Five questions about evidence, authority and accountability help determine how much work procurement can delegate to AI agents.
An agent can assemble a supplier assessment quickly while leaving the buyer uncertain whether it concerns the right legal entity or the factory supplying a critical part. That uncertainty becomes consequential when the output shapes an award or a production commitment.
At DPW Amsterdam 2026, the implementation problem had a clear place on the agenda. In its Siemens AI session preview, DPW framed Siemens CPO Klaus Staubitzer’s “Recoding Siemens with AI: The Good, the Bad and the Ugly” around legacy systems, workforce readiness and the economics of scaling agents. These are operating requirements procurement leaders need to examine when comparing AI proposals.
Five questions can help CPOs decide where agents can contribute, what evidence they need and which judgments require other functions. The answers should be specific to the decision being delegated: a routine bid comparison, a supplier assessment and a continuity response require different depths of information and authority.
What Does the Agent Know About the Supplier
A supplier may appear under several names across sourcing, purchasing and risk systems. Ownership can sit with another legal entity, while a relevant facility or material is absent from the purchasing record. An assessment built on the wrong identity can overlook exposure or create an unnecessary escalation.
The public summary of HFS Research and DPW’s procurement AI readiness survey of 150 executives reports that only 4% had a unified data foundation they trusted for AI. That finding makes supplier identity and evidence quality central evaluation questions.
Procurement needs a connected understanding of the supplier and the relationships relevant to the decision. The evidence should show where a finding came from, how recent it is and where important gaps remain. Those details let a category team distinguish an established fact from a question that requires further review.
The information required will vary. Comparing bids may depend on normalizing units and commercial terms. Assessing a supplier change may require ownership and facility information. Leaders should define the evidence needed for the work they intend to delegate and establish who resolves conflicting records.
Which Decisions Can the Agent Make
Collecting documents, preparing a comparison and changing an approved supplier carry different consequences. The authority assigned to an agent should reflect those differences.
In “Speed Wins: How Procurement Leaders Are Compressing the Sourcing Cycle,” Time Manufacturing’s Emil Radu Cojocaru described the challenge of extending a procurement agent across four ERP systems and five data languages. The team recognized that a broad rollout could generate enough exceptions to require more people to manage them. He described a phased approach beginning with information gathering and FAQs before adding more rules and scope.
That example makes exception handling part of the economics. A team needs to understand the effort required to check, correct and escalate the agent’s work before assuming it has released capacity.
A team might permit routine information gathering under defined rules while requiring review before supplier outreach or a change to a commercial record. A consequential eligibility decision needs an accountable person and a recorded rationale. The organization should also determine what happens when an agent encounters missing evidence or an exception outside its authority.
These choices affect how much work a team can delegate. Sending every routine step for approval can preserve the original bottleneck. Giving an agent broad discretion without clear limits can create commitments the organization has not agreed to make. Define review points around the actual stakes of the decision.
Does an Alternative Remove the Dependency
When a supplier cannot deliver, another approved name can appear to solve the problem. Both companies may depend on the same upstream producer or specialized facility. The proposed switch can leave the original exposure intact.
Procurement needs to assess the relationships behind the alternative, as well as its ability to meet specifications and deliver the required volume on time. An agent can help assemble the evidence, but the assessment must reflect the business’s requirements and the limits of what is known.
This is where procurement and engineering need a shared basis for evaluation. A commercially available substitute may require testing or a design change. A qualified part may have no available capacity. The team needs to know which condition applies before it presents an option as usable.
How Does New Evidence Reach a Decision Already in Motion
A supplier decision remains exposed to change after an award. An ownership change, disruption or new restriction can affect the case for continuing a purchase that was reasonable when approved.
The relevant question is how that evidence reaches the active decision. Procurement may need to identify affected orders and products, determine which commitments require attention and assign someone to assess the response. An alert without that connection can leave the original purchase proceeding on an outdated assumption.
Teams should decide which changes reopen review, who receives the evidence and what the agent may prepare while a decision is pending. The response also needs to show where uncertainty remains so the business can judge whether it has enough information to act.
Who Owns the Response Across Functions
A finding may begin in risk, alter a sourcing recommendation and require engineering or finance approval. Several systems can hold part of the record. The business needs an owner for the issue and a clear account of the resulting decision.
In “Why Procurement Is Moving Up the CFO Agenda,” Swisscom Group CPO and Finance Transformation Lead Michaela Mayrhofer described how process redesign becomes difficult when it changes individual mandates and responsibilities. Delegating work to agents brings the same issue into focus.
Procurement may own the sourcing recommendation while engineering approves a substitution, compliance interprets a restriction and the business owner decides whether to accept the commercial consequence. The agent’s role should fit that division of responsibility. Consequential exceptions need an identified reviewer, a recorded decision and a way to verify that the agreed action occurred.
The response needs to reach the systems where the company records its approved suppliers, contracts and orders. Reviewers should be able to follow the evidence and understand why the organization chose a particular course of action.
Build the Foundation Around a Consequential Decision
Test the approach against a situation that crosses these boundaries: a disruption at a critical upstream facility while a product launch is approaching. Work through the evidence, alternatives and approvals the organization would need. Decide which tasks an agent can perform and where human judgment changes the outcome.
That exercise gives an AI initiative a concrete purpose. It also makes the requirements for an AI-native foundation more precise: connected evidence, defined authority and a way to carry the work through to resolution.
1EXIGER.AI connects risk intelligence with agents that gather evidence and coordinate work under customer-defined controls. Customers determine the permitted sources, actions and approvals. That gives procurement leaders a concrete basis for discussing what they want to delegate and where specialist or business review must remain.
To discuss a demonstration focused on your procurement decisions, contact Exiger.
Table of Contents
Get in Touch
Learn how you can build a more resilient supply chain.
