Agentic AI and Procurement (Part 3): What Agentic AI Can Actually Do

For procurement leaders evaluating agentic AI, the most pressing question is no longer “What is it and does it drive returns?” but “What can it actually do today?” The term ‘agent’ is often used loosely to describe any automation feature with embedded intelligence. However, true agentic AI goes further: agents are defined by their ability to dynamically determine how to solve problems by selecting the most relevant tools (including other agents), adjusting steps and executing decisions based on evolving inputs and goals. That distinction is crucial when evaluating the business case and identifying use cases.

Part 3 of our agentic AI mini-series examines each stage of the procurement lifecycle, highlights agentic-enabled opportunities and presents examples of how one or more agents might operate in practice.

Purchase requisition and intake

Agentic AI can automate and augment the entire intake process. Intake agents can interpret incoming purchase requests (emails, forms or portals), classify them and enforce policy compliance. They can check if a request exceeds budget thresholds or involves off-contract items, flag missing information and automatically route the request to the correct approval workflow. The agents can even generate missing data by querying internal systems, e.g., retrieving past order history, and sending clarifying questions. With orchestration, multiple agents coordinate: one agent validates the request, another assigns the cost center and category and a third triggers approvals or purchase order (PO) creation as needed.

Illustrative example: A procurement intake agent detects a new request for software licenses. It autonomously checks the company’s contract database and sees an existing enterprise agreement. The agent then classifies the request under the proper category, fills in default terms from that contract and routes it directly to auto-approval since it meets policy. Meanwhile, a compliance agent ensures the purchase conforms to guidelines, e.g., the vendor is on the approved list. By handling these steps automatically, agents eliminate routine back-and-forth, reducing lead time from request to order while enforcing governance in real time.

Sourcing and negotiation

Agentic AI can manage sourcing by constantly scanning markets and data to find optimal suppliers and opportunities. A sourcing agent can access real-time market data, historical spend and supplier performance information and then identify best-fit vendors or product alternatives. For example, a sourcing agent might automatically analyze demand forecasts and market prices, detect price variations across suppliers or goods and suggest volume-based discounts or substitute products to cut costs.  Such an agent could then compile a shortlist of suppliers (even those outside existing panels), draft an RFx and send it to the potential companies. Over multiple execution cycles, it would ‘learn’ from past outcomes to refine criteria and target suppliers more effectively.

Illustrative example: Suppose a company needs to source capacitors. A sourcing agent continuously monitors industry news and market databases/catalogs, notices a surplus in a particular region and proactively issues an RFx to high‑quality suppliers in that region. It automatically compares incoming bids against historical and market costs and quality data, identifies the lowest price within risk tolerance and may even negotiate minor price improvements with the supplier. The agent could then generate a contract signature if thresholds are met. This agentic approach accelerates sourcing cycles, identifies savings by identifying volume discounts or alternatives and adapts the sourcing strategy as market conditions change.

Supplier discovery and onboarding

Agentic systems can accelerate the process of finding and vetting new suppliers. A supplier discovery agent continuously crawls public data and supplier databases for companies that match sourcing needs. It might use machine learning to detect and flag promising new vendors or alternatives, e.g., a startup with better technology. Once identified, an onboarding agent automates pre-qualification: sending out forms, verifying documents (e.g., insurance) and cross-checking references. These agents collaborate to cut through the ‘long tail’ of potential suppliers.

Illustrative example: When a business suddenly needs eco-friendly packaging materials, a discovery agent scours sustainability forums and supplier networks, finds a certified recycled plastic vendor not in the current panel and alerts procurement. A partnered onboarding agent then initiates the vetting process by auto-filling data known from internet profiles, sending a compliance questionnaire and integrating the new supplier into the ERP once approval is received. This orchestration can shorten supplier discovery and qualification time by orders of magnitude and expand the supplier pool with minimal manual effort.

Contract negotiation and compliance

Agentic AI can both negotiate and ensure compliance at contract time. A negotiation agent can understand an organization’s overall business strategy and category-level expectations to identify business objectives and acceptable terms; it autonomously runs negotiation rounds with suppliers. Using historical contracts and market benchmarks, it can propose trade-offs (e.g., longer payment terms for a volume discount) or counteroffers. Simultaneously, a contract agent scans legal clauses in real time. It highlights deviations from preferred terms, flags hidden risks and even suggests redlines. Because these agents learn continuously, each negotiation becomes more efficient and aligned with best practices.

Illustrative example: For a recurring services contract, an autonomous negotiation agent first evaluates a supplier’s offer against historical pricing and market indices. When the supplier counters, the agent uses predictive analytics to anticipate further counter-responses and automatically adjusts its strategy. It may accept a higher service fee in exchange for a penalty waiver if data show an overall value gain. Meanwhile, the contract agent reads every draft clause, comparing it against regulatory checklists and company policy. It immediately flags any non-compliant language, such as a cancellation clause above the risk threshold, and suggests alternatives. By the end, the combined agent team delivers a final contract draft with optimal terms and built-in compliance, often completing the process in days that are significantly faster than traditional cycles.

Risk management

Agentic AI can enable proactive, continuous risk management throughout procurement. Risk monitoring agents can ingest diverse data (news feeds, sanctions lists, supplier financials, ESG reports) and model complex risk scenarios. They continuously score and flag suppliers on multiple risk dimensions (financial stability, geopolitical exposure, ESG compliance). When a high-risk signal like a credit downgrade appears, these agents automatically alert stakeholders and trigger mitigation, such as finding alternative suppliers or adjusting order schedules. By predicting problems before they occur, agentic AI turns risk management from reactive to pre-emptive.

Illustrative example: A supplier risk agent tracks a key component maker in real time. It notices rising news about a raw material shortage that would affect that company. The agent correlates this with shipping delays in the supplier’s data and downgrades their risk score. It then proactively alerts sourcing teams (or discovery and sourcing agents) to find and qualify an alternative supplier. Similarly, if a supplier fails a quality audit, the agent could automatically reassign future orders to other approved vendors. By handling risk signals 24/7, such agents help ensure business continuity and compliance.

Spend analysis and optimization

Agentic AI can continuously ‘keep its finger on the pulse’ of spending data. A spend analytics agent aggregates POs, invoices and contracts in real time, looking for optimization opportunities. Unlike static reports, it can detect emerging trends or anomalies (such as a sudden price spike or unauthorized purchases) and act. For example, if it identifies two departments purchasing the same item from different suppliers at different prices, it can recommend or even initiate consolidation to secure volume discounts. It may also use sales and operations data to forecast demand and optimize inventory and procurement timing, thereby reducing carrying costs.

Illustrative example: An analytics agent notices that last quarter’s spend on office supplies jumped 15% in one division. Digging in, it finds many small off-contract orders. It then automatically nudges those buyers to a preferred contract supplier or even triggers a PO under the corporate agreement. Meanwhile, another agent analyzes seasonal demand patterns and vendor lead times to pre-order critical materials before prices rise, aligning order timing with market cycles. These agent-led optimizations capture savings that might otherwise elude human analysts, who review only after the fact.

Supplier performance management

Agentic AI can autonomously track and improve supplier performance. A performance agent continuously monitors quantitative and qualitative KPIs: delivery times, quality metrics, invoice accuracy and sentiment, e.g., stakeholder feedback. If, for example, a supplier’s on-time delivery drops or defect rates climb, the agent would flag it immediately and could suggest corrective actions (like re-negotiating terms or organizing a quality meeting) or trigger another agent (for identifying an alternative). It may also identify top performers, prompting recognition or a strategic partnership. The key is that agents work 24/7 to keep supplier metrics within targets and escalate any issues before they impact operations.

Illustrative example: A supplier performance agent analyzes ongoing shipments and notices a trend of late deliveries from one vendor. It cross-checks this against warehouse logs and escalates an alert to procurement and logistics. It might then automatically schedule a conference call with the supplier or even re-route some orders to a second source. By using live data to optimize supplier relationships, agents help maintain high performance across the supply base.

Strategic procurement planning

Agentic AI can act as a strategic partner, enabling long-term planning and scenario analysis. Planning agents integrate macroeconomic forecasts, commodity price trends, internal demand projections and risk models to propose strategic initiatives that align with these factors. They can simulate ‘what-if’ scenarios, e.g., the impact of a tariff change, and recommend category strategies or alternative sourcing plans. In effect, they help procurement teams move from reactive cost-cutting to proactive value creation.

Illustrative example: A strategic planning agent analyzes multiyear purchase history, forecasts and external market indices. It might predict a future aluminum price surge and suggest locking in a longer-term contract now. The agent could also develop a category roadmap by identifying which product lines should be centralized or localized based on projected growth or risk. Over time, it learns which strategies drove savings or resilience, refining its guidance. Such agents ‘predict needs before they arise’ and even propose new sourcing strategies, effectively becoming a procurement ‘thought partner.’ By continuously learning, they help align procurement plans with business goals and emerging market conditions.

Tail spend and low-value procurement

Agentic systems are particularly well-suited to managing tail spending (the fragmented, low-volume purchases that typically escape procurement oversight due to resource constraints). Historically, procurement teams have faced a trade-off between enforcing compliance in this area and allocating bandwidth to more strategic categories, which is why many vendors focus on bringing smart and AI-based capabilities to automate tail spend sourcing. 

Agentic AI represents the next step, as it can be used to broadly automate the end-to-end process beyond sourcing, thanks to the orchestration of multiple specialized agents. Such agents can autonomously validate purchase requests, identify preferred suppliers, generate POs and route them for approval according to policy, with minimal or no human intervention. Agents can also be used to close the P2P process for goods receipts and invoice processing.

These agents enforce rules and ensure policy coverage in areas where humans lack the time or incentive to intervene, dramatically increasing compliance and control across long-tail spend.

Illustrative example: A requisitioner submits a request for promotional merchandise. An agent detects this as a tail-spend category, classifies it appropriately and checks for existing contracts. Finding a preferred supplier on file, it auto-fills pricing, initiates a PO and routes it for approval under preset thresholds. If the spend is below the auto-approval limit, the agent issues the PO directly and logs the transaction for audit purposes. If no preferred supplier is available, the agent can suggest pre-vetted companies based on category and past transactions or trigger discovery and sourcing agents. Over time, the agent learns which purchases tend to bypass policy and proactively flags or redirects them, thereby closing compliance gaps without increasing the team’s workload.

Procure-to-pay exception handling

Even in well-automated environments, exceptions in the P2P process, such as invoice discrepancies, unmatched POs or missing receipts, remain a persistent source of friction. Agentic AI introduces a new level of responsiveness and learning to these cases. Exception-handling agents use business rules, transaction context and historical resolution patterns to resolve discrepancies autonomously. They can reconcile mismatched documents, escalate issues only when necessary and, in many cases, resolve the issue end-to-end without requiring human review. 

Illustrative example: An invoice arrives for 100 units, but the PO shows 95. A P2P exception-handling agent spots the mismatch, checks the goods receipt history and finds that 100 units were received and accepted. It verifies that the price and terms match contract conditions, logs a discrepancy note and clears the invoice for payment. In another case, when a receipt is missing, the agent triggers a reminder to the receiver or automatically creates a receipt if it has collected information that shows the goods were actually received. Over time, it learns to prioritize common exception patterns and resolve them faster, freeing AP and procurement staff from manual triage.

Getting started

Several organizations are already piloting agentic tools. Some use RFx copilots and contract summarization assistants; others have expanded into areas like tariff management and compliance monitoring. Supplier management is another area where organizations have put in place agents that can detect declining supplier performance and automatically retrieve the delivery history, query alternative vendors and generate a continuity plan without requiring a single prompt.

The examples above show the power agentic AI has in the orchestration across systems and processes. Agents do not replace the system of record but activate it. And where legacy tools follow linear rules, agents adapt in real time and operate across silos.

While many deployments are still scoped and tightly monitored, the core capabilities (autonomy, goal orientation and feedback learning) are operational. 

It should also be noted that not all systems currently labeled ‘agentic’ operate with full autonomy. Capabilities exist along a spectrum of capabilities (from fixed sequences with dynamic inputs to fully autonomous agents capable of planning, executing, and learning). 

Understanding where a use case falls on this spectrum is essential when evaluating maturity and ROI potential. A report by The Hackett Group^® supports this pragmatic and ramp-up approach. While nearly half of procurement teams piloted Gen AI in 2024, only 4% reached scaled deployment, which highlights the importance of aligning ambition with operational readiness and maturity.

For procurement teams, the task and challenges are to identify use cases specific to their organization that are structured, measurable and rich in accessible data to demonstrate the value of agentic systems. The Hackett Group’s report gives valuable insights into these aspects. 53% of procurement leaders expressed concern about unrealistic expectations, and many cited data quality and integration complexity as top barriers.

Also, from a technology selection standpoint, teams should assess whether a vendor’s system offers true agentic features like dynamic tool selection, real-time context management and multi-step reasoning or whether it simply automates decision trees. It is because these elements will drive the business case. It will be stronger when autonomy and adaptation are real, not just claimed.