The promise of a digital twin of an organization rests on an appealing assumption: The more completely a business can observe and model itself through data, the better it can understand the present while preparing for its future. Yet an organization is not a machine, but a social system whose members interpret being observed and change their behavior in response. As I will argue in this article, this social circumstance creates a fidelity paradox.

The pervasive observation required for a perfect digital twin would alter the organization of humans being represented. The model would become more complete as a data system, but less faithful to the social reality it was built to capture. The useful alternative is not to abandon organizational twins, but to design them as selective models with explicit boundaries.

The desire to control an organization’s present and future

The idea of a digital twin of an organization is enticing. It promises corporate leaders a unified, virtual live view of operations, utilization, productivity, efficiency, employee happiness and other indicators of organizational health (Parmar et al., 2020). At first glance, the limits of modeling are assumed to be set primarily by the availability and quality of appropriate data. Yet the literature identifies deeper constraints, including model quality, observability, synchronization and the integration of different representations (Jones et al., 2020).

The greater fantasy is that a comprehensive and integrated digital representation would allow management to know and control what is happening. As popular management reasoning goes, anything that can be observed, measured or interpreted could theoretically be modified toward a desirable outcome (Franco-Santos et al., 2012). In the strongest version of the digital-twin vision, interventions made through the model would produce predictable changes in the organization itself (Lyytinen et al., 2024).

A perfect digital twin could also go beyond dashboards that monitor the status quo. It could create a virtual sandbox for simulating future states and testing the impact of potential interventions. A business could assess its past and present while anticipating risks, building scenarios and planning for the future. That makes the concept attractive on both tactical and strategic levels. Commercial process-mining and business-process-management providers are already moving their tools in this direction, even though a complete digital twin of an organization remains more vision than operational reality (van der Aalst et al., 2021).

Organizations aren’t machines but social systems of humans

Unlike physical turbines or automated assembly lines that can be digitally twinned with the right models and sensors (Dai et al., 2026), organizations involve human beings with minds of their own. Machine tools neither care nor interpret being monitored, but employees generally do. They interpret rules in their own ways, follow disparate agendas, negotiate meanings and react strategically to measurement.

Business organizations formulate tasks, goals, roles, processes, hierarchies and rules by which people are assigned activities in an outcome-oriented way. Organizational sociology describes these as “highly formalized social structures” (Weick, 1995), defined along the lines of division of labor, hierarchy of authority, and formalized rules and procedures (Hatch & Cunliffe, 2006).

Formalized structures are comparatively easy to digitally twin because they are made explicit and binding. They make organized human action more determinable, traceable and repeatable (Luhmann, 1964). Their behavioral expectations are tied to formal membership roles, and employees agree to play along. Non-compliance can ultimately result in demotion or termination.

There’s an informal side to every organization that cannot be governed

Yet anyone who has worked in an organization knows that it rarely operates exactly as intended on paper. Not every aspect can or should be regulated. Employees participate in situations governed by rationales unspecified by the organization, known as the informal sphere (Luhmann, 1964). Here, they act as office friends, unofficial mentors, gossip buddies or informants.

The informal sphere is not simply what happens on the sidelines. Most companies are imperfectly designed, so workers create workarounds. As organizational ethnographer John Brewer (2006) points out:

“Workers often try to follow the formal rules but have to engage informally in practical reasoning and ad hoc practices to operationalize them when the formal rules are incapable of meeting the job at hand, such that fulfillment on the organization’s formal goals requires informal organizational rules, tacit knowledge and discretion. However, workers have to make it look as if the formal rules were followed as part of the organization’s coercive control, so engage in further ad hoc practices to ensure the paperwork conforms the procedures.” (p. 314)

The informal sphere therefore helps to keep the formal organization running. The two are recursive and essential to its daily operation. Twinning the informal sphere, however, creates an ethical and practical challenge.

What data would a “perfect” twin actually require?

Consider what a perfect digital twin would require. It would begin with managerial-operational layers that can be formalized along the production process. It would include formal social structures; resources and inputs; activities and processes; outputs, outcomes, performance and control; and strategy and planning:

• Division of labor: organization charts, business units, functions, roles and tasks
• Hierarchy of authority: chains of command, delegation, compensation, promotion, punishment and accountability
• Formalized rules and procedures: codes of conduct, policies, and work and communication processes
• Resources and inputs: materials, labor and working time; capital and budgets, tools, technology and infrastructure, data, information and knowledge; external services and suppliers
• Activities and processes: workflows, production and service activities; resource allocation; dependencies between functions; decision-making and coordination
• Outputs and outcomes: products, services, transactions, revenue and stakeholder outcomes
• Performance and control: capacity, utilization, productivity, cost, and efficiency; Quality, error rates, and goal attainment
• Strategy and planning: objectives, targets, investment and staffing plans; product roadmaps, risks and forecasts

These structural, capacity-related and processual building blocks are comparatively innocuous. They are standard, often quantifiable categories already digitized in mature organizations. Unsurprisingly, popular concepts of organizational digital twins focus on them (Parmar et al., 2020; van der Aalst et al., 2021).

What about the people who work in an organization?

The difficulty begins with data about people as social beings who interact and collaborate, and as individuals with minds of their own. Both dimensions affect how an organization operates in practice. They form part of an ethnographic framework I first developed in my business anthropological dissertation about AUDI AG (Mai, 2015) and can also be approached through organizational culture: the complex social formation and symbolic system of meanings largely shaped by people (Martin, 2002).

The social dimension encompasses how people relate to and work with each other:

• Social groups and structures, such as tenure-based allegiances and gatekeepers of internal knowledge
• Work-related collective identities, such as “us engineers versus those marketers”
• Informal hierarchies, including power structures erected by office politicians
• Informal working practices and workarounds that circumvent dysfunctional processes
• Daily interaction and communication between members of the organization

The psychological dimension concerns how actors think and perceive organizational reality. It includes, for instance:

• Values and beliefs
• Ingrained rules
• Patterns of thinking and self-perceptions

This dimension is harder to trace because it becomes visible only when people externalize their thinking through speech or writing.

Unlike the largely structured and quantifiable data in the managerial-operational layer, these dimensions can only be grasped through unstructured, qualitative data. To represent them, a twin would need to capture:

• Official communication in meetings, emails, calls and messages
• Informal conversations in corridors, cafeterias and after-work settings
• Organizational stories and anecdotes that indicate norms
• Mental frameworks that people use to make sense of organizational reality
• Tacit practices that people embrace to conduct their work
• Employee movement and interaction on company premises
• Private messages in which organizational matters are discussed.

Methodologically, this involves a high degree of observation via audio, video, textual documentation, and even semi-structured interviewing. Data completeness therefore implies pervasive observation of human activity across formal, informal and even private spheres. This is not entirely a surveillance pipe dream. Some companies already use AI to analyze sentiment in Slack messages, emails and Teams conversations to map organizational health and identify “toxic pockets” (Pham, 2026, February 23).

The fidelity paradox between complete representation and stylized performance

Pervasive observation does not merely create ethical and legal problems. It contaminates the model. Even if employees consented to having every movement, utterance and interaction logged and analyzed, they would become self-conscious.

Once people know that private conversations, workarounds and informal relationships are being captured, those practices change. They manage their visibility, sanitize communication and produce behavior expected to appear favorable to the measurement system. After all, management might sanction observed behavior it considers counterproductive, even when it was well intended.

This recalls Michel Foucault’s (1977) explanation of panopticism: prison inmates are permanently observable but do not know what the observer actually sees. The result is a constant, embodied sense of observation, anxiety and self-discipline. In contrast, business organizations depend on voluntary, partial membership and incentives rather than imprisonment and punishment, but such surveillance would still signal an absence of trust and diminished autonomy. Put simply, no one wants to work in a surveillance hellhole.

When monitoring creates stylized performance, the twin increasingly models the organization’s response to surveillance rather than the organization that existed before it. Its pursuit of fidelity thus undermines its own accuracy and the organic workings of the business.

What we miss when leaving out the informal layer

Why not simply exclude coffee-machine conversations, private messages and other informal exchanges then? Because informal organizational life is not peripheral noise. Instead, it is often what allows the formal organization to function. People use tacit knowledge, discretion and ad hoc practices precisely because formal procedures cannot anticipate every situation.

Nor do all members view organizational reality in the same terms. Organizations are places of multivocality and dynamism (Strati, 2000). There is no single “true organizational state” waiting to be measured. An assembly-line worker may legitimately experience the same company differently from a CEO or mid-level manager. A perfect twin would have to account for multiple, simultaneous logics, interests and motivations. Otherwise, it would impose a dominant view that is not representative of how the organization actually works.

From perfect twins to useful twins

The fidelity paradox does not mean organizational digital twins are useless. It means a useful twin should have declared boundaries. Instead of claiming to represent everything, the model should specify the questions it is designed to answer, the phenomena it captures well, the dimensions it excludes and the uncertainties that remain.

A workforce-planning twin might model capabilities, workloads and demographic risks without claiming to model organizational culture. A process twin might identify bottlenecks without pretending that communication and movement metadata reveal employee sentiment or live interaction. The twin then becomes an instrument for selective inquiry rather than an all-knowing representation of the company. Its value lies not in being perfect, but in being explicit about what it can and cannot show.

References

Brewer, J. D. (2006). Ethnography. In C. Cassel & G. Symon (Eds.), Essential guide to qualitative methods in organizational research (pp. 312–322). London: Sage.

Dai, W., Wang, G., Shan, X., Li, D., & Zhou, S. (2026). Digital twin of machine tools: A review and perspective. The International Journal of Advanced Manufacturing Technology, 145, 54-72.

Foucault, M. (1977). Discipline and punish: The birth of the prison. New York City, NY: Vintage Books.

Franco-Santos, M., Lucianetti, L., & Bourne, M. (2012). Contemporary performance measurement systems: A review of their consequences and a framework for research. Management Accounting Research, 23(2), 79-119.

Hatch, M. J., & Cunliffe, A. L. (2006). Organization theory: Modern, symbolic, and postmodern perspectives (2nd ed.). Oxford: Oxford University Press.

Jones, D., Snider, C., Nassehi, A., Yon, J., & Hicks, B. (2020). Characterising the Digital Twin: A systematic literature review. CIRP Journal of Manufacturing Science and Technology, 29, 36-52.

Luhmann, Niklas (1964). Funktionen und Folgen formaler Organisation. Berlin: Duncker & Humblot.

Lyytinen, K., Weber, B., Becker, M. C., & Pentland, B. T. (2024). Digital twins of organization: Implications for organization design. Journal of Organization Design, 13, 77-93.

Mai, D. (2015). Organizational Cultures of Remembrance: Exploring the Relationships between Memory, Identity, and Image in an Automobile Company. Berlin/New York: De Gruyter.

Martin, J. (2002). Organizational culture: Mapping the terrain. Thousand Oaks, CA: Sage.
Parmar, R., Leiponen, A., & Thomas, L. D. W. (2020). Building an organizational digital twin. Business Horizons, 63(6), 725-736.

Pham, W. (2026, February 23). AI tools for detecting toxic workplace behavior. Gaslighting Check. Retrieved from https://www.gaslightingcheck.com/blog/ai-tools-detecting-toxic-workplace-behavior.

Strati, A. (2000). Theory and method in organization studies: Paradigms and choices. London: Sage.

van der Aalst, W. M. P., Hinz, O., & Weinhardt, C. (2021). Resilient digital twins: Organizations need to prepare for the unexpected. Business & Information Systems Engineering, 63, 615–619.

Weick, K. E. (1995). Sensemaking in organizations. London: Sage.