When AI Arrives in Procurement, the Greatest Resistance Is Not in the Software
There is a pattern that repeats itself in almost every organization going through a deep technological transformation: the hardest part was not choosing the platform. It was discovering, weeks after the launch, that the underlying problem was not technological at all.
In the case of artificial intelligence applied to procurement and sourcing functions — what the industry calls procurement — that pattern is becoming so common that it already has a name of its own. McKinsey describes it with surgical precision: the organizations that manage to scale AI in procurement are not those that chose the best software, but those that redesigned their workflows from end to end before asking a model to automate them. Those that did not discovered that AI does not fix operational fragmentation. It amplifies it.
What is happening in the world of corporate acquisitions is not a tool upgrade. It is a reordering of who makes decisions, with what information, at what speed, and with what level of human intervention. That cannot be solved by purchasing a license. It demands that the organization change what it understands by value, by control, and by accountability.
---
The Illusion of the Successful Pilot
Prajkta Waditwar, senior director of technology sourcing at Box and member of the Forbes Technology Council, described a scenario that any operations leader will recognize: a global organization implemented AI to improve supplier visibility and automate risk assessment across different regions. The models performed well in the test environment. But when the time came to scale, the system exposed inconsistent supplier data, fragmented approval workflows, and disconnected enterprise management systems that had survived for years thanks to human tolerance for ambiguity.
This type of post-pilot failure has an internal logic that is worth naming explicitly: pilots work because variables are controlled. An orderly spending category is selected, a motivated team, a collaborative supplier. The AI shines. The investment is validated. The decision to scale is made. And then the system comes into contact with the full operational reality, with all its accumulated history of improvised processes, mislabeled data, and decisions that used to be made over the phone.
The problem is not that the organization did something wrong. The problem is that for years, efficiency was sustained by the human judgment of people who knew the shortcuts, the exceptions, and the suppliers that were difficult to classify. That tacit knowledge was never documented, never structured, never converted into data. AI cannot operate on what does not exist as data.
Deloitte documents this in its 2025 global survey of chief procurement officers: organizations with greater digital maturity are obtaining significantly higher returns on their investments in generative artificial intelligence. The gap is not in who has access to the technology. It is in who built the foundations beneath it.
---
The Structural Problem That No One Has Named Yet
There is something that organizations tend to underestimate when embarking on this transformation, and it deserves to be named precisely: operational fragmentation in procurement is not an accident. It is the result of years of rational decisions made locally.
Every region that negotiated its own contracts had reasons to do so. Every business unit that built its own approval process was solving a real problem with the resources it had. Every team that maintained a spreadsheet parallel to the corporate system did so because the corporate system did not respond with the speed they needed. Fragmentation is, in many cases, the digital footprint of an organization that grew faster than its governance capacity.
The moment that organization introduces artificial intelligence into its procurement processes is precisely the moment when that history is exposed. And what is exposed is not only technical inefficiency. What is exposed is a governance model that depended on the individual judgment of specific people in order to function.
This connects to something McKinsey points to when describing the evolution toward what they call "AI agents": systems that can ingest context, plan complex tasks, and act with a certain degree of autonomy across multiple systems simultaneously. When that agent attempts to operate in an environment where supplier data exists in three different versions depending on which system you consult, where approval policies vary by region without any documented logic, and where the master contract sits on a local server that only one person knew about — someone who no longer works at the company — the agent does not fail due to technological limitations. It fails because the environment does not have the minimum architecture required to sustain automated decisions.
The question this raises for senior leadership is not whether to implement AI in procurement. It is how honest they are being about the true state of their data infrastructure and governance before asking the system to decide.
---
When the Procurement Function Stops Being Operational
Zycus, in its 2026 artificial intelligence guide for procurement, describes the transition in terms that initially sound like marketing copy but, when read carefully, reveal something more structural: AI is not arriving in procurement to make what already exists more efficient. It is arriving to absorb the bulk of transactional work and free up human capacity for something different.
That "something different" is what Waditwar describes clearly from her direct experience: procurement teams are being brought in increasingly earlier in strategic conversations, not to negotiate prices, but to assess the long-term operational implications of a supplier decision. How much dependency does a deep integration with a software vendor generate? How complex would it be to exit that contract in three years? Does the technology architecture being purchased increase or reduce future flexibility?
Those are not questions that historically belonged to a procurement function. They are questions of strategic risk management. And the fact that they are now part of the area's agenda reveals something important about what is changing: the automation of transactional work does not only free up time — it redistributes authority.
HFS Research puts it in more direct terms: AI platforms are enabling procurement leadership to move from operational execution toward strategic enablement. That means the competency profile needed in the function is changing, that the indicators used to measure its success will have to change, and that the relationship between procurement, finance, legal, and operations will have to be redesigned, because the boundaries between those functions become more porous when there is a connected intelligence system that cuts across all of them.
McKinsey estimates that a procurement function that completes this transformation can be between 25% and 40% more efficient than current models. But that figure should not be read as a projected headcount reduction. It should be read as a reallocation of capacity: fewer people processing transactions, more people making decisions that systems still cannot make on their own.
---
The Leadership This Change Needs and the One It Usually Finds
This is where the transformation becomes most interesting to analyze from an organizational perspective, because the leadership profile that historically dominated procurement functions was built around very specific competencies: hard negotiation, deep supplier knowledge, the ability to move contracts under pressure, the institutional memory of which supplier failed in which context ten years ago.
That profile has value. But it is not the same profile needed by a procurement function whose greatest contribution to the business lies in the quality of its risk analyses, the speed with which it can integrate market signals into sourcing decisions, and the ability to work with systems that generate recommendations that must be questioned when context demands it.
The transition is not comfortable, and it would be naive to describe it as if it were simply a growth opportunity for everyone. There are people with twenty years of experience in procurement whose core value lay in doing well the things that a system can now do faster and with more consistency. That generates real resistance, and that resistance is not irrational. It is a comprehensible response from someone who sees that the rules of the game have changed without anyone consulting them.
Deloitte points to something that deserves attention: organizations that invest in preparing their teams alongside technological modernization consistently outperform those that focus exclusively on technology deployment. That is not a surprising finding. But the way it translates into organizational practice does matter. It is not about offering AI courses to teams who simultaneously watch their primary tasks being automated. It is about redesigning roles in such a way that people understand what kind of human judgment remains irreplaceable and at what point in the process that judgment is most valuable.
The risk facing many organizations is not that their procurement teams will reject AI. It is that they will adopt it superficially, using it to accelerate what they were already doing without changing the underlying logic, and that in the process they will lose the opportunity to build a function that truly operates as a layer of strategic intelligence within the business.
---
Structural Maturity Cannot Be Improvised After Deployment
What is happening in procurement is, at its core, a very specific version of something organizations face in almost all of their deep technological transformations: the gap between the architecture they have and the architecture they need in order to sustain what they want to build.
AI in procurement is no exception. It is the use case where that gap becomes most visible most quickly, because the consequences of a poorly automated sourcing decision are concrete and costly. A supplier selected by an algorithm that operated on outdated data. A contract automatically renewed because the system did not have access to the risk signal that already existed in another system. An approval that processed itself because no one had clearly defined what level of spending required human oversight.
Those are not AI errors. They are organizational design errors that AI executes with perfect precision.
The argument that deserves the most attention in this entire debate is not whether artificial intelligence is going to transform corporate procurement. That outcome appears sufficiently supported by the available evidence. The argument that deserves the most attention is how many organizations will arrive at that transformation with the data architecture, governance processes, and role redesign needed for the system to function as promised — and how many will discover that they installed sophisticated technology on a foundation that was not yet ready to support it.
The answer to that does not depend on the software vendor they choose. It depends on how much institutional honesty they are willing to apply before deployment, not after.
