Procurement Has an AI Problem That Isn't About the Technology
AI is already doing real work inside procurement. It can compress tasks that once took days into minutes — analyzing spend, reviewing contracts, flagging supplier risk, and steering employees toward compliant purchases. That productivity gain is genuine.
So is the new exposure it creates. Every automated action runs on whatever data sits behind it. When supplier records are incomplete, contract terms are scattered, or spend data lives in three systems, the AI operates on assumptions. It still produces an output. It just may not be a reliable one.
SAP's Gordon Donovan, vice president of Research for Procurement and External Workforce, published a feature on August 10, 2026 laying out this tension directly. The context is a study SAP sponsored through Economist Enterprise, covering 2,648 C-suite leaders and published as Procurement at a Crossroads: From Optimism to Realism.
What the Study Found
The headline number is 56% of executives named AI strategy as the main catalyst for procurement's digital agenda. That is a strong signal about where investment and attention are going.
The more consequential finding sits beside it. The same study recorded lower confidence in the function's ability to translate technology investments into consistently better outcomes. Executives are spending more and feeling less certain it is working. Many reported that AI has yet to meaningfully improve procurement decision-making quality despite growing investment.
Rising investment paired with flat or uncertain outcomes points away from a technology gap and toward a governance and data readiness gap.
The Operating Risks ERP Owners Should Take Seriously
Two structural vulnerabilities run through the SAP argument.
The first is data fragmentation. Fragmented supplier records, contract terms, spend information, and risk signals mean an AI agent lacks the context it needs to act reliably. Without unified data governance — covering common definitions, ownership rules, access controls, and traceable sources — the system fills gaps with assumptions. Those assumptions reach decisions before a human reviews them.
The practical ERP question is whether the data model your procurement processes depend on is actually coherent. Supplier master records that differ between your ERP and your procurement tool, contract terms stored outside any governed system, or spend categories defined differently across business units all create this problem. They do not need to be perfect.
But organizations should be explicit about uncertainty — surfacing data gaps to users rather than letting the system proceed as if the information is complete.
The second vulnerability is accountability design. SAP's position is that governance should be built into the operating model before deployment, not added as a final approval layer. That means defining which actions AI may complete, which recommendations require human review, and which decisions must remain under human control before a use case goes live. It also means setting escalation rules for exceptions involving sensitive data, high-value commitments, supplier concentration, or regulatory obligations.
The study adds one more boundary worth noting. Fewer than one in 10 respondents would give AI the lead across most procurement choices within three years. By contrast, 46% expect AI to assist with tactical work while people retain authority over strategic matters. That split reflects something experienced procurement professionals understand well: data can identify a favorable supplier on price and performance metrics.
It cannot tell you how that supplier will behave during a disruption or whether they will treat your business as a priority when capacity is constrained. Those assessments depend on relationship history and commercial judgment no system fully captures.
Measuring the Right Things
The SAP argument is also a measurement argument. Organizations need metrics for both value and risk — defined before deployment, not after. Cycle-time reduction, contract compliance, spend under management, user adoption, supplier performance, and risk response can show whether a use case is delivering. But those measures need to be paired with exception rates, human overrides, data-quality failures, and control breaches.
The reason to track both simultaneously is that efficiency metrics can look good while decision quality stays flat. If your AI is completing more purchase recommendations faster but your exception rate is climbing, the speed is obscuring a quality problem. The visible metric hides the real one.
What Finance and ERP Leaders Should Do Now
If your organization is expanding AI use in procurement, three questions are worth working through before the next deployment decision.
First, is your data foundation stable enough? Map where supplier records, contract terms, and spend data actually live. If they are fragmented across your ERP, a procurement platform, and unstructured storage, that is a prerequisite problem — not something AI will resolve on its own.
Second, have you defined decision rights before deployment? The SAP framework calls for explicit thresholds: what AI completes autonomously, what it recommends for human review, and what it escalates. Building those rules into the workflow design is materially different from reviewing AI outputs after the fact.
Third, are your success metrics paired with risk indicators? A dashboard showing faster cycle times without corresponding visibility into override rates and data-quality failures will not tell you whether the AI is actually improving outcomes.
The organizations that get this right will not simply be the ones that automate the fastest. They will be the ones where AI amplifies the judgment and relationship experience procurement professionals already have. And they will be the ones where accountability for consequential decisions stays clearly assigned to a person.