AI has improved what procurement produces – It hasn't changed how procurement decides

This blog is part of a series from Closing the Gap: How Procurement Moves from Insight to Action, a whitepaper produced by Beroe in partnership with Kearney. The series explores what it will take to close the gap between intelligence and action to ensure procurement’s operating model meets its expanded mandate.  

In our previous post, we examined how procurement has become stuck in firefighting mode: reactive by default, overwhelmed by signals, and unable to convert visibility into timely decisions. In this post, we turn to what many assumed would solve that problem: AI. And why, despite genuine productivity gains, it hasn’t. 

Procurement teams are already using AI to accelerate a wide range of activities, from summarizing market intelligence to drafting sourcing documents and analyzing supplier responses. These are not trivial improvements. They reduce effort, compress timelines, and help teams move faster through tasks that previously might have consumed hours or days. 

AI represents a significant step forward for procurement, particularly in accelerating analysis and expanding access to insight. The challenge is not its capability, but how it is applied within the broader system: it has been applied either at the wrong layer of the problem, or to specifically address single issues in isolation. Procurement has tended to treat AI as a means to create a growing ecosystem of specialized tools, each addressing its own issue, when what it actually needed was a unified decision system.

Why AI adoption in procurement is delivering productivity gains but not better decisions

The current wave of AI adoption in procurement has improved the speed and quality of output. It has allowed category managers to produce analyses faster, analysts to generate detailed insights in an hour that previously might have taken months, and leaders to access synthesized information in seconds. At the same time, there has been significant investment across procurement’s broader technology stack, delivering improved visibility via enhanced data platforms, expanding insight with market intelligence tools, and digitizing execution with S2P platforms. On the surface, this looks like transformation but in practice, the impact on decision-making has been more limited. 

The issue is not an absence of tools, but how they are connected. Insights are generated using internal and external data, workflows can be executed, but the process of moving from that data to a decision to action remains fragmented. Faster output does not automatically translate into better procurement decisions. A market summary is not a decision, and even a recommendation is not a decision unless it is contextualized, assessed, explained, aligned, executed, and measured. Today, the steps by which that process is performed are disconnected. 

This is where many AI initiatives stall. They accelerate individual tasks, but do not redesign how procurement senses, prioritizes, chooses, acts, and learns. Put simply, they are improving parts of the decision-making process, but not the process itself. 

Today, many procurement professionals use generative AI tools regularly, and a significant proportion of leaders report measurable productivity gains in daily tasks. But this is just solving the first part of the problem: producing information and analysis. That final step of making an informed and timely decision requires context, judgment, coordination, and accountability. It is also where most of the time and complexity sits. 

This points to a deeper shift. The real frontier is not productivity alone, but the economics of attention. In a world of abundant signals, the scarcest resource in procurement is no longer information, but the time and focus required for teams to make decisions. The question is no longer how to produce more analysis, but which decisions are worth a procurement professional’s attention, and how that attention should best be allocated.

Why procurement’s decision-making process is still built on outdated foundations

To understand why AI hasn’t solved procurement’s decision problem, it helps to understand what sits beneath the surface. Many procurement processes were designed in the ERP era, built around structured workflows, sequential approvals, and periodic planning cycles. These assumptions work for transactional control, but they are poorly suited to continuous market adaptation. 

Strategic procurement decisions do not move cleanly through linear workflows. They are iterative, require judgment under uncertainty, and often involve trade-offs rather than right answers. 

Adding a conversational interface to an ERP-era process does not make the process continuous, contextual, or decision-centric. A chatbot that answers questions about a dashboard may improve usability and access to information, but it does not change how signals are prioritized, whether the organization acts in time, or how outcomes are captured.

Why generic AI tools cannot make defensible procurement decisions

Generic LLMs are impressive because they are broad. They can explain concepts, draft content, summarize documents, structure arguments, compare options, and create plausible analysis across almost any domain. For low-risk tasks, this breadth of application is valuable. 

However, strategic procurement decisions are highly contextual. Recommendations to change suppliers, for example, depend on a myriad of factors: category dynamics, the supplier’s capability and track record, contractual constraints, switching costs, changing regulatory requirements, or evolving stakeholder priorities. These are not variables generic models can inherently understand or reliably infer. 

A generic LLM may summarize the state of a market, but it will not, by default, know whether the potential new supplier has available capacity, whether new quality approvals would take six months, whether the business can tolerate a specification change, or whether the timing aligns with a sourcing window. These are not minor details; they are often the difference between an interesting insight and a defensible decision. 

To improve decision-making, AI requires an understanding of the specific context of the business, the ability to evaluate trade-offs, align stakeholders, and apply judgment under uncertainty. It also requires learning: the ability to understand what happened when similar decisions were made in the past, and to use that to inform what should happen next. This is where current approaches fall short. They generate answers, but they do not consistently capture context, embed experience, or learn from outcomes in a way that improves future decisions. 

Critically, these limitations do not diminish the value of AI; they clarify its role. As Gartner predicts, by 2030 AI will orchestrate procurement in 30% of organizations, triaging tasks to humans or AI agents based on their unique strengths. AI can accelerate how procurement produces insight, but it does not, on its own, resolve how procurement makes decisions. That requires a different layer, one that connects signals to action in a continuous, contextual, and scalable way.

Real-time procurement data does not translate into faster decisions

Procurement’s operating rhythms were designed for a world where periodic optimization was sufficient. Annual category plans, quarterly business reviews, scheduled sourcing waves, and point-in-time consulting assessments all assume that the market will remain stable enough for decisions to retain relevance over time. That assumption no longer holds. 

Many procurement leaders could reasonably argue they already have continuous visibility through market alerts, feeds, dashboards, trackers, business intelligence reporting, and AI assistants that can summarize inputs on demand. But continuous observation is not continuous decision-making. 

The value lies not in the signal itself, but in determining whether that change matters, what it means, what options exist, what trade-offs are acceptable, what action should be taken, and how value will be measured. Most procurement systems today are better at signals than they are at decisions.

How decision latency is causing procurement teams to miss value capture opportunities

Category strategy illustrates the structural challenge clearly. Strategies are typically developed as point-in-time assessments, comprehensive at the moment of creation but static thereafter, while the conditions they describe continue to change. As markets move and supplier conditions evolve, the assumptions underpinning those strategies degrade. Yet in many instances the strategy itself remains unchanged until the next formal review cycle. 

One of the biggest differences between episodic and continuous procurement is the window to act. In slower markets, procurement could identify an opportunity, analyze it, align stakeholders, and execute over weeks or months. In faster markets where commodity prices are volatile or tariff exposure unavoidable, the value window can close before the traditional process has even started. 

Speed matters because market signals decay quickly. A signal is most valuable when it is early enough to change the outcome. Once the market has fully adjusted, procurement is no longer acting on insight; it is reacting to consequences. 

The result is structural decision latency. Value is not lost because procurement lacks information, but because it cannot consistently convert that information into timely action. The procurement teams that pull ahead will be those built to make better decisions, continuously, at scale, and before the market forces their hand.

This is the fourth in a series of blogs drawn from Closing the Gap: How Procurement Moves from Insight to Action, a whitepaper authored by Beroe and Kearney. To read the full whitepaper, click here.

To find out how Beroe MAX^TM powered by Kearney, is closing the gap between visibility and action, read the full press release here or watch our on-demand webinar here.