Procurement consists of a long chain of tasks: from signaling that something is needed, to negotiating, ordering, receiving and invoicing. Some of these tasks are largely administrative and follow a fixed pattern. Others require judgment, relationship sense, or a trade-off that cannot be captured in rules. Our taxonomy counts 62 tasks within this domain, divided into three categories. This article shows the pattern: which type of work falls where, and why.
Tasks that AI can in practice perform independently generally have two characteristics: the input is structured and there is a clear, verifiable correct answer. Consider checking order confirmation. The price, quantity and delivery time on the confirmation are compared with what was ordered, and a discrepancy can be objectively established. No room for interpretation is needed.
The same applies to placing an order with a supplier once a purchase request has been approved. Converting an approved request into an official order is a matter of correctly transferring data into the right system and format, not a matter of judgment.
Processing purchase invoices also largely falls under this category: matching the invoice, order and receipt is a three-way comparison with fixed criteria. As long as the documents are digital and legible, this is a task where a system matches faster and more consistently than a human doing it manually. What these tasks have in common is that the source of truth is already established in other systems: the order, the request, the contract. AI compares and processes, it does not invent anything.
The largest group of tasks in the procurement department combines an AI proposal with a human assessment that approves or rejects, with reasoning. That is a different division of labor than full takeover: the system makes the first move, the human bears responsibility for the decision.
Requesting and comparing quotes is a good example here. A system can effortlessly lay quotes side by side on price, terms and delivery time, and that overview is valuable. But which supplier best fits the situation also depends on things that are not always in the quote: previous experiences, strategic importance, risk appetite. That trade-off requires a human who can approve or reject the outcome.
Investigating price discrepancies also falls into this category. A system can signal that an invoice price deviates from the agreed price and even suggest a likely cause. But actually resolving it with the supplier, and the question of whether the discrepancy is acceptable, remains a judgment that someone with knowledge of the relationship and the contract must make.
Evaluating supplier performance also belongs here. Data on delivery time, quality and complaints can be well collected and summarized, but the judgment about what that means for the continuation of the relationship is a decision with consequences that a human must be able to substantiate. This category is in practice the largest, because procurement is a field where figures provide an indication, but not the whole story.
A third group of tasks does not lend itself to takeover, not even with oversight, because the core of the work revolves around persuading, building trust, or making a decision with long-term consequences that does not follow from data alone.
Negotiating purchase prices is the clearest example. Negotiating is a dynamic, social process in which tone, timing and mutual trust are at least as important as the figures. An AI system can provide input, but conducting the conversation itself is something else.
Selecting a supplier for an important or long-running contract also remains human work, especially when the choice goes beyond price and delivery time. Strategic dependency, reputational risk and the quality of long-term collaboration are factors that are difficult to quantify.
In addition, onboarding a supplier remains a task where human control is needed over compliance and payment data, precisely because errors here create financial risk. This connects to a broader theme: when the use of AI touches on decisions about who performs work or how an organization is structured, separate legal requirements apply, as can also be read in the AI regulation and your employees.
What separates the three categories is not the complexity of the task itself, but the nature of the decision it contains. Comparing, matching and taking over already established data is suitable for full automation. Assessing based on multiple, partly soft criteria requires oversight. Persuading, negotiating and making strategic choices remains human work. This pattern is not unique to procurement: in what work can AI take over in logistics and in HR tasks, as described in what work can AI take over in the HR department, the same three-way division recurs: structured work goes into category 1, assessment work into category 2, relationship and decision work into category 3.
How many hours this frees up in concrete terms depends heavily on the size of the procurement department, the systems that are already connected, and the number of suppliers and contracts. No general percentage can be given for this without knowing the specific situation.
The classification in this article shows the pattern, not the precise situation in your own procurement department. For an initial indication, there is the free quickscan from ftetoai: twelve questions, no account required, with an indication of what portion of the hours in your profile can be taken over by AI today. The full work scan, which goes deeper into specific tasks and processes, is still under construction; we cannot yet offer it here, but we will announce it as soon as it becomes available.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.