Most enterprises are piloting or have already deployed agentic AI in tech procurement. In fact, the primary question is no longer whether to use agentic AI but rather how to adapt the procurement organization’s design to keep up with AI advances and make the best use of the technology.
As part of BCG’s 2026 tech procurement study, we surveyed more than 200 CIOs, procurement leaders, and specialized IT and technology procurement buyers across industries in North America, Europe, and the Asia-Pacific region. We also conducted in-depth interviews with procurement and AI and tech leaders across industries, platform and product vendors, and global system integrators. Three themes emerged across the research:
- Internal operational benefits tend to appear earlier than external supplier-facing commercial benefits.
- Production deployments typically deliver stronger outcomes than pilots.
- Organizations that combine technology deployment with capability building, process redesign, and governance report stronger outcomes than those that focus on technology alone.
Differentiating Between Internal and External Value
The study distinguishes between two broad categories of outcomes from agentic AI in procurement.
Internal value refers to improvements within the organization, such as productivity gains, reduced cycle times, better process discipline, reduced manual effort, and more compliance.
External value refers to improvements in market-facing outcomes, such as optimized software licensing terms, reduced SaaS sprawl, improved cloud contract conditions, lower vendor lock-in risk, improved negotiation outcomes, access to new suppliers, improved supplier quality, reduced risk exposure, and overall commercial advantage.
This distinction is important. Respondents report high levels of internal value more frequently than high levels of external value for the six types of use cases we studied: vendor discovery, request for proposal generation, contract analysis, spending analysis, risk monitoring, and supplier performance management. (See Exhibit 1.)
Moreover, internal benefits often appear sooner because they stem from improvements in efficiency and productivity. Procurement teams can process information faster, complete more work, and spend less time on repetitive tasks. External benefits, by comparison, frequently depend on broader organizational change. Improving negotiations, supplier management, or sourcing outcomes often requires process redesign, governance changes, and new ways of working alongside the technology itself.
When evaluating success, it’s important for procurement leaders to anticipate that favorable outcomes are likely to appear in operational performance before they show up in supplier-facing commercial results.
Favorable outcomes are likely to appear in operational performance before they show up in supplier-facing commercial results.
Top Benefits from Agentic AI
The distinction between internal and external value becomes clearer when we look at the specific benefits that organizations report from agentic AI. (See Exhibit 2.)
Respondents most often cite improvements related to speed, throughput, and reductions in manual effort. These outcomes are operational in nature and help explain why internal value is reported more frequently than external value across use cases.
By contrast, benefits such as broader supplier coverage, fewer negotiation rounds, and stronger negotiation outcomes are reported less frequently. These results depend not only on the technology itself, but also on how procurement processes, supplier engagement models, and decision-making practices evolve around it.
Interviewees frequently described this progression of benefits as a journey. Operational improvements are often visible soon after deployment because they stem directly from automating or augmenting procurement activities. Supplier-facing benefits tend to emerge more gradually as organizations redesign workflows, integrate systems, and build confidence in new ways of working. Ultimately, external value has the potential to far exceed internal value.
One automotive manufacturer illustrates the point. Its early tender-assistant tools quickly produced local productivity gains. Then, by redesigning the procurement life cycle around AI rather than automating individual tasks, it converted those productivity gains into commercial results—such as negotiation uplift and EBIT contribution—worth more than ten times the internal value.
Pilots Underperform Deployments
Another interesting finding from our survey and interviews with industry leaders is that deployments generate much stronger internal value than pilots do. That doesn’t mean pilots are ineffective, however. They are designed to test a limited scope—often a single process, team, or data source. Their primary purpose is to assess feasibility and identify implementation challenges. Nevertheless, it’s instructive to see just how much returns from deployment outperform pilots. (See Exhibit 3.)
The reason for this difference is that many agentic AI benefits emerge only when organizations connect multiple workflows and redesign the surrounding operating model. As a result, pilots can validate individual use cases but often do not reflect the full potential of scaled deployment.
Pilots can validate individual use cases but often do not reflect the full potential of scaled deployment.
BCG worked with an automotive manufacturer executive who stated, “The early pilots delivered real productivity gains. But we couldn’t monetize that freed-up time inside our legacy processes. Only once we redesigned the entire procurement life cycle around AI did those gains start translating into commercial impact.”
Three Modes of AI Deployment
When organizations do deploy agentic AI in procurement, they run it in one of three modes:
- Assistive (Recommendations Only). The agent provides analysis and recommendations, while people make all of the decisions and take all of the actions.
- Low-Level Autonomy (Human in the Loop). The agent carries out the work, but people review and approve significant decisions before they take effect.
- High-Level Autonomy. The agent decides and acts on its own, within boundaries set by the organization.
A clear majority of companies keep a human in control, and this intervention keeps autonomy from advancing beyond an early stage. Most organizations with less than three years of AI experience still rely on human oversight. (See Exhibit 4.)
Organizations that report more advanced deployments frequently note both stronger AI capabilities and larger AI investments. (See Exhibit 5.) Leaders from those companies also emphasize the need for close collaboration between tech procurement, IT operations, and business stakeholders.
Our research suggests that investments and capabilities are complementary rather than interchangeable. Investment alone is insufficient if teams lack the skills, experience, and governance structures needed to deploy and manage agentic systems effectively. Likewise, the most capable teams will struggle to scale without sustained investment in technology, integration, and change management.
“Suppose you have a procurement team with no agentic capability today. You build a platform that identifies their process—but if they don’t use it effectively, how will it work?”
– Interview, global system integrator
Obstacles to AI Deployment
Even when technology performs well, four common obstacles often prevent faster cycles and lower costs. (See Exhibit 6.)
Data Foundations. Heterogeneous inputs and inconsistent data quality are the most cited technical blocker. About a third of respondents name the cost of data preparation and system modernization as a barrier, and system integrators describe data readiness as the single largest source of delay in deployment.
Integration with Existing Systems. Nearly half of respondents cite difficulty integrating AI with legacy systems such as enterprise resource planning (ERP) and procure to pay, and a further 47% cite integration complexity as a common technical barrier. Agents can create value only when connected to the live procurement flow, and building that connection is slow and costly. One global system integrator described a 12-month implementation, with each component tested in isolation and then stitched into a single orchestrated system
Legacy Processes. This is the underlying reason productivity in pilots doesn’t always translate into commercial outcomes. Four in ten respondents cite the difficulty of running the transformation alongside business as usual, and
one-third point to an ineffective operating model. If the process around the agent is unchanged, freed-up capacity has nowhere to go.
Governance. This is the most often cited area of constraint. The biggest organizational barrier is trust in autonomous decision making (71%), and close behind are security and IP risks (66%). More than half of respondents cite regulatory uncertainty (57%), accountability for agent actions (53%), and auditability of decisions (48%), while 42% point to the absence of structured KPIs. Trust isn’t built by deploying more agents; it’s built by governance and measurement systems that credit agents for what they deliver.
“Trust in full automation is limited. Humans still spend time validating outputs and tracing data issues, leaving efficiency on the table.”
– Interview, European retailer
Five Actions for Tech Procurement Leaders
The obstacles to scaling agentic AI fall across two fronts: technical, covering data and integration, and organizational, covering process redesign, governance, and capability building. The AI models themselves already perform well. What holds deployment back isn’t the technology but the enterprise’s ability to adapt around it. The following five actions address both dimensions.
Redesign processes first—AI can’t create commercial value on its own. Deploying AI into an unchanged process is one of the most common ways organizations limit their own returns. Efficiency gains can accumulate quickly, but they rarely translate into commercial outcomes if the surrounding workflow was never designed to capture them. Freed-up capacity creates little value when the operating model remains unchanged.
Procurement leaders should therefore begin not with selecting technology but with the commercial outcomes they want to achieve. From there, they should redesign the underlying processes and involve IT, finance, and business stakeholders as active co-designers rather than downstream recipients of the result. The greatest value comes not from automating existing work but from rethinking how work gets done.
Build capability alongside technology. Technology alone is insufficient to modernize procurement. Organizations also need to improve data quality, integrate systems, redesign processes, and govern AI responsibly.
Leaders should treat capability building as a broad organizational effort, embedding new skills across tech procurement teams rather than relying on isolated training programs or technology investments. Capability determines not only how effectively an organization uses AI, but also how far it can responsibly extend agent autonomy as deployments mature. Notably, organizations that report the most advanced deployments consistently demonstrate both stronger AI capabilities and higher levels of AI investment.
Evaluate operational and commercial outcomes. The gap between pilot productivity gains and realized commercial value means that a single headline metric can be misleading. Operational benefits typically emerge first because they result from automating or augmenting existing tasks. Supplier-facing outcomes, such as stronger negotiations, better pricing, and improved value capture, take longer to materialize because they depend on redesigned processes and new ways of working.
Leaders should therefore measure both operational and commercial outcomes from the outset. The objective is not simply to manage expectations but to ensure that early productivity gains become the starting point for transformation rather than its endpoint.
Build the data and integration foundations. Inconsistent data and disconnected systems remain the most frequently cited sources of delay, and a successful pilot resolves neither. Agentic systems create value only when connected to the live procurement environment—ERP platforms, procure-to-pay systems, and the broader ecosystem that procurement and adjacent functions depend on. Building those connections is often complex, time consuming, and expensive.
Leaders should sequence data readiness and integration ahead of broad deployment and evaluate pilots not only on how well the AI performed in isolation, but also on whether they established a credible path to production.
“Many expect the tool alone to solve the problem quickly. But without integration into the application ecosystem, the processes, and the data foundation, the value does not materialize.”
– Interview, former CIO of a European media operator
Redefine what procurement is for—not just how it operates. The AI models themselves are no longer the primary constraint. Vendors and system integrators consistently describe agentic capabilities as ready to deliver value at scale. Organizations remain cautious, however, and their concerns around autonomous decision making, accountability, and legal exposure are legitimate. Those concerns largely reflect the challenge of deploying AI within today’s organization, not of designing tomorrow’s.
Those that generate the highest returns haven’t simply automated their existing procurement function, they’ve reimagined it, from a function defined by transaction execution and process compliance to one defined by market intelligence, commercial judgment, and the ability to turn supplier relationships into competitive advantage.
The challenge is ultimately less about advancing the technology and more about deciding what to become using the technology. The most successful organizations will strive to pursue something greater than just a more efficient version of today’s procurement function.
The five actions we have outlined here are designed to help close the gap between where most organizations are today and the leading adopters. Rather than waiting for the technology to improve, procurement leaders should focus on building the organizational capacity to use what is already available to capture a more strategic commercial future.
