What Is an Operating Model? The Seven Parts, and What AI Changes.

A company buys the AI tools, runs the pilots and trains everyone. Six months later the work moves exactly as it did before: the same approvals, the same handoffs, the same people deciding the same things in the same meeting.

The tools changed and the operating model did not. That is the most common way AI programs stall in 2026, and it is why "we need a new operating model" has become a sentence every executive team hears, usually with no definition attached.

In Rewired Leadership I describe leadership in an AI-enabled company as an operating system, one that "determines how information flows, how decisions get made, what happens automatically, where humans intervene, and how exceptions are handled." An operating model is that system, written down.

One note on terms: in finance, an "operating model" is a spreadsheet that forecasts a company's revenue, costs and cash. This article is about the other meaning, how an organization runs.

Operating model vs business model, strategy and org structure.

Four terms get used as if they were one. They answer different questions, and mixing them up is how a company redraws its org chart and announces a new operating model.

The operating model vs business model distinction is the one people trip on most. Two companies can sell the same thing to the same customers at the same price, and still run completely differently.

One decides at the edge in hours. The other escalates everything and waits weeks. Same business model, different operating model, and over time very different results.

The org structure is the other trap. It is one part of the operating model, the most visible and the easiest to change, so most "new operating models" turn out to be reorganizations that redraw the boxes and leave the decision rights where they were.

The components of an operating model.

Published operating model frameworks run from three parts (people, process and technology) to a dozen. These seven are the ones that decide how work actually moves, and the fourth is the one AI added.

1. Work and processes. How work flows from request to result, end to end and across functions. Most delay lives in the handoffs between teams, which no single function owns.
2. Structure and roles. The units, reporting lines, spans and layers, and what each role is accountable for. More and more, roles are defined by capability rather than title, so people can move as the work moves.
3. Decision rights and governance. Who decides what, inside which guardrails, and what escalates. This is where most operating models fail first, because it is rarely written down.
4. People and AI: who does which work. Which work agents execute, which they prepare for a person, and which judgments stay with people. In my book this is Judgment Allocation: "AI handles execution, analysis, and coordination. Humans handle ambiguity, ethics, trade-offs, and meaning."
5. Technology and data. The systems the work runs on, and the data that tells people and agents what is happening. This is the part companies change first, and often the only one.
6. Performance and rewards. What gets measured, rated and paid for. If this part contradicts the other six, it wins, which is why a review that grades last year's job can quietly undo a redesign.
7. Cadence and correction. The rhythm of meetings, reviews and decisions, and how the model learns from what goes wrong. Strong models fix the guardrail that failed instead of blaming the person who hit it.

The company in the opening changed part five and expected the other six to follow. They rarely do: habits, incentives and calendars hold them in place.

Why the operating model matters more in 2026.

The limit on what AI is worth to a company has moved from the technology to the way the company runs, and the people closest to the technology now say so themselves.

In Deloitte's 2026 study of 662 senior technology leaders at companies with $1 billion or more in revenue, 81% said they can deploy and govern AI at scale today. Nearly 75% also said their operating model must change within 12 to 18 months to sustain that progress (Deloitte, June 2026).

Only about a quarter change their operating model continuously; the most common cadence is quarterly. Deloitte's own conclusion is worth keeping: "the future operating model is less about organizational structure and more about coordination."

Operations leaders see the same gap from the floor. In PwC's 2026 survey of 767 US operations and supply chain leaders, 83% said AI agents and automation will speed up the breakdown of functional silos.

Only 41% run a horizontal, networked operating model today (PwC, 2026).

Where the operating model lags: how US operations leaders see AI agents in 2026

Expect AI agents and automation to break down functional silos 83%

Run a horizontal, networked operating model today 41%

Comfortable letting AI agents run full end-to-end processes 37%

Have fully embedded an AI strategy across business units 27%

That 42 point gap is the operating model problem in one number. Agents will cross functional lines whether or not the structure, decision rights and measures were built for it, and where they were not, the old model wins.

What changes in an AI operating model.

An AI operating model is not a separate model bolted on for the technology team. It is the same seven parts, redesigned on the assumption that agents do part of the work.

BCG's study of about 50 AI front-runner firms, built on interviews with their leaders between March and August 2026, found the same five shifts again and again (BCG, September 2026).

Those firms outperform their peers by more than 11 percentage points a year in shareholder returns. An interview study shows correlation, not cause, but the pattern in how they run is consistent, and it maps onto the parts of the model.

Decision rights come first, because every other row depends on them. In BCG's study, reversible, lower-risk calls are pushed down to whoever is closest to the work, and irreversible ones tend to stay up. Authority follows the cost of being wrong, not rank.

In the book I put the same idea in one line: "Guardrails replace permissions. Escalation becomes the exception, not the default."

The judgment row is the one most companies skip. Deciding which decisions stay with a person is a design choice, and if nobody makes it, it gets made by default, one workflow at a time.

Leadership failure occurs when judgment is either hoarded at the top or abdicated to machines.

The rewards row is where redesigns quietly die. Some three-quarters of BCG's front-runners weigh judgment over activity when they evaluate people. Most companies still count output, and if the scorecard rewards volume, people keep producing volume, now with agents helping.

The managers row is the hardest to staff. A majority of BCG's interviewees described middle managers becoming player-coaches who still do part of the work, a job few of them were trained for.

How to design an operating model.

Operating model design goes wrong most often in its order: teams draw the structure first and argue about decisions later. These seven steps, the order I use with executive teams, run the other way.

The same steps work whether you are creating an operating model for a new unit or redesigning one that has run the same way for a decade.

01Start from the few outcomes the strategy depends on

Name the three to five outcomes the business must deliver in the next two years, and the work that carries each one. Start there, not from the org chart.

It is the same list a talent strategy starts from, which keeps the two plans pointed at the same work.

02Map how the work really flows today

Follow a few real pieces of work end to end, across functions, and write down every handoff, approval and wait. That is your current operating model, which is rarely the one on the slide.

People inside a system rarely see their own workarounds. An outside read of where the current model breaks finds them faster.

03Decide which work goes to agents, and which stays with people

The Judgment Audit from Rewired Leadership sorts each core function's output into three groups: execute, which you automate; synthesize, which you augment; and judge, which you keep with people and elevate.

The point is to free people for the third group. Cutting the first and stopping there produces a cheaper version of the old model, not a new one.

04Write the decision rights before you draw the boxes

For each important decision, write who decides, inside which guardrails, and what escalates. Set the rights by reversibility: easy to undo goes to whoever is closest to the work, hard to undo stays higher.

Writing them down does more than speed decisions up. Clear decision rights and guardrails are how people learn that their judgment is trusted.

05Then set the structure and roles around the flow

Now draw the boxes, around the work you mapped rather than the functions you had. Keep spans wide enough that managers coach instead of route, and define roles by capability rather than title, so people can move as the work does.

06Line up measures and rewards

Change what the review scores before the new model launches, not a year after. Measure the quality of judgment, not the volume of activity.

Then check every bonus and incentive for anything that still pays for the old way of working. One leftover metric can hold a whole redesign in place.

07Pilot it in one unit, then set the cadence for changing it

Run the new model in one unit for a quarter, fix what breaks, then extend it. Treat the target operating model as a direction, not a finish line.

Review it at least quarterly. The work will keep moving as the agents improve, and a model that was right in January can be wrong by June.

Why operating model changes fail.

I have watched executive teams invest in new platforms, new tools and new operating models while the experience on the ground barely changed. The reason is the one I give in the book: "leadership behavior never rewired to match the new reality."

The opposite mistake fails too. "Changing leadership behavior without changing leadership design rarely works," because under pressure people revert to whatever the system rewards. The design and the habits have to move together.

Five signs, the failure signals from the operating system at the back of my book, tell you the model is the problem, not the people.

• Leaders are repeatedly asked to restate direction.
• Leaders become approval bottlenecks.
• Leaders are buried in decisions AI could already support.
• Leaders only hear about problems when they become crises.
• The same problems repeat quarter after quarter.

The second sign is the most common at the top. If the CEO keeps checking every decision, "the organization will move at the speed of the CEO's inbox."

Fixing that takes leaders whose habits match the new design, not another reorganization.

An operating model example.

Most operating model examples online describe business models: subscriptions, marketplaces, franchises. This one shows the parts of one team's model before and after agents arrive. It is illustrative, not a client case.

Take a customer operations team of 200 people handling refunds and complaints.

• Work. Before: people read every ticket. After: agents triage and draft replies, and people take what the agents flag.
• Decision rights. Before: every refund over $100 needs a supervisor. After: agents settle refunds under a set limit, and people own the rest and every exception.
• Structure and managers. Before: layers of supervisors who route work. After: team leads become player-coaches who work the hard cases with their people.
• Performance. Before: tickets closed per hour. After: the quality of escalations and what happened to the customer.
• Cadence. Before: a process review once a year. After: a weekly look at where the guardrails were too tight or too loose.

Two real cases from the BCG study show the parts moving together. At Reserv, an insurance technology company where the staff closest to a problem design the fix, the time to build a new learning and development program fell from 18 months to four weeks.

At Walmart, employees often move roles every 18 to 24 months, so experts document how the work actually gets done, and the company is creating orchestration roles to manage its AI systems. When people move that often, the model has to carry the knowledge.

An operating model on one page.

An operating model template does not need a consultant's canvas or a slide deck. If you can answer these seven questions on a single page, you have one. If you cannot, you have found where to start.

1. Outcomes. Which three to five outcomes does the strategy depend on?
2. Work. How does the work behind each one flow, end to end?
3. Decisions. Who decides what, inside which guardrails, and what escalates?
4. People and AI. Which work do agents do, and which judgments stay with people?
5. Structure. Which teams and roles carry the work, and who coaches them?
6. Measures. What gets measured and paid for, and does it reward judgment?
7. Cadence. How often do you review the model, and when is the next review?

Frequently asked questions.

The short version.

An operating model is how a strategy turns into daily work: how work flows, who decides, which work goes to people and which to agents, and what gets rewarded. Most companies are changing their tools faster than their model.

Start with decision rights and the split of work between people and AI, then build the structure around them. That is the core of redesigning structure, roles and decision rights for the way work runs now.