AI-driven procurement faces architectural hurdles as fragmentation hampers trust and scalability

Procurement’s enthusiasm for AI is no longer the main issue. The harder question, as Ardent Partners’ research shows in a report sponsored by Ivalua, is whether the function has built an operating environment capable of supporting it. In its latest analysis of the move towards AI-first procurement, CPO Rising argues that the limiting factor is increasingly architectural: data, processes and supplier activity remain scattered across systems that were never designed to work as one.<...

That matters because AI performs best when it can draw on connected context. According to the report, only 11% of organisations say they operate with a unified data model, while 42% describe themselves as integrated. But integration is not the same as true connectivity. As the analysis notes, manual reconciliation may help teams keep operations moving, but it does not create a seamless foundation for intelligent execution. In practice, procurement is still living with managed fragmentation.

This is where AI changes the conversation. For years, system architecture was often treated as an IT concern, with buyers focusing more on features, usability and basic integration. Now, as AI is being asked to evaluate suppliers, interpret contracts and support transactional decisions, the quality of the underlying architecture has become a procurement issue in its own right. If the data model is inconsistent, inaccessible or poorly governed, then AI may be able to produce an answer, but not necessarily a trustworthy one.

The biggest concern raised in the Ardent Partners research is hallucination, cited by 51% of respondents. That fear is understandable: AI can produce convincing but incorrect outputs even when the source data is sound. When the data itself is incomplete or inconsistent, the risk increases further. In procurement, where such outputs may shape supplier assessments, commercial terms or risk judgements, the consequences of confident error can be financial as well as reputational.

Data sovereignty is another pressure point, at 20%, reflecting unease about where procurement data goes once it is used to train or power AI systems. The article also suggests that rigidity becomes a more visible obstacle earlier in the deployment journey, implying that many organisations can pilot AI tools, but struggle to scale them because their architecture cannot absorb the change.

That helps explain why general-purpose AI tools and enterprise copilots are currently the most common access point for procurement teams, used by 63% of organisations. By contrast, AI embedded directly into core procurement platforms is present in only 27%. The result is a market that has adopted AI in broad terms, but has not yet woven it deeply into daily workflows. Most teams are still working alongside AI rather than through it.

The wider technology commentary reinforces that point. TechTarget has argued that fragmented AI infrastructure creates governance gaps, delays and operational inefficiencies, and that enterprises need a converged architecture with shared governance and access controls if they want AI to scale reliably. Promethium, meanwhile, says fragmented data across warehouses, CRM systems and on-premise databases remains one of the main reasons AI projects stall. In procurement specifically, Speclens has highlighted the need for clean supplier identifiers, category codes and current spend data before AI can produce useful analysis, while Futura Solutions warns that incomplete datasets can still steer buying decisions in the wrong direction. Levelpath similarly says that many enterprises remain stuck because procurement data is trapped in silos rather than organised around a single source of truth.

The common theme is that AI does not remove operational complexity; it exposes it. TechRadar has made a similar case, arguing that AI often amplifies weak data and unclear governance rather than masking them. For procurement leaders, that means the route to AI-first maturity is not simply to buy more tools, but to reduce the burden created by disconnected systems and poor data discipline.

Ardent Partners’ broader message is that procurement’s AI journey begins with the infrastructure beneath the technology. Until organisations can connect their data, align governance and make their operating environment more coherent, AI will remain useful at the edges but limited at the core.

Source: Noah Wire Services