When everyone has AI, your operating model will set you apart

Most operating models still treat tasks as the unit of value creation. As execution becomes more automated, a greater share of value creation instead depends on the decisions that set direction, allocate resources and resolve tradeoffs. As AI takes on more execution, the differentiator becomes the decisions that determine what gets done, as well as the proprietary data and human judgment behind them.

The decisions that matter most are often choices that directly affect important business outcomes: what price to offer, how much inventory to position or which clinical trial site to select. Getting one of these decisions right can have a material impact on the outcome, often disproportionate to the value available from optimizing the work around it. This makes the decision itself a particularly important source of value in an AI-enabled organization. Companies, therefore, need to identify the relatively small number of decisions that matter most and decide how people and AI should make them.

For each decision, leaders should specify the outcome it serves, who is accountable and how AI should contribute to the decision. As Figure 1 shows, the right decision mode (that is, how people and AI share the work of making a decision) depends on its level of risk and complexity. Leaders should also consider where people need to retain control for reasons such as ethics, accountability, legitimacy or talent development. Named human stewards and clear guardrails help keep decisions traceable and auditable.

Figure 1: The decision mode framework

The right balance of human and AI roles depends on the decision.

Most companies have spent years improving performance within individual functions. Now, with decisions better defined, the bigger opportunity lies across the organization, where handoffs create delay, duplication and customer frustration. Companies can capture that value by organizing work around outcomes that cut across functions.

Companies can give greater weight to horizontal value streams, or “mega-processes,” that run across functions. How far they go will vary, from fully resourced value streams with their own budgets and accountability to a more balanced model. In either case, the shift extends beyond reporting lines to the operating model itself, including structure, processes, talent and incentives. Customer journeys and end-to-end flows (such as signal to shelf, quote to cash, claim to settlement and patient to outcome) then become another way of organizing and managing the work.

Within each value stream, work is organized around a shared outcome rather than the priorities of individual functions. As Figure 2 illustrates, people and AI can play different roles in the decisions along the way, while more of the execution becomes automated. And functions still own the strategy, standards and expertise on which the value stream depends.

Figure 2: Inside the value stream

In AI-enabled operating models, value streams cut across functions while functions keep their core responsibilities

As work is redesigned around AI, companies also need to rethink the distribution of work. AI agents already perform meaningful work across many enterprises, resolving customer issues and managing supply chain tasks and processes. In fact, almost every large company deployed at least one agent last year.¹ Yet most agents have no named owner, clear line of accountability or place in the workforce plan. As agents take on more work and a greater share of routine decisions, gaps in ownership make it harder to govern their performance, resolve exceptions and determine who ultimately owns the outcome.

This will become harder to sustain as companies redistribute more work between people and AI. In our recent survey of 3,000 C-suite leaders at some of the world’s largest companies, 58% of executives said their organizations plan to change how work is distributed between people and AI in the next 12 months, more than double the share who said the same at the start of 2026.²

Managing people and AI as one workforce begins with accountability. Every agent should have a named human owner who is ultimately accountable for what it does. Policies and guardrails should also apply consistently across people and machines, with a clear record of how authority is delegated.

Beyond accountability, planning changes too. The team, not the individual, has always been the real unit of performance, and its composition has always mattered more than the capabilities of any one member. What AI adds is a new layer to solve for: the team is now a mix of people and agents, and that mix has to be designed rather than assumed. Once leaders know the outcomes and decisions a team owns, they can set the balance of human judgment and machine capacity, to deliver at scale. Within the team, human work concentrates where it is irreplaceable: knowing when to trust the system, when to step in, and when empathy matters most.

As human and machine capacity are managed together, companies also need a better way to measure what that capacity costs. Yet most financial and workforce systems still budget capacity in two columns: payroll and contractors.

While AI agents have become a third source of productive capacity, their costs behave differently from those of employees and contractors.

Employee labor is largely a committed cost, while machine capacity is consumption-based: it can expand and contract with use. Its economics resemble cloud infrastructure more than payroll. AI spending also does not scale predictably with headcount. In a recent Accenture study of AI deployments, fewer than 10% of users and workflows drove the vast majority of AI spending.⁴ As a result, two similarly sized workforces can have very different AI costs depending on where and how extensively AI is used. Most planning systems, however, are still built around headcount and are not designed to account for AI’s variable usage-based costs or the uneven distribution of AI spending across the workforce.

As more work is organized around critical decisions and horizontal value streams, cost should be measured against the outcomes that those value streams deliver. The right measure is, thus, cost per outcome: What did the organization achieve, and what did it cost to achieve it?

Measuring cost per outcome requires people, agents and inference to sit on more comparable terms in planning systems. More broadly, all five shifts described in this article depend on something few enterprises have today: a connected view of the people and agents doing the work, the decisions that shape it, what that work costs and the governance around it.

Without that connection, leaders can’t reliably determine the cost of an outcome or make informed choices about the mix of human and machine capacity behind it. This is both an operating model issue and a finance issue, so the CHRO and CTO need to address it with the CFO. Until they have a common view, the economics of the operating model remain invisible.

That connected view also creates the conditions for something more valuable: an organization that gets smarter with every decision it makes. Most organizations record outcomes but rarely capture the context behind them, including what managers knew, what they expected, what AI recommended, why people accepted or overrode that recommendation and what happened as a result.

Institutional knowledge lives in the heads of experienced people, not in the systems around them. AI can change the economics of organizational learning by making that knowledge easier to capture, share and apply. Indeed, management thinkers have long argued that organizations improve by learning continuously and correcting errors: Peter Senge popularized the idea of the “learning organization,”⁸ while philosopher Karl Popper emphasized progress through testing ideas and learning from mistakes.⁹

AI can make those principles more practical at scale by helping organizations retain what they learn and apply it to future decisions and work. “Decision history,” the accumulated record of the context, recommendations, judgments, actions and outcomes associated with pivotal decisions, can inform both future decision makers and future AI systems.