Most operations functions today are crowded with disconnected point solutions, some effective, some not. A company might have a world-class forecasting engine and a cutting-edge logistics optimizer, but they rarely work in unison. AI agents change that. By acting as intelligent orchestrators across inventory, procurement, and the commercial organization, they empower COOs to move beyond optimizing individual parts of the value chain. Finally, the whole system can respond as one.
Operations has been using advanced analytics for years. So what is different about this wave of AI?
Operations has been at the leading edge of analytics for a long time because it is such a data-rich environment. Biopharma companies began using sophisticated algorithms to optimize yield more than a decade ago. Supply chain teams across industries have also made real advances in forecasting, inventory, logistics, and network optimization.
When generative AI first emerged, people naturally looked for language-heavy applications such as document search, quality processes, and regulatory submissions. Today, the conversation has moved on. Companies now see an opportunity to connect the point solutions they already have across operations. Orchestrator agents can bring those pieces together along an end-to-end supply chain, a make-to-order process, or the full journey from source to pay. That is the larger opportunity with AI that companies are now trying to understand.
When you look at the companies furthest ahead, what does good look like? Where are they finding value beyond straightforward efficiency gains?
The strongest companies have a clear view of where the value pools are and where human efficiency is a material part of the opportunity.
Take procurement. A company may spend $10 million to $20 million running a function that manages $5 billion in external spend. Making those processes more efficient is important, but the primary aim is rarely to reduce headcount. It’s to help the team manage that $5 billion more effectively and capture savings from the much larger spend base.
Supply chain is similar. In many companies, reducing the number of planners will not be a game changer, but improving service and cost metrics will. That balance can shift in large supply-chain organizations with thousands of planners in high-cost locations. But the point is to be clear about where human efficiency deserves attention and where the larger opportunity lies in the other metrics the function manages.
Can you bring that to life in supply chain and logistics? What are leading companies now able to do that they could not do before?
Historically, a company might conduct a network optimization every three years or when a significant event required it. As its capabilities improved, the company might optimize routes every six months or once a year.
The best companies are now thinking about the entire value chain as being always on. They have a network, an inventory strategy, a changing demand forecast, and some flexibility in how they use their assets. Those elements can be optimized together through dynamic routing, dynamic product placement, and dynamic inventory optimization. As conditions change, the company changes where it positions inventory and how it routes vehicles to customers.
This is where the point solutions companies developed in the past can be brought together by orchestrator agents. Otherwise, multiple teams are separately trying to understand the forecast, inventory positions, service levels, and the best way to route shipments, and someone then has to piece all of that together. With AI, the company can address those decisions together and become more agile in responding to customers while managing costs, working capital, and asset utilization.
Some leading companies already have this capability. In cross-functional, fully orchestrated deployments, we are seeing on-time and in-full service improve by roughly 2% to 3%. We are seeing improvements of about 15% to 20% in shipment logistics, excluding warehousing. Inventory improvements range from 10% to 25% when raw materials through finished goods are optimized together, and scrap can fall by 20% to 50%. Those numbers are why companies have to be careful about following workforce efficiency alone.
Could we see a nearly all-AI logistics company, or parts of operations becoming almost entirely AI-driven? Is that realistic?
Yes, it is realistic. But I think we get a lot more with a human in the loop playing a different role. Today, most demand planners spend their time adjusting statistical models, trying to figure out the right inputs and the best fit. We’ve progressed to the point where AI can handle that heavy lifting—it can identify the best-fit model and determine whether the data needs to be adjusted.
What I want the demand planners to do now is become orchestrators and shapers of demand. I want them to sit down with the commercial team and ask where demand is likely to soften, which markets might need a promotion, and what uplift to expect. They should be working directly with retail and B2B customers to capture on-the-ground insights we can feed right back into the AI to improve forecast accuracy.
You can automate the statistics, but you get much more value when the human uses that baseline to actively shape the business.
As people begin to play different roles, where are companies running into the greatest organizational challenges?
The functional boundaries start to move. When a planner becomes more involved in shaping demand, the commercial team may ask where its role begins and ends. When procurement uses AI to negotiate terms beyond price and challenges technical specifications, other teams may find themselves working in the same territory.
The organization then has to rethink its decision rights. What decisions are we trying to make? What actions follow from those decisions? Which skills and stakeholders should be involved?
That is a better starting point than defending existing swim lanes. It may also require companies to reconsider how the functional silos themselves are constructed. I have not seen anyone fully solve that yet.
Middle management is especially important. Many middle managers have built their expertise around being the steward of a process and improving it over time. Now the company is asking them to work in a very different way, while still relying on the institutional knowledge they hold. Leaders often understand the strategic argument quickly. The larger challenge is bringing the organization with them and having a direct conversation about how roles will change.
What does an agent-ready operations function need to have in place before it can deploy AI meaningfully at scale?
I would start by considering three questions. What value pools are we trying to address? What are the important decisions we need to make? What data do we have?
The data does not have to be perfectly clean. You need a reasonable understanding of its structure, where it is strong, and where it will limit you. Companies can improve the data as they move rather than waiting for a multiyear data cleansing program to be completed.
I would also resist the instinct to begin by mapping the current process in detail and trying to fix it. That can place artificial constraints on the design. Teams should look at the existing sequence of steps and ask how AI can make each one faster.
Begin with value, decisions, and data. That gives you more freedom to rethink how the work should happen.
Pilots are important, but they can also leave companies with another collection of point solutions. How do you design a pilot with scale in mind from the start?
The purpose of a pilot is to learn how to scale. The scope can be narrow, but it should sit within an enterprise-wide or operations-wide priority and use common technology standards from the start.
Challenges arise when every subfunction launches its own pilot with different standards. A demand-forecasting pilot may work, as may a yield-optimization or crewing pilot. But if they use different platforms, data structures, security rules, and architectures, those improvements may never scale. You have a thousand blooming flowers. They may look pretty and work fine on their own, but when the company tries to connect the pieces, they break apart and the pilots become throwaway work.
To manage all this, you need a capable, cross-functional central team. It may be only four or five people, with representation from information security, technology architecture, data management, legal or ethics, and the business. The company also cannot treat that team as a temporary project. Its members have to keep pace with changes in the technology, security requirements, and the bounds within which AI can be deployed.
All of these approaches seem to be changing what it means to be an effective COO. How is the role changing, and what does that mean for how companies develop future COOs?
The traditional COO was often the deep technical expert in the company—the manufacturing or supply chain specialist. But with agentic AI taking over that technical execution, the COO has to become more of an orchestrator across the enterprise. Cost and quality remain fundamental, but the role can now contribute much more to how the company competes.
With AI managing those daily processes, the COO needs to be a general manager of sorts. That requires aligning directly with commercial leaders on broader market strategies and margin expansion. It means working with the CHRO on how to deploy human talent alongside these new AI systems, and partnering with the CIO or CTO on platforms, standards, and make-versus-buy decisions. You can’t just develop future COOs to be narrow functional experts anymore. You need leaders who can pull all of those pieces together.
