---
lang: en
template: insight-detail
---

# **Management without self-interest**
> Jadey Insight | June 12, 2026 | Jörg Hubacher

Transformation capability as a problem of the operating model: compliance logic, outcome ownership and the question of coordination without positional self-interest.

This article is a personal contribution to the debate on the development of organisations, not an original empirical study.

Note: This article formulates a testable thesis on organisational transformation capability. Documented organisational research, plausible mechanisms and a further, empirically untested design thesis are deliberately kept separate. It is not product copy.

## Abstract

According to recurring survey data, especially McKinsey surveys, the success rate of large transformation programmes is often around one third. These figures are survey evidence, not a hard empirical constant; newer data for specific types of initiatives report improvements. What still requires explanation is the persistence of certain failure patterns over decades. This article argues that this persistence is not primarily an execution problem that can be solved through better change management, but a structural feature of the prevailing operating model.

Three mechanisms are distinguished. First, compliance logic shifts the evaluation standard from outcomes to demonstrable rule conformity, because conformity is individually safer than responsibility. Second, the coordination layer of organisations is structurally interest-bound: decisions about budgets, roles and priorities are also decisions about the position of those who decide. Third, this order stabilises itself because every transformation has to be implemented and reported by the layer whose status it changes.

From this, the article derives a design thesis: transformation capability requires a coordination and allocation instance that has no career, budget or status interest of its own in the outcome of its decisions, here called coordination without positional self-interest. According to the current state of organisational research, human hierarchies cannot systematically guarantee this property. Agentic AI systems could exhibit it, though not automatically, but only under architectural and governance conditions that the article names alongside the objections to the thesis. Outcome ownership, objective setting and liability remain with humans in this model.

Keywords: transformation capability; operating model; compliance; outcome ownership; principal-agent theory; organisational politics; bureaucracy; change management; agentic AI; algorithmic management

## Problem statement: a failure rate that barely moves

Few popular management diagnoses have been repeated as consistently as the low success rate of organisational transformations. Empirically, however, this finding is heterogeneous and strongly shaped by survey data. Kotter formulated the influential diagnosis in 1995 that the majority of change initiatives fail to achieve their objectives (Kotter, 1995). McKinsey surveys over roughly 15 years and several thousand respondents repeatedly indicate a success rate of about 30 percent; the authors themselves note that this value has practically not changed across survey waves (McKinsey, 2021). In large-scale reorganisations, only 21 percent of directly involved executives rated the result as successful in 2014 (McKinsey, 2014).

These figures need methodological qualification. Hughes showed that the widely cited 70 percent failure formula has a weaker empirical basis than its citation frequency suggests: it lacks a valid and reliable empirical foundation and rests mainly on self-reports, inconsistent definitions of success and consulting data, not on controlled longitudinal studies (Hughes, 2011). Cândido and Santos reach a similar conclusion in a systematic literature review: the circulating failure rates of strategic initiatives are largely based on outdated, fragmented or weak evidence; the actual rate remains open (Cândido and Santos, 2015). This critique is justified.

The evidence also does not show a uniform picture. Recurring surveys on large transformations report low success rates over longer periods; at the same time, newer data for specific initiative types point to improvements. For operating-model redesigns, McKinsey reported in 2025 that 79 percent of initiatives were completed and 63 percent achieved most objectives and improved performance, a clear increase compared with the 21 percent from 2014, although only 24 percent count as highly successful and a substantial share of companies still fails to capture the full value potential (McKinsey, 2025).

The issue requiring explanation is therefore not a universally constant failure rate, but the persistence of certain failure patterns over three decades of method development: coordination failure between units, distorted reporting paths, blocked reallocations and diffused outcome ownership. These patterns recur in the surveys and studies cited here, including where aggregate success rates improve. When the same mechanisms are named as main causes of failure over decades, it is not plausible to assume that companies are merely repeating the same craft errors. It is more plausible that the cause does not lie in execution, but in the operating model itself: in the way organisations distribute responsibility, information and decision rights.

### Core thesis

In short: transformation fails less because of resistance from individual people than because it has to be implemented through a coordination layer with three structural properties: it is measured by conformity instead of outcomes, it decides in matters that affect itself, and it controls the information about its own condition. As long as this layer consists of actors with self-interests, the persistence of the failure patterns described here is not an execution error, but an equilibrium state.

## Conceptual clarification: compliance, outcome ownership, operating model

Three terms carry the argument and are used here in a specific sense.

Compliance in this article does not refer to the legally required observance of laws, whose necessity is not in question. The point is compliance logic as an organisational principle: the shift of the evaluation standard from the question of whether an appropriate outcome was achieved to the question of whether the prescribed process was demonstrably followed. Organisational sociology has described this mechanism since the 1970s. Meyer and Rowan showed that formal structures in organisations often have a ceremonial function: they create legitimacy toward the environment and are decoupled from the actual work level (Meyer and Rowan, 1977). Power continued this diagnosis under the term audit society: auditability becomes a production goal in its own right; organisations spend growing shares of their resources signalling control without this necessarily providing evidence of performance (Power, 1997). DiMaggio and Powell added the diffusion mechanism: structures are copied between organisations because deviation costs legitimacy, often regardless of whether their effectiveness is proven (DiMaggio and Powell, 1983).

Outcome ownership means that a result can be attributed to a person who actually had access to the means required to achieve it. Under compliance logic, this coupling tends to dissolve: responsibility is distributed across processes, committees, approval chains and matrix structures until no result can be attributed to any one person, while every involved person can demonstrate rule-conforming conduct. Sull, Homkes and Sull showed in a multi-year study of strategy execution that the limiting factor is less a lack of agreement with the strategy than coordination failure between units; the reliability of reciprocal commitments between departments is consistently low (Sull, Homkes and Sull, 2015). Beer and Eisenstat identified recurring barriers to strategy execution, including unclear priorities, ineffective leadership teams and broken vertical communication, factors that were known in the organisations studied but not openly addressable (Beer and Eisenstat, 2000).

Operating model, finally, does not mean the business model, but the architecture of coordination: who sets objectives, who allocates resources, who measures, who escalates and who decides on exceptions. The thesis of this article concerns that architecture, not individual processes, tools or culture programmes.

## Evidence: three mechanisms from organisational research

### Conformity is individually rational, responsibility is individually risky

From the perspective of an individual executive, compliance logic is not a malfunction, but a rational response to incentives. Whoever owns an outcome can fail; whoever follows a process is protected. The denser the network of policies, approvals and documentation duties, the more the individual optimum shifts from initiative to self-protection. Audit research describes the aggregated consequence as a relocation of organisational resources: a growing share of work serves the production of auditability rather than the production of performance (Power, 1997).

There are estimates of the magnitude of this effect. Hamel and Zanini put the cost of excess bureaucracy for the United States at more than three trillion US dollars in lost economic output per year, and around nine trillion for OECD countries; large organisations typically have eight or more management layers (Hamel and Zanini, 2016; Hamel and Zanini, 2020). These values rest on contestable assumptions and should be read as an order of magnitude, not as a measurement. They do, however, quantify a qualitatively well-documented finding: a substantial part of work in large organisations coordinates, controls and documents other work.

### The coordination layer decides in matters that affect itself

The theoretical foundation is provided by principal-agent research: once ownership and management are separated, agency costs arise because agents systematically have room to pursue their own interests, while complete interest alignment and control can become so expensive that residual agency costs are rationally accepted (Jensen and Meckling, 1976). Niskanen modelled public administrations as budget-maximising bureaucracies, providing an influential public-choice explanation for why administrative units can tend to expand their resource base independently of the substantive mandate (Niskanen, 1971). Cyert and March described firms not as goal-rational units, but as coalitions of competing interest groups whose objectives are negotiation outcomes (Cyert and March, 1963). Pfeffer documented that resource allocation in organisations can follow power, coalition and influence logics to a considerable degree rather than pure substantive logic (Pfeffer, 1992).

For transformation, this finding is central, because transformation is by definition redistribution: of budgets, positions, responsibilities and status. Each of these redistributions has to be decided and implemented by decision-makers whose own position is affected. Christensen showed that established companies miss disruptive developments not primarily because of a lack of information, but because their internal resource allocation processes rationally favour existing customers, margins and therefore existing internal structures (Christensen, 1997). March framed the same relationship more generally: organisations systematically displace exploration in favour of exploitation because its returns are closer in time, more certain and attributable to current actors (March, 1991).

### The order stabilises itself

Hannan and Freeman described structural inertia as a selection feature: organisations are selected for reliability and reproducibility, properties that make change harder (Hannan and Freeman, 1984). This is compounded by an information problem: the coordination layer that is supposed to be changed is also the layer that reports on the progress of the change. Status reports and success messages pass through the interests they are supposed to evaluate. The unaddressability of central barriers described by Beer and Eisenstat is the consequence of this constellation (Beer and Eisenstat, 2000).

Change management in the common sense, communication, participation and qualification, addresses attitudes and capabilities. It does not address the incentive and information structure. The success rate that has barely moved over decades is consistent with this gap.

This interim conclusion requires no reference to artificial intelligence: the limited transformation capability of companies is well explained by five decades of organisational research. It follows from compliance incentives, the interest-bound nature of decision-makers and the self-reference of reporting paths. A durable increase in transformation capability would have to address these three properties architecturally.

## An alternative operating model: the design thesis of coordination without positional self-interest

The following section formulates the hypothetical part of the argument. It should be read as a design thesis, not as a proven result.

### The reference case: decentralisation without a technological foundation

The attempt to solve the problem described here without new technology is documented. Under Zhang Ruimin, Haier divided the corporation into several thousand micro-enterprises coordinated through internal market mechanisms rather than hierarchy; Hamel and Zanini describe this RenDanHeYi model as one of the most prominent and radical documented examples of replacing bureaucratic hierarchy with market-like coordination between micro-enterprises (Hamel and Zanini, 2018). The finding is ambivalent. The model works, but a broad adoption by other large companies is not documented in the management literature. The obvious explanation is the structural problem described above in sharpened form: the dismantling of the coordination layer has to be decided and carried out by the coordination layer. At Haier, this required an exceptional power position of corporate leadership that had grown over decades. Such a constellation cannot be systematically replicated.

### Properties of an agentic coordination layer

Agentic AI systems change the starting point here. For the first time, there is an instance that can perform coordination work, allocate resources, track priorities, measure goal achievement, escalate deviations and document decisions, while potentially exhibiting three properties that human coordination layers, according to the research cited here, do not exhibit:

- No positional interests: An agentic system has no career, budget or status interests of its own. It loses nothing if the area it coordinates shrinks, and gains nothing if it grows. Budget and resource maximisation in Niskanen's sense disappears as its own motive, though not as a motive of those who configure the system; the article returns to that point in the objections.
- No self-reference of reporting paths: Allocations, evaluations and escalations can, with the right architecture, be versioned, logged and reviewed by third parties. This auditability is not automatic in agentic AI, but a governance and system-design requirement; in practice, black-box components, incomplete logs or unversioned context changes can undermine it. Where it is established, however, the reporting instance is not simultaneously the subject and filter of the evaluation.
- No attachment to its own structure: A system executing a reorganisation has no interest in preserving the structure being replaced.

The thesis is explicitly not that AI should run companies. Objective setting, risk trade-offs, liability and outcome ownership remain with humans. What is transferred is the interest-bound middle: the translation of objectives into allocation, coordination between units, measurement and follow-up. Without positional self-interest means precisely this, the absence of the system's own career, status and budget interests, not freedom from interests in every sense. Whether the decisions of such a system are actually more neutral, transparent and auditable than human coordination depends on objective definition, data access, logging, role model and governance. In this division of labour, human outcome ownership is not reduced, but made more precise, because it can no longer be diffused across committees and approval chains.

### Compliance as boundary condition instead of work layer

In such an operating model, compliance also changes character. Rule conformity is not created through downstream documentation, but built into execution as a boundary condition: a process step that requires approval cannot proceed without that approval, and its process log is also its evidence. This idea is not new: workflow systems, ERP approvals, business-process management and policy-as-code approaches have implemented parts of it for years. Agentic AI could extend these forms of executable compliance by dynamically coordinating not only individual control points, but also objective translation, prioritisation, escalation and follow-up. The verification work described by Power would then move into execution itself and no longer require a separate work layer. This removes the structural possibility of using process conformity as a substitute for outcome ownership.

## Objections and alternative explanations

The design thesis is strongly formulated and has to face the most serious objections.

The first objection concerns the premise of freedom from interests. AI systems are not interest-free; at most, they are interest-neutral to the degree that their objective functions are. Every agentic system optimises for goals that humans define, using data that humans select, under constraints that humans set. The principal-agent problem therefore does not disappear, but shifts to the configuration level: to the question of who sets the objectives of the coordination layer and controls its parameters. This objection is valid and explains why this article speaks of an operating model rather than a tool. Without governance of objective setting, an agentic coordination layer reproduces the interests of those who configure it. The difference from the status quo nevertheless remains and can be named precisely: an objective function is explicit, versioned and auditable; the interest position of a management layer is not.

The second objection comes from research on algorithmic management. Kellogg, Valentine and Christin show in their review that algorithmic control in practice often does not operate as neutral coordination, but as expanded control, more comprehensive, more granular and less transparent than human supervision (Kellogg, Valentine and Christin, 2020). Findings from the platform economy support this assessment. The objection applies to every implementation that equates coordination with surveillance. It does not necessarily apply to an architecture in which the agentic layer is the most fully inspectable part of the organisation and decisions on exceptions, hardship cases and objectives remain with humans. Whether this separation can be kept stable in practice is an open empirical question.

The third objection concerns responsibility in the legal sense. An AI system cannot bear liability and cannot be held accountable. That is correct and is understood here not as a weakness, but as the reason for the proposed division of labour: precisely because the coordination instance cannot bear liability, it must not carry outcome ownership, but only execute coordination whose results are clearly attributable to persons. The requirements of the European AI Act for high-risk systems, human oversight, logging and transparency, are compatible with this line (European Union, 2024). Whether an agentic coordination system is to be classified as a high-risk system under the regulation in an individual case depends on the specific use; such a classification is especially likely when the system evaluates employees, assigns tasks, measures performance or prepares decisions with employment effects.

The fourth objection is the most fundamental: the self-stabilisation problem described here also applies to the introduction of the proposed model. An agentic coordination layer would also have to be approved by the existing coordination layer. This objection speaks against expectations of fast, broad-based conversion. A displacement path through the edges is more plausible: newly founded units, subsidiaries and mid-sized companies without an entrenched coordination layer that build agentically organised operations and create adaptation pressure on established structures through cost and speed differences. This path structurally corresponds to the disruption pattern described by Christensen, applied to the operating model itself.

Finally, an alternative explanation of the initial findings has to be considered: the low success rates could simply reflect the difficulty of the task, not a structural defect. Transformations are complex, multi-year initiatives under uncertainty; a success rate of one third would then not be an equilibrium finding, but a property of the task class. The improvements reported for operating-model redesigns (McKinsey, 2025) could also be read this way: as a learning curve for a difficult but learnable task. Against a complete explanation through task complexity stands the fact that the research cited here locates the failure mechanisms not primarily in complexity, but repeatedly in allocation politics, coordination failure and reporting distortion, that is, in variables of the operating model. The alternative explanation cannot be fully excluded with the available data, however.

## Consequences for organisational design, governance and measurement

### Organisational design: concentrate responsibility, externalise coordination

If the analysis is correct, the first consequence is a design principle for responsibility: outcome ownership belongs undivided to individual persons who actually have access to the required means. Committee decisions without personal attribution are, by this standard, not a form of responsibility but its dissolution. Coordination, allocation, measurement and follow-up, by contrast, belong in an agentic layer whose objective functions, rules and protocols are fully inspectable.

### Governance: address the shifted principal-agent problem

Objective setting for the agentic layer requires its own governance structure that explicitly addresses the principal-agent problem shifted to the configuration level: documented objective definitions, versioned changes, separated roles for objective setting and benefit, and external auditability. Compliance should be integrated into processes as an executable boundary condition rather than organised as a downstream documentation function.

### Measurement: indicators of transformation capability

Such a model becomes testable through metrics on which the existing model, according to the evidence cited here, is systematically weak:

- Share of working time spent on coordination and evidence production relative to total effort.
- Time from an objective decision to the actual reallocation of resources.
- Share of decisions with one clearly named outcome owner.
- Ratio of implemented to decided structural changes.
- Deviation between internal reported status and externally validated status of ongoing initiatives.

## Conclusion

The transformation debate mostly treats the failure of change as an execution problem: insufficient communication, lack of participation, weak sponsorship. Organisational research from the past five decades suggests a structural reading. Compliance logic rewards self-protection instead of outcomes, decision-makers decide in matters that affect themselves, and reporting paths run through the interests they are supposed to evaluate. Under these conditions, the persistence of coordination failure, distorted reports and blocked reallocations, which continue even where aggregate success rates improve, is not a failure of the people involved, but an equilibrium state of the operating model.

This equilibrium does not shift through appeals, but through architecture. The design thesis formulated here is that agentic AI could for the first time allow the construction of a coordination layer without its own positional, career, status or budget interests in the outcome of its allocations, and that outcome ownership would not disappear as a result, but become clearly attributable to persons again.

The thesis is testable. If it is wrong, organisations with an agentic coordination layer should show the same patterns as today's organisations: diffused responsibility, growing evidence work, blocked reallocations, distorted reports. If it is correct, agentically organised units will in the coming years reallocate measurably faster, operate with lower coordination effort and implement decided structural changes at higher rates. The adaptation pressure on the classical operating model would then arise not from theory, but from cost comparison.

What is open is not whether today's coordination layer has self-interests; that is well documented. What is open is whether a technically constructed coordination layer, under real governance conditions, actually works with less positional self-interest than the human layer it would replace. This question deserves empirical research before it is decided ideologically, in either direction.

## Source review for this version

The central statements were checked against primary sources or established bibliographic records. The sources do different things: peer-reviewed research and monographs carry the theoretical basis (agency theory, sociological institutionalism, inertia and politics research, algorithmic management). Consulting and survey data document only recurring low success rates, not their causes; they do not replace controlled studies. The 2025 McKinsey counter-finding on operating-model redesigns is deliberately included to avoid selective data choice. Hamel and Zanini's estimates of bureaucracy costs are used as an order of magnitude, not as a measurement. The design thesis of the article is marked as such and does not claim empirical proof.

## Literature and sources

Beer, M. and Eisenstat, R. A. (2000): The Silent Killers of Strategy Implementation and Learning. MIT Sloan Management Review, 41(4), 29-40.

Cândido, C. J. F. and Santos, S. P. (2015): Strategy Implementation: What Is the Failure Rate? Journal of Management & Organization, 21(2), 237-262. DOI: 10.1017/jmo.2014.77.

Christensen, C. M. (1997): The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail. Boston: Harvard Business School Press.

Cyert, R. M. and March, J. G. (1963): A Behavioral Theory of the Firm. Englewood Cliffs: Prentice-Hall.

DiMaggio, P. J. and Powell, W. W. (1983): The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields. American Sociological Review, 48(2), 147-160. DOI: 10.2307/2095101.

European Union (2024): Regulation (EU) 2024/1689 (AI Act). Official Journal of the European Union, July 12, 2024. Online: [eur-lex.europa.eu/eli/reg/2024/1689/oj](https://eur-lex.europa.eu/eli/reg/2024/1689/oj) (accessed June 12, 2026).

Hamel, G. and Zanini, M. (2016): Excess Management Is Costing the U.S. $3 Trillion Per Year. Harvard Business Review (online), September 5, 2016. Online: [hbr.org/2016/09/excess-management-is-costing-the-us-3-trillion-per-year](https://hbr.org/2016/09/excess-management-is-costing-the-us-3-trillion-per-year) (accessed June 12, 2026).

Hamel, G. and Zanini, M. (2018): The End of Bureaucracy. Harvard Business Review, 96(6), 50-59. Online: [hbr.org/2018/11/the-end-of-bureaucracy](https://hbr.org/2018/11/the-end-of-bureaucracy) (accessed June 12, 2026).

Hamel, G. and Zanini, M. (2020): Humanocracy: Creating Organizations as Amazing as the People Inside Them. Boston: Harvard Business Review Press.

Hannan, M. T. and Freeman, J. (1984): Structural Inertia and Organizational Change. American Sociological Review, 49(2), 149-164. DOI: 10.2307/2095567.

Hughes, M. (2011): Do 70 Per Cent of All Organizational Change Initiatives Really Fail? Journal of Change Management, 11(4), 451-464. DOI: 10.1080/14697017.2011.630506.

Jensen, M. C. and Meckling, W. H. (1976): Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. Journal of Financial Economics, 3(4), 305-360. DOI: 10.1016/0304-405X(76)90026-X.

Kellogg, K. C., Valentine, M. A. and Christin, A. (2020): Algorithms at Work: The New Contested Terrain of Control. Academy of Management Annals, 14(1), 366-410. DOI: 10.5465/annals.2018.0174.

Kotter, J. P. (1995): Leading Change: Why Transformation Efforts Fail. Harvard Business Review, 73(2), 59-67.

March, J. G. (1991): Exploration and Exploitation in Organizational Learning. Organization Science, 2(1), 71-87. DOI: 10.1287/orsc.2.1.71.

McKinsey & Company (2014): The Secrets of Successful Organizational Redesigns: McKinsey Global Survey Results. Online: [mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-secrets-of-successful-organizational-redesigns-mckinsey-global-survey-results](https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-secrets-of-successful-organizational-redesigns-mckinsey-global-survey-results) (accessed June 12, 2026).

McKinsey & Company (2021): Losing from Day One: Why Even Successful Transformations Fall Short. Online: [mckinsey.com/capabilities/people-and-organizational-performance/our-insights/successful-transformations](https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/successful-transformations) (accessed June 12, 2026).

McKinsey & Company (2025): The New Rules for Getting Your Operating Model Redesign Right. Online: [mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-new-rules-for-getting-your-operating-model-redesign-right](https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-new-rules-for-getting-your-operating-model-redesign-right) (accessed June 12, 2026).

Meyer, J. W. and Rowan, B. (1977): Institutionalized Organizations: Formal Structure as Myth and Ceremony. American Journal of Sociology, 83(2), 340-363. DOI: 10.1086/226550.

Niskanen, W. A. (1971): Bureaucracy and Representative Government. Chicago: Aldine-Atherton.

Pfeffer, J. (1992): Managing with Power: Politics and Influence in Organizations. Boston: Harvard Business School Press.

Power, M. (1997): The Audit Society: Rituals of Verification. Oxford: Oxford University Press.

Sull, D., Homkes, R. and Sull, C. (2015): Why Strategy Execution Unravels - and What to Do About It. Harvard Business Review, 93(3), 57-66. Online: [hbr.org/2015/03/why-strategy-execution-unravelsand-what-to-do-about-it](https://hbr.org/2015/03/why-strategy-execution-unravelsand-what-to-do-about-it) (accessed June 12, 2026).