The COO of a regional energy network operator receives the company's AI portfolio for the first time. Until now it lived with the CIO and a small innovation team. She scrolls through forty initiatives and notices something odd: most of them are about her own area. Outage triage, crew planning, asset inspections, customer callbacks after a fault. AI in operations turns out to be most of the portfolio, while she has never been asked to approve a single item on it.
That situation is common, and the MetroMinds Library shows why. When you map thousands of AI use cases to the they change, operations comes out on top by a wide margin. This article looks at what that means for the COO: where the ideas cluster, which types of AI are heading your way first, and why operational use cases need a different kind of ownership.
Half of all AI ideas are operational
The MetroMinds Library contains 4,980 AI use cases, each linked to a business capability from L1 to L3. Of those, 1,971 sit in Operations and Delivery and another 553 in Supply Chain and Sourcing. Together that is 2,524 use cases, or 51 percent of the Library.
The other half is spread across ten L1 capabilities, from finance and HR to legal, strategy and technology. None of them comes close. If you want to know where AI will change daily work first, the short answer is: in the parts of the organisation that run the service, the plant or the network.
That does not mean AI is an operations project. It means the business outcome of most AI investments depends on operational processes, operational data and operational people. A portfolio that is governed without the COO is missing the person who has to make most of it work.
In some industries, AI is almost entirely operational
The share differs a lot by industry. In transportation and logistics, 73 percent of the use cases in the Library sit in operations or supply chain. In manufacturing it is 69 percent, in communications 64 percent. At the other end, professional services (28 percent) and media and entertainment (31 percent) have more of their AI ideas in sales, content and knowledge work.
The mix inside that share also differs. In logistics, supply chain makes up 40 percentage points on its own. In communications and manufacturing, almost all of it is operations and delivery: network assurance, field service, production and quality. So even within operations, the COO of a telecom operator and the COO of a logistics company will face a rather different set of ideas.
Where in operations the ideas cluster
Operations is a wide area, but the AI ideas are not spread evenly across it. The 2,524 operational use cases in the Library touch 49 different L3 capabilities, out of 209 in total. So fewer than a quarter of all capabilities hold half of all AI ideas.
The top of the list is about keeping a service running: handling cases and exceptions, managing service delivery and keeping assets reliable and compliant. Resource allocation, service levels and order fulfilment follow. For a COO that is useful information, because it shows where several initiatives are likely to compete for the same process, the same data and the same teams.
Agents and optimisation will reach your area first
Some types of AI are much more operational than others. Of the 600 agentic use cases in the Library, 430 sit in operations or supply chain, which is 72 percent. For optimisation it is 75 percent and for simulation 62.5 percent. Content creation is the opposite: only 17 percent of those use cases are operational.
That matters because agents and optimisation change how work is organised, not only how fast it is done. Supply chain leaders already see this coming. In a Gartner survey of 509 supply chain leaders, 55 percent expected agentic AI to reduce the need to hire for entry-level positions, and 86 percent agreed it will require new processes for developing future talent [2]. Those are operating model questions, and they land on the COO's desk.
As valuable as other ideas, but harder to deliver
Operational use cases in the Library score about the same on business value as the rest: 3.81 against 3.83 on a five-point scale. On feasibility the difference is clear: 2.98 for operational use cases against 3.63 for all others. Operational AI has to work inside live processes, with real-time data, shift patterns, safety rules and physical assets. That makes it harder to deliver than, for example, a drafting assistant for the marketing team.
It also helps explain where companies use AI today. According to Eurostat, marketing or sales was the most common purpose among EU enterprises using AI in 2025, at 34.70 percent, while logistics was the least common at 6.08 percent [1]. The easy wins are being taken outside operations, while most of the remaining value sits inside it.
The box below shows how a COO can use this in a first portfolio review.
What this asks of the COO
If half of the AI portfolio changes operational work, the COO needs a formal role in it. In practice that comes down to a few concrete responsibilities.
- Co-own the portfolio with the CIO, with operations deciding which use cases move to approved for execution.
- Name a process owner for every operational use case, who signs off on the expected benefit and the change in the way of working.
- Look at the capabilities where many initiatives cluster and decide which one goes first.
- Plan the people side early: changed roles, new review tasks and the skills needed to supervise AI in a live process.
- Treat operational data as a shared asset, so that one data improvement can serve several use cases at once.
The COO does not need to become an AI expert for this. The discipline that operations already applies to capital projects, capacity and risk is exactly what an operational AI portfolio needs.
How MetroMinds helps the COO see the operational portfolio
Every use case in MetroMinds is linked to a capability from L1 to L3, so the operational part of a portfolio is visible at a glance. In the Library you can filter by industry, business function, process, data readiness and time to value, which makes it quick to find the operational ideas that fit your organisation.
Within a portfolio, the value and feasibility matrix shows which operational use cases are ready to scale and which are strategic bets that need preparation. Capability impact and workforce impact show which teams and are affected. Portfolio managers and process owners can then agree statuses together, with a recorded reason for every hold or rejection.
For the CIO's side of the same story, read why your AI portfolio is mostly an operations portfolio. And if exceptions and queues are your daily reality, see exception handling: the hidden cost AI can reach first.
Want to see which part of your AI portfolio lands in operations? Book a demo
Sources
- Eurostat, "Use of artificial intelligence in enterprises", Statistics Explained, data extracted December 2025.
- Gartner, "Gartner Survey Shows 55% of Supply Chain Leaders Expect Agentic AI to Reduce Entry-Level Hiring Needs", press release, 25 February 2026.
- MetroMinds Library, 4,980 AI use cases with benchmark ranges based on real-world implementations, August 2026.
