How innovation teams are structured

Start with what has to ship and where it needs authority to move. Start there. The structure answers the rest, from budget ownership to approval rights to the number of handoffs a new idea has to survive, so the safest default is still a small setup with extra governance added only when the work starts breaking.

Use the org charts on this page as a decision aid. Seven models, one problem. A centralized lab, venture studio, CoE, embedded team, transformation office, Corporate VC unit, and open innovation layer each place authority in a different spot, which changes how fast work moves and who has to survive the politics around it.

TL;DR

• Structure decides what innovation work can ship.
• Labs and CoEs buy legitimacy faster than adoption.
• Embedded teams move faster, but fragment without shared rules.
• Venture studios need patient capital and direct sponsorship.
• Corporate VC buys options, not product delivery.

Teams choosing an innovation team structure need to avoid the model that looks credible in a board deck and then stalls in execution.

What does your innovation structure actually commit you to?

An innovation team structure is the operating design that decides where authority, budget, and delivery risk sit. That choice determines whether new work behaves like a staff function, a product team, an investment arm, or a coordination layer before titles and headcount settle the details.

"When people ask me what Amazon's organizational structure looks like, I often say it's a federation of startups more than it is a single entity."

— Andy Troutman, AWS re:Invent 2020: Two-pizza teams: Organizing for innovation

As Andriopoulos and Lewis argued in 2008, structural ambidexterity works by separating exploration from exploitation instead of pretending one team can do both on the same cadence. The charts below show whether innovation is expected to act like a staff function, product unit, venture investor, or coordination layer.

Start with the job, not the label. A lab protects uncertain work. A CoE standardizes how other teams work. An embedded team owns work inside the line, then lives with the adoption consequences.

If the mandate is cross-BU exploration, a lab, venture studio, or open innovation layer makes sense. If the mandate is product change inside an operating unit, embedded teams or a federated innovation model survive planning better than a distant central group.

Where do the seven models sit on the centralized-to-distributed spectrum?

The seven models sit on a spectrum from centralized control to embedded execution. As authority moves closer to the business unit that must absorb the work, shipping usually gets easier, while enterprise coordination, standardization, and portfolio visibility get harder to maintain across units.

In 2017, Tõnurist and colleagues found that innovation labs were often created for legitimization and environmental complexity, not just for execution. In 2011, Petroni and colleagues described open innovation as a shift toward matrix and network organizational models.

Why do centralized innovation labs keep producing activity instead of shipped products?

Centralized innovation labs are easier to defend in an executive review than to wire into a business unit's release cycle. They can produce demos, prototypes, tours, and portfolio artifacts quickly, but they still depend on another operating unit to absorb delivery risk, fund rollout, and make the work real in market.

Thinkers50's older warning remains useful because it named the failure pattern directly:

"activity and vanity metrics; events hosted, press mentions, experiments run, number of customer conversations, number of hackathons and minimum viable products launched. Number of validated business models? Zero."

— Thinkers50, Thinkers50

Steve Blank's test is cleaner:

"If you're interested in innovation inside your company, you should be looking around and asking, are we creating innovation theater or are we actually delivering innovation?"

— Steve Blank, Notice Disruption and Innovate Through It, with Steve Blank

"Because of the bureaucratic nature of newsroom authority, ideas created in the intrapreneurial unit are ultimately isolated from the lifeblood of the broader news organization."

— Naldi et al., The Isolation of Innovation

As Naldi and colleagues documented in 2015, ideas born inside a separate innovation unit can end up isolated from the operating core that would have to adopt them. BCG's 2024 innovation study found that only 3% of surveyed companies qualified as innovation ready even though 83% ranked innovation among their top three priorities. If a unit must ask another unit for adoption, you did not design a delivery structure. You designed a handoff.

Centralization earns its keep when talent is scarce, infrastructure is expensive, or the work needs insulation from quarterly delivery pressure. The mistake is using a lab for product-adjacent work that should have lived closer to the line.

What are the seven innovation team structures, and what is each one built to do?

These seven structures solve different organizational problems, so the right choice depends on mandate, time horizon, and where adoption authority lives. The useful comparison is not which model sounds modern. It is which one can do the job without creating a fatal handoff.

Labs protect uncertain work. CoEs standardize and coach. Embedded teams own work inside an operating unit. If that three-way distinction is fuzzy, the rest of the design usually will be too.

Innovation Lab. Best for protected exploratory work the core cannot absorb yet. Breaks when adoption depends on reluctant business units.

Venture Studio. Best for net-new ventures with dedicated operators and patient capital. In March 2026, MIT Sloan Management Review argued that internal venture studios draw on strategically relevant pools of ideas, resources, and talent that a corporation has already cultivated.

"Internal venture studios allow organizations to tap strategically relevant pools of ideas, resources, and talent that have been cultivated over time."

— Constanze Coelsch-Foisner and Fiona E. Murray, MIT Sloan Management Review, SC Ventures' Venture Studio: Centralized Corp…

It breaks when leadership wants venture upside on annual-budget timing.

Innovation CoE. Best for standards, methods, portfolio visibility, and capability building. It breaks the moment a governance team is told to ship product itself. Linda Hill put the risk directly:

"When you tell people the innovators are in that center of excellence and they're the executors, you get a problem."

— Linda A. Hill, Linda Hill on the ABCs of Innovation

Embedded Teams. Best for product-adjacent innovation inside a business unit that already owns customers, budget, and adoption. They break when autonomy exists in name only or when the company expects cross-BU coordination.

Transformation Office. Best for finite, sponsor-backed enterprise change. It breaks when the office owns deadlines but not the teams, systems, or P&L needed to execute them.

Corporate VC. Best for option value and external learning. It breaks when executives expect it to behave like an internal product team.

Open Innovation. Not a department. It works best as a company-wide layer that practices continuous foresight, taking weak external signals and turning them into pilots the business can actually test. Once the partnership starts, the model breaks if nobody owns the handoff.

What does the data show about innovation structure outcomes?

Outcome data helps when it clarifies which bottleneck a structure solves, not when it pretends every model should be judged by the same scoreboard. Offices, labs, venture studios, and CVC units optimize for different outputs, so the useful comparison is mandate fit plus downstream adoption.

Wellspring's 2022 analysis of corporate innovation offices found that 38% of the companies that outpaced peers on revenue growth had a CINO-led innovation office, versus 14% of other companies.

BCG 2024: Only 3% of surveyed companies qualified as innovation ready even though 83% ranked innovation among their top three priorities.

VCI / GCV 2024-2025: About 40% of CVC units do not survive past year three, which makes sponsor durability part of the structural design problem, not just an execution problem.

Taken as a whole, the evidence shows that a clear mandate helps, but it does not fix adoption. So a centralized office does not prove a centralized lab will ship.

How did Amazon build embedded innovation without calling it that?

Amazon is the clearest example of embedded innovation working because the team design bundled authority, customer ownership, and execution inside small units. The lesson is not simply to stay small. It is to reduce coordination overhead while keeping enough shared rules to stop local autonomy from fragmenting, a lesson that matters again when teams move from invention to ownership.

In 2022, Daniel Slater described Amazon's single-threaded ownership model as a team focused on one customer set and one service end to end. Carletta Ooton's 2018 explanation of the two-pizza team logic remains useful because it ties team size to communication overhead, not fashion.

That is why Andy Troutman could describe Amazon as a federation of startups instead of a single entity.

The 2000s two-pizza model only worked because later shared API rules kept local autonomy from turning into sprawl. Give the team real ownership, then add only the minimum shared rules needed to keep neighboring teams from colliding.

What does your reporting line reveal about what the organization actually expects?

A reporting line is not just admin plumbing. It is the cleanest signal of what the organization expects innovation to do. The boss who protects the team usually reveals whether the function is meant to govern, build, invest, or persuade across business units.

CEO or strategy reporting usually means portfolio visibility and cross-BU reach. It fits venture studios, labs, and enterprise CoEs, but it is fragile if one sponsor is doing all the carrying.

CTO or R&D reporting usually means technical depth. It is a strong home for labs, CoEs, and open innovation layers, but it often strands work upstream from adoption.

Business-unit reporting usually means near-term product or operational change. That is where embedded teams work best because the team sits close to customers and adoption authority.

CFO or transformation-sponsor reporting usually means program control. It can work for a time-bounded transformation office, but it often leaves the office accountable for change it cannot directly execute.

Which structures can run which kinds of work, and which ones cannot?

Structure should follow the kind of work that must happen, the cadence that work needs, and the number of handoffs the organization can tolerate. A model that fits Horizon 3 exploration often fails at Horizon 1 shipping, and the reverse failure is just as common.

Embedded teams are strongest at Horizon 1 and close-to-core Horizon 2 work because the same unit can test, decide, and ship. Innovation labs and venture studios fit protected Horizon 2 and Horizon 3 work.

These are coordination structures, not product-delivery structures. Innovation CoEs and transformation offices work best when the problem is shared method, portfolio visibility, or sequencing across teams. They struggle when what the business actually needs is end-to-end product ownership.

Corporate VC and open innovation are external-facing structures. They buy access, options, and partnerships. They do not remove the need for an internal owner who can absorb the result. A Corporate VC can buy options on the future. It cannot substitute for a team that has to ship next quarter.

What happens when the executive sponsor leaves?

Leadership turnover is the stress test most innovation structures eventually fail. When a sponsor leaves, the surviving question is whether the model still has authority, budget, and a visible route from experiment to owned execution, or whether the whole function was really one champion's project.

"CVC fails not because it does not work, but because it is rarely architected to survive governance transitions."

— VCI Institute, The CVC Reckoning: Munich Re Ventures Shutdown Exposes the Structural Fragility of Corporate Innovation

In January 2025, VCI Institute argued that CVC fails when it is not designed to survive governance transitions. Global Corporate Venturing reported in late 2024 that SAP.iO, Verizon Ventures, and other units still closed despite visible portfolio activity.

Transformation offices get squeezed differently. Executive Connections and IMD both describe the same pattern: IT often owns infrastructure, business units own products, and the office in the middle owns neither. That is the innovation sandwich, the squeeze between IT control and BU control that leaves the office accountable without giving it much direct power. Judy Estrin's warning is the practical version: once the leadership function turns off, the innovation function does too.

What do people get wrong about innovation team structure?

The biggest mistake is confusing labels, roles, and structures. A lab, CoE, venture arm, or open-innovation program can carry the wrong name and still behave like something else in practice. The job is to diagnose the operating model, not trust the label on the org chart or turn the whole design problem into a mind mapping exercise.

• A lab and a CoE are the same thing. They are not. A lab protects exploratory work. A CoE standardizes methods and capability across other teams.
• Corporate VC is just another innovation team. It is an investment structure, not a delivery structure. Patent output and strategic option value do not equal internal shipping capacity.
• Open innovation can stand alone. The external pipeline only matters if there is an internal owner for evaluation, market validation, and handoff.
• Every hackathon is theater. Robyn Bolton's 2025 defense is a fair correction. Rituals can build shared language, but they still fail as a substitute for adoption power, which is why wisdom of the crowd is not a reporting line.

Teams waste time here. They debate roles before they name the bottleneck. Headcount feels real, but structure decides whether those people can do anything useful.

When do the standard models break down?

Every model breaks when the mandate, authority, and operating conditions stop matching. The useful edge-case question is not whether a structure is good in general. It is what fails first when the team is asked to do work the model was never designed to carry.

Embedded teams break when the business unit treats them like a service bench rather than a product-owning team. Bessem Ayari's warning on culture, distance, and trust captures the human side of the same failure.

Open innovation breaks when partnerships arrive without an internal owner for invention, integration, or commercial handoff. Venture studios break when leadership wants startup upside without venture-style time horizons.

The centralized-lab exception is time-bound and task-bound: Amazon built Lab126 in the 2000s for hardware work that rewarded concentrated specialist capability.

Astro Teller made the case for X in June 2025 with a blunt output test:

"Google Brain alone has produced enough value to more than cover the total costs of X."

— Astro Teller, X Development, X Development profile

Pisano's 2010 analysis of the decline of the corporate lab still matters because it showed that the centralized model broke when organizational form stopped matching the science and commercialization task.

How do you choose a structure your organization can actually execute?

Choose the smallest structure that can still do the job. Every extra reporting line adds another place a pilot can stall, another approval path to manage, and another sponsor who can quietly kill momentum. Structure should match mandate, coordination cost, and time horizon before it matches prestige.

Start with a small lab or venture studio if the mandate is cross-BU exploration. A CoE fits standards and coaching. For product change inside an operating unit, embed the work there. Use Corporate VC or open innovation for external options only when an internal owner already exists and the team can move from a rough thesis to market validation without adding another orphaned handoff.

1. Name the mandate before the model

Define whether the job is mainly governance, venture building, or product change. Linda Hill's architect, bridger, catalyst framing helps because it separates leadership roles from delivery roles.

2. Match size to coordination cost

If the organization is small or politically centralized, a full embedded model is often theater turned inside out: lots of local titles, no durable authority. Embedded teams become viable when local units can actually own budgets and roadmaps.

3. Check maturity and time horizon

McKinsey's 2025 committed innovators research points to a mixed model with centralized scale and localized control. The Three Horizons separation rule lands in much the same place: keep Horizon 1 close to the business, while Horizon 3 usually needs insulation, patient capital, or both.

4. Choose the tradeoff you can defend in year three

If the structure needs one heroic sponsor forever, it is fragile. If it cannot explain how a pilot becomes an owned operating asset, it is fragile.

How do you know if your innovation structure is working?

A working innovation structure produces evidence that survives beyond activity counts. It creates a repeatable path from exploration to owned execution, so the test is whether managers can point to shipped outcomes, explain adoption without heroics, and keep the system alive when sponsorship changes.

Three checks matter most. First, the team should be able to point to shipped products, implemented process change, or commercially live partnerships, not just experiments. Second, can someone in the line organization explain how work gets adopted without inventing a hero story every time? Third, does the structure keep working when sponsorship changes?

Nappi and colleagues made the point in 2021. When a dashboard still leads with events, idea counts, patents filed, or press coverage, it usually shows a structure built to look orderly rather than help managers act. Adoption, learning rate, and innovation feedback loops that transfer work into the business point in a better direction.

FAQ

What's the difference between an innovation lab and an innovation CoE?

An innovation lab protects exploratory work that the core business cannot absorb yet, while an innovation CoE standardizes methods, governance, and capability across other teams. The confusion matters because companies that mix them up end up asking a standards function to deliver products or a lab to govern everyone else.

Should innovation report to the CEO or a business unit head?

Report to the CEO or strategy lead when the mandate is enterprise-wide portfolio change, venture building, or cross-BU coordination, because that work needs air cover across units. Report to a business unit head when the job is to improve an existing product or operating model, because adoption authority matters more than symbolic reach.

How do embedded innovation teams work and where do they fail?

Embedded teams work when they sit inside a domain that already owns customer relationships, budget, and a clear path to adoption. They fail once they become service staff for other leaders or when the organization expects enterprise-wide change without adding the shared governance that local teams alone cannot provide.

When does a centralized innovation lab make sense?

A centralized innovation lab makes sense when the work sits far from the core, needs insulation from quarterly targets, or depends on specialist capability that would be wasteful to duplicate in every unit. Hardware programs and moonshot-style bets are the strongest exceptions because concentrated expertise and patient timelines matter more than local adoption speed, especially when the work resembles long-horizon disruptive innovation rather than next-quarter delivery.

What does a Corporate VC model actually do for an enterprise?

A Corporate VC unit invests in external startups to buy learning, optionality, and strategic signal. That can sharpen external sensing and access to new markets, but it does not replace an internal product team or venture studio because investment exposure is not the same thing as owning execution inside the company.

How do you choose between federated and centralized innovation?

Choose federated or embedded structures when the work must ship close to local products, customers, or operating units that already own adoption. Choose centralized layers when the main need is common standards, portfolio visibility, shared infrastructure, or shared external access that multiple units would otherwise duplicate badly.

How big should an innovation team be for a company of our size?

There is no universal headcount rule that travels well across mandates, because team size follows the work, the adoption burden, and the number of units that must coordinate. Size the structure to the coordination problem rather than to prestige, and add headcount only when the current model cannot protect the work or move it forward.