make it possible to identify the data that can best train AI agents to improve business processes. To minimize the cost and effort of such training, Menne of apsolut Group recommends using LLMs from major software vendors that are pretrained to handle procurement issues.

Data quality: Make it usable. While not a new challenge, ensuring that procurement data is clean and categorized for analysis and discovery remains a top concern. In a 2024 McKinsey survey, even CPOs who say they have built a single data “source of truth” acknowledged their data is not cleaned and categorized, may lack data about purchase quality, and might not include external data from suppliers and customers.

Data privacy and security: Mitigate risks. As with any IT innovation, it pays to consider risk factors. Agentic AI poses many of the same privacy, security, and compliance challenges as other types of AI (such as guarding against bias, protecting individuals’ personal information, and avoiding the use of stolen data).

But there’s more to consider. Like LLMs, some agents transmit data to the cloud for processing, which may expose the data to unauthorized third parties, writes Daniel Berrick, a senior AI policy counsel at the Future of Privacy Forum. Berrick also notes that AI agents can be susceptible to prompt injection attacks, a technique in which LLMs receive malicious prompts that cause them to misbehave, for example, by installing malware or redirecting them to deceptive websites.

CPOs would be wise to confer with their legal, compliance, and information security teams to determine their best approach, and to draft privacy policies to share both inside the organization and with suppliers.

Automation is not new. Artificial intelligence applications for procurement—including those using machine learning, robotic process automation, optical character recognition, and generative AI—have all made a mark. But automation is expanding with agentic AI. With smart data management and effective deployment, AI agents have the potential to strongly support CPOs by reducing procurement costs, reducing errors, and finding insights that can uncover new business opportunities.

Read more