Procurement teams are experimenting widely with generative AI, but few have pushed the technology into full-scale operations. New findings from EFESO’s CPO Pulse Report show a widening gap between early enthusiasm and enterprise-ready deployment.
A Growing Divide Between Experimentation and Scaled Impact
EFESO Management Consultants’ latest CPO Pulse Report, based on interviews with 50 chief procurement officers, shows that generative AI has become a routine component of procurement strategy discussions. Nearly all surveyed leaders report experimenting with the technology, and many describe it as an increasingly reliable tool for productivity, cost efficiency, and analytical support.
Yet adoption at scale remains elusive. According to the report, just 5% of procurement organizations have industrialized GenAI across their operations, a striking contrast to the 75% still confined to experimentation. Forty percent remain in early exploration, while 35% are running pilots that have yet to mature into enterprise-wide rollouts.
The gap is not driven by lack of interest. It reflects a tightening focus on feasibility, governance requirements, and economic justification. EFESO notes that although GenAI is no longer treated as an emerging concept, the sector is moving beyond broad testing toward selective deployment, where data readiness and operational control can be maintained.
Recent market research echoes this shift. Studies from multiple advisory firms show rising investment in foundational data work, particularly taxonomy alignment, master data cleanup, and privacy controls, as organizations discover that advanced GenAI applications stall without stable inputs. That groundwork, while critical, is slowing the pace of visible transformation.
Where GenAI Works Today and Where It Doesn’t
The Pulse Report reinforces that GenAI’s clearest value is concentrated in high-density, high-structure use cases. Contract analysis and summarization lead with a 69% success rate, followed by supplier and market intelligence (61%) and automation within sourcing workflows (55%). These areas share common characteristics: abundant accessible data, lower integration complexity, and quick validation cycles.
More advanced applications are proving harder to industrialize. Only 35% of procurement leaders reported value from AI-supported negotiation, reflecting the model-training requirements, governance controls, and behavioral variability embedded in negotiation processes. EFESO’s Principal, Gaël Sandrin, describes this shift as a move toward disciplined decision-making: leaders are less concerned with whether GenAI works and more with where it can work reliably and cost-effectively.
That discipline is also shaping investment decisions outside the immediate study scope. According to trade reports, companies piloting GenAI-driven risk assessments, particularly in supply continuity and ESG analysis, are discovering that models require deeper integration with supplier data than existing systems can provide. The friction reinforces EFESO’s finding: the constraint is structural readiness, not technological ambition.
Confidence Gaps and Barriers to Scaling
Despite widespread experimentation, satisfaction levels remain mixed. Only 34% of surveyed leaders say GenAI has delivered value commensurate with early expectations, while nearly half report partial satisfaction and one in five express disappointment.
Major barriers persist. Concerns about data reliability (68%) and confidentiality and compliance (67%) continue to limit broader deployment. Skills gaps (57%) and inconsistent data quality (55%) further slow progress. These challenges mirror what has emerged in other digitally intensive functions: scaling AI requires more than standalone tools, it demands robust governance, secure integration, and domain-specific talent.
EFESO argues that this does not signal a slowdown in adoption but rather a maturation of expectations. Organizations are transitioning from testing GenAI everywhere to deploying it only where operational readiness, economic logic, and governance allow it to create measurable lift.
A Shift Toward Data Conditions That Strengthen AI Performance
Procurement research published over the past year shows a steady rise in investment toward supplier data quality, controlled vocabularies, and audit-ready lineage tracking, areas that once sat outside mainstream transformation programs. These efforts are beginning to show measurable benefits in organizations that have reported faster model tuning cycles and more stable AI-generated outputs. As more teams build these conditions into their operating routines, GenAI’s practical value tends to increase in tandem. This pattern suggests a growing recognition that durable progress often comes from improving the structures that AI relies on, rather than expanding the number of use cases in play.