MIT NANDA’s report last month that revealed a 95% rate of failure of Gen AI pilots sparked considerable media coverage and, no doubt, many deep enterprise conversations asking the question: Which AI business use cases do succeed at delivering ROI?
The correct use cases can deliver a “Yes” answer to this simple litmus test: Does adding more human labor to scale the function have the reverse of the desired effect, reducing productivity by adding unacceptable workflow bottlenecks, costs and complexity?
One such use case is enterprise procurement. And because it’s also too strategic and nuanced to hand over entirely to machines, procurement gains additional credibility and veracity from the humans in the loop. The perfect symmetry here is that the human capacity to provide this necessary oversight is actually freed up by agentic AI doing the drudge work.
By applying agentic AI to large enterprise procurement scenarios, companies such as HP, Fidelity Investments, BT, Santander Bank and Tesco have measured ROI figures. Tesco’s Save to Invest program, for example, was called out in the retailer’s 2024 Annual Report for helping to deliver £1.2bn in savings, enabling the company to “offset inflation and create headroom to fund investments.”
If we agree that procurement is a solid use case, the other ingredient for ROI is ensuring AI is engineered to produce optimal outcomes at the speed of business. With agentic AI, the temptation is to go for vendor or home-grown systems that use lots of agents to handle specialized tasks like POs, supplier search, SOW-writing or negotiation. This multi-agent approach, however, reduces agents to siloed task executors that lack perspective, intelligence and range.
The Many Advantages of Range
In one of my favorite books, “Range: Why Generalists Triumph in a Specialized World,” the author David Epstein argues that in a complex and unpredictable world, the most successful experts often start as generalists. Exploring broadly across different domains, rather than specializing early, enables deeper creativity and better problem-solving, he maintains.
Range challenges conventional thinking behind specialization and Malcolm Gladwell’s “10,000-hour rule,” demonstrating the relative advantages of broad experience across sports, science and business. It explains why many company leaders ultimately become innovative thinkers and gain long-term success after coming through cross-functional management development programs.
Why Range Matters to AI Agents in Procurement
We humans learn linearly, building expertise over time through trial and error. Even the most versatile polymaths need years to develop true cross-functional prowess. AI agents, by contrast, can instantly ingest and synthesize the experience of thousands, switch roles fluidly and see patterns humans can’t.
That difference highlights why scaling procurement solely with human labor is so difficult. Procurement may be a specialist function, but it spans an unusually wide set of activities. On a typical day, a procurement executive might hop from category management to sourcing to supplier relationship management to ensuring supply continuity. To manage billions of dollars in trade, there’s a dizzying range of skills, experience, knowledge and strategic execution required. The domain expertise spans sourcing, supplier discovery, stakeholder intake, requirement definition, RFP creation, negotiation, contract management, compliance, risk assessment, legal alignment, finance integration and ongoing performance management. All these activities generate huge amounts of documents and data.
Traditionally, this work is split across teams, creating countless silos and hand-offs, which is why scaling manually becomes so counterproductive. Using AI agents solves this dilemma because you can deploy a single procurement agent with the range and depth to handle multiple interconnected processes. Using a single agent for procurement can deliver what users need: a cohesive, intelligent partner designed to reason, learn and operate across every stage of the sourcing journey.
What to Look for and What to Avoid
To gain the best ROI outcomes, look for a single AI agent with a wide range of procurement expertise and broad contextual understanding. The agent should be designed to reason, learn and operate across every stage of the lifecycle, from “I have a need” to “I am now engaged with my supplier to satisfy my need.” It should navigate simple and complex sourcing requests that span industries, regulations and internal dynamics, direct and indirect, across sectors like telecom, financial services, consumer goods and retail, technology and enterprise IT, and others. It should manage everything from tail-spend and complex services to multi-billion-dollar strategic sourcing, covering thousands of categories with deep expertise and autonomous decision-making.
What to avoid? Anyone selling lots of task-based AI agents. It might help them to “land-and-expand” their sales engagement with you, but it will never help you achieve that elusive ROI.
