A target operating model describes how a business will run once a planned change is complete: which processes it performs and how, who does the work and where, what information and systems support it, which suppliers it relies on, and how it is managed. It is the design document behind a reorganization, a merger integration or a new strategy, and the record of such designs is poor. A 2015 McKinsey study found that less than a quarter of organizational redesigns succeed, with 44 percent running out of steam after getting under way and a third failing to meet their objectives after implementation. The target operating model now has a new question to answer, which is how much of the work agents will do, and McKinsey's survey data suggests the answer depends on design: redesigning workflows had the biggest effect on whether organizations saw EBIT impact from generative AI, out of 25 attributes the firm tested.
What goes into a target operating model
A widely used framework for the contents is Andrew Campbell's Operating Model Canvas, written with Mikel Gutierrez and Mark Lancelott and published by Van Haren in 2017. It breaks an operating model into six elements whose initials spell POLISM. Processes are the work and its sequence, from order entry to the month-end close. Organization is the people, roles, reporting lines and decision rights. Locations are where the work happens, including offshore centers and outsourced sites. Information is the systems and data the work depends on. Suppliers are the parts of the work bought from outside, from software to outsourced services. The management system is how the whole is planned, measured and governed.
The order matters. Processes come first because the other five elements exist to serve them: the organization is designed around the work, the systems carry the work's information, and the measures track the work's results. A target operating model that starts from the organization chart designs the reporting lines first and fits the work around them, which is how many of them end up describing boxes rather than work.
How a target operating model is built
The usual method runs in five steps. The strategy is turned into a short list of design principles, statements such as serving mid-market customers from one location or closing the books in five days, which every later choice must satisfy. Then the current operating model is described, the future state is designed element by element, and the gap between the two is identified, before the changes are sequenced into a transition plan with owners, dates and measures.
The level of detail varies more than the method does. Campbell's own guidance is that the canvas is for high-level operating models of one page or 10 pages rather than detailed ones of 100 pages or 1,000, and that distinction is where many projects go wrong. A one-page model is enough to agree a direction and too thin to run a business on, and the detailed design that would be enough is often left to the implementation team, after the consultants who drew the one-page version have gone.
Why so many target operating models stay on slides
The McKinsey figures above cover organizational redesigns in general, and the reasons such designs fail are familiar to anyone who has been through a reorganization. The design is drawn from interviews and workshops, so it describes the work as managers understand it rather than as the work runs, and the difference shows up only when the new model meets the old exceptions. The design is delivered as a document, so the organization receives a description of the future and a plan, and nobody is accountable for the future once the plan is handed over. We describe the consulting model that produces this outcome in why your AI consultants left you a deck.
A design that starts from evidence of the work avoids the first problem. The current state can be mapped from system logs and captured desktop work rather than from memory, with the exception paths and their frequencies included, by the method we describe in business process mapping. A future state designed against that map is designed against the real volume and the real exceptions.
Where agents now sit in the target operating model
The processes element used to have one kind of performer: a person, sometimes helped by a system that stored the record. It now has three. People do the work that needs judgment, authority or a relationship. Rules and conventional automation handle the routine steps with structured inputs. Agents handle steps that involve reading documents, making routine decisions and working across systems. Gartner predicts that by 2028 at least 15 percent of day-to-day work decisions will be made autonomously through agentic AI, up from 0 percent in 2024, and the Gartner analyst Anushree Verma offers a useful rule of thumb for assigning the work: agents where decisions are needed, automation for routine workflows and assistants for simple retrieval.
A target operating model that includes agents has to answer four questions for each process step: who or what performs it, who supervises the performer, where the exceptions go, and what control confirms the step was done correctly. The organization element changes with the answers. Roles that consisted of executing routine steps become roles that supervise agents and resolve their exceptions, which is the shift we set out in what the operator's role becomes after automation. The companies that get the most from AI are the ones making these changes: McKinsey's 2026 survey found that nearly three-quarters of its high performers report fundamentally redesigning workflows because of their AI use, against a quarter of other respondents.
A target operating model template for one process
A template that forces the design down to the level of the work fits on one page per process. For each step in the process, it records eight things: what the step does, who or what performs it, who supervises the performer, where the exceptions go, what control confirms the step was done correctly, which system holds the result, where the work is done, and which supplier, if any, is involved. Above the steps sit three measures for the process as a whole, usually volume, cycle time and the share of cases that reach a person, each with its value today and its target.
The template does two jobs. It makes the design testable, because every row can be checked against the process as it runs, and it makes the transition plan concrete, because each change is a change to a row. The measures have to be taken before the design is finalized, by the method we set out in baselining a process for outcomes, or the target operating model has no way to show that the future state is better than the present one.
A reasonable counter, answered
A reasonable counter is that agents are too immature to design an operating model around: Gartner itself predicts that over forty percent of agentic AI projects will be canceled by the end of 2027, and McKinsey's 2025 survey found that in any given business function, no more than 10 percent of respondents said their organizations were scaling AI agents, a figure the 2026 edition shows rising, most of all at large companies. Both findings are real, and a target operating model that assumes agents will run whole functions by next year is a bad design. The answer is to design at the level of the step. An agent that reads supplier invoices and prepares the match is proven today, and an agent that negotiates with a supplier is not, so the first belongs in the design and the second belongs in a later version. The four questions above work the same way whether a step is performed by a person or an agent, which means a model designed this way can move steps between performers as the technology proves itself, without redrawing the organization each time.
Common questions
- What is a target operating model?
- A target operating model is a description of how a business will run once a planned change is complete. It sets out which processes the business performs and how, who does the work and where, what information and systems support it, which suppliers it relies on, and how the whole is managed and measured. Companies draw one before a reorganization, a merger integration, a new strategy or a major technology change. It sits between the strategy, which says what the business is trying to achieve, and the implementation plan, which says how to get there.
- What is included in a target operating model?
- Andrew Campbell's Operating Model Canvas, a widely used framework, breaks an operating model into six parts whose initials spell POLISM: processes, organization, locations, information, suppliers and management system. Processes describe the work and its sequence; organization describes the people, roles and reporting lines; locations describe where work happens; information covers the systems and data; suppliers covers what is bought from outside; and management system covers planning, measurement and governance. A 2026 target operating model also has to say which process steps agents perform, who supervises them and where their exceptions go.
- How do you create a target operating model?
- Start with the strategy and turn it into a small number of design principles that the operating model must satisfy. Map the current operating model from evidence of how the work runs, then design the future state element by element, process first, because the other elements follow from the work. Compare the two to find the gaps, and sequence the changes into a transition plan with owners and measures. Test the design on one process before rolling it out, because a design that has never run is a hypothesis.