← Back to Blog
31 August 2026

Dunbar's Number: What the Science Says and What Leaders Miss

Most leaders believe they can manage large teams simply by working harder, communicating more, or implementing better systems. Yet organizations consistently hit invisible walls where coordination breaks down, trust erodes, and performance suffers, regardless of how polished the strategy looks on paper.

The culprit is often something far more fundamental: human cognitive limits.

The dunbar number, a concept developed by British anthropologist Robin Dunbar in the 1990s, suggests that humans can maintain stable social relationships with roughly 150 people at any given time. This figure has since become one of the most cited, debated, and misunderstood ideas in organizational psychology. Leaders invoke it constantly, yet few understand what the research actually supports, or where it falls short.

In this analysis, we will cut through the popular oversimplifications and examine what the science genuinely says about social group limits. You will learn how these cognitive boundaries operate across different relationship layers, why most leadership teams apply this concept incorrectly, and what a more nuanced understanding means for how you structure teams, build culture, and scale your organization effectively.

The Science Behind the ~150 Figure

The Science Behind the ~150 Figure

Robin Dunbar's foundational insight did not arrive as a flash of intuition about human society. It emerged from a methodical comparative analysis published in 1992, in which Dunbar plotted neocortex ratios against observed group sizes across 36 primate genera. The neocortex ratio he used was specifically the volume of the neocortex divided by the volume of the rest of the brain, a measure chosen because it produced the strongest regression fit among several candidate variables. When he extrapolated that regression line to the neocortex ratio of Homo sapiens, the predicted maximum stable group size landed at approximately 148, a figure that popular discourse promptly rounded to 150. Critically, this was an extrapolation beyond the observed data range, not an interpolation within it, which means the prediction carries more uncertainty than the clean round number implies. Dunbar's own follow-up work acknowledged a 95% confidence interval spanning roughly 100 to 231, wide enough to accommodate meaningfully different social realities depending on where within that range a given human population sits.

The Cognitive Mechanism

The underlying logic is about processing capacity, not personality or culture. As a group grows, the cognitive load of tracking each individual's identity, status, and relationships with every other member increases non-linearly. The neocortex, the brain region most associated with higher-order cognition and language, appears to set a ceiling on how much of this relational bookkeeping can be sustained simultaneously. Dunbar tested ecological alternatives, including ranging area and diet composition, and found neither accounted for the brain-size variation across primates as convincingly as social group complexity did. The implication is striking: human intelligence may have expanded not primarily to solve physical problems but to manage an increasingly dense web of social relationships. Language, on this view, evolved partly as a more efficient substitute for the one-to-one grooming that other primates use to maintain bonds, allowing humans to tend multiple relationships in parallel through conversation.

Four Layers, Not One Ceiling

What often gets lost in popular retellings is that Dunbar never proposed a single number. He described four nested social circles: approximately 5 intimate relationships, 15 close friends and trusted collaborators, 50 meaningful contacts, and 150 stable casual connections. Each layer is roughly three times larger than the one inside it, and the frequency and emotional intensity of contact drops at each step outward. These are not sharp categorical boundaries; they are zones of graduated relational investment. The innermost circle of five represents the people you would call in a crisis. The 150-person layer is better understood as the ceiling for groups that cohere through mutual familiarity alone, without needing formal rules, hierarchies, or enforcement structures to hold them together. Academic scrutiny has intensified around the precise figures, with alternative phylogenetic analyses producing substantially different estimates, but the layered architecture of the model has proven more resilient than any single number within it.

From Academic Paper to Cultural Shorthand

The reason most readers arrive already familiar with "Dunbar's number" owes less to academic journals than to two landmark popular works. Malcolm Gladwell featured the concept in The Tipping Point, framing 150 as a practical threshold for organisational cohesion. Yuval Noah Harari returned to it in Sapiens, connecting it to the limits of pre-institutional human cooperation. This cultural penetration means intermediate readers typically need extension and application, not a basic introduction. The more productive analytical question is not whether 150 is precisely correct but what the layered structure beneath it reveals about how relational capacity is actually distributed across different levels of an organisation, and where structural design either honours or ignores those limits.

Historical and Operational Corroboration

Historical and Operational Corroboration

The case for a cognitive ceiling on stable social relationships does not rest on neuroscience alone. Across ethnographic records, military doctrine, ancient governance, and modern corporate engineering, the same threshold recurs with a consistency that demands explanation beyond statistical coincidence. Each domain arrived at roughly 150 through its own logic, its own pressures, and its own centuries of trial and error. The convergence is the real finding.

Ethnographic and Historical Evidence

Hunter-gatherer bands, Neolithic village settlements, 11th-century English parishes, and even the mundane practice of sending Christmas cards all cluster in the 100 to 200 range. Robin Dunbar himself noted that "the community level of organization turned out to be almost exactly 150" when he tested his regression-derived figure against anthropological data. Amish communities observe a similar upper boundary before schism or relocation occurs naturally. These are not controlled experiments; they are the accumulated outcomes of ordinary human social life, and they sit squarely within the 100 to 231 confidence interval the original 1992 paper produced. Importantly, critics who have challenged the statistical derivation of 150 from primate neurology have inadvertently strengthened the argument for operational corroboration: if the number cannot be cleanly derived from brain ratios alone, its persistent appearance in lived human organization becomes even more significant, not less.

Military Organization as Independent Cross-Validation

Perhaps the strongest independent cross-validation comes from military history. The Roman centuria comprised roughly 80 to 100 soldiers in active deployment, scaling to approximately 160 with support personnel, a structure that endured for centuries because it functioned under genuine battlefield stress. Modern military company units across NATO and other major armed forces converge near the same figure of 150 soldiers. Commanders did not read cognitive neuroscience; they learned through operational failure what size of unit a leader could know personally, trust implicitly, and coordinate without bureaucratic scaffolding. The fact that the Roman legion and the contemporary infantry company reached the same answer, separated by two millennia and radically different technologies, points to a constraint that is architectural rather than cultural.

An Ancient Governance Framework

The alignment extends further back still. In Exodus 18, Jethro counsels Moses to appoint leaders of thousands, hundreds, fifties, and tens, distributing the relational and judicial load that a single leader cannot sustainably carry alone. The nested groupings, thousands, hundreds, fifties, and tens, map with striking precision onto Dunbar's four social layers of 150, 50, 15, and 5. Jethro's advice was practical and urgent; Moses was exhausting himself trying to hold every relationship and decision personally. The prescribed remedy was structural: span limits at every level, with authority genuinely delegated rather than merely notionally shared. This is not a loose analogy. It is evidence that the relational constraints Dunbar formalized in 1992 were observed, diagnosed, and managed as documented in analyses of Dunbar's number by practitioners of governance thousands of years before the vocabulary of cognitive science existed.

Corporate Practice: Gore and Amazon

W. L. Gore and Associates, the company behind Gore-Tex, institutionalized the same threshold as explicit policy. Founder Bill Gore observed directly: "We found again and again that things get clumsy at a hundred and fifty." When any facility approaches that headcount, a new plant is built rather than the existing one expanded. The rationale is not sentimental; beyond 150, personal accountability erodes, informal coordination breaks down, and formal management layers multiply to compensate. Gore has sustained this practice across a global workforce of more than 9,500 employees, generating over three billion dollars in annual revenue, suggesting the constraint is commercially viable at scale, not merely idealistic.

Amazon's two-pizza team principle operationalizes the innermost Dunbar layers rather than the outer boundary. Teams of five to ten people remain small enough that every member holds a genuine, current mental model of every colleague's work, priorities, and reliability. Psychological safety research consistently supports this range as optimal for trust-based collaboration. The design choice is deliberate: coordination overhead, not capability, limits what larger teams can achieve.

What conventional explainers of Dunbar's number routinely miss is that the triangulation itself carries the argument. No single source, whether ancient scripture, Roman military doctrine, Hutterite farming practice, or Silicon Valley engineering policy, proves the cognitive limit independently. But the recurrence of the same threshold across domains that share nothing except the human beings within them points to a structural reality that organizational design cannot afford to ignore.

What the Critics Get Right

The corroboration from historical villages, military units, and Gore-Tex factories is compelling, but intellectual honesty demands equal attention to what the evidence does not support. Several substantive critiques have accumulated over the past three decades, and engaging with them directly strengthens rather than weakens the practical utility of the concept.

The Statistics Do Not Support a Single Number

The most consequential critique is statistical, and it is hiding in plain sight. Dunbar's own 1993 paper reported a 95% confidence interval for the predicted human group size running from approximately 100 to 231, a span wide enough to encompass a small department and a military company simultaneously. That interval is not a footnote; it is the data speaking. When popular coverage routinely presents 150 as a precise cognitive threshold, it misrepresents what the original regression actually demonstrated, which is a plausible central tendency within a broad statistical band, not a hard biological ceiling. A 2021 paper in Biology Letters by Lindenfors, Wartel, and Lind intensified this concern considerably. Using Bayesian and generalized least-squares phylogenetic methods, the Stockholm University researchers generated 95% confidence intervals stretching from 4 to 520 and 2 to 336 respectively, leading them to conclude that specifying any single number is, in their words, "futile." That conclusion deserves to be taken seriously, not dismissed as academic pedantry.

Extrapolation Is Not the Same as Prediction

The second critique concerns methodology rather than statistics. Dunbar's 1992 paper built its regression from data on 36 primate genera and then extended the line outward to predict a value for Homo sapiens, a data point that lies outside the observed range entirely. This is extrapolation, not interpolation, and the distinction matters. Interpolation makes predictions within the boundaries of known data, where model assumptions are tested and uncertainty is constrained. Extrapolation projects beyond those boundaries, where small errors in the slope compound and the model may no longer describe the underlying relationship. As Michael Goff's analysis makes clear, the headline figure rests on an out-of-sample projection, a methodological limitation that most popular treatments never flag.

Causation Remains Contested

A third line of criticism targets the causal story itself. Dunbar's framework depends on the claim that social complexity, rather than ecological pressure, drove neocortex expansion across primates. Some evolutionary researchers continue to contest this, arguing that foraging demands and habitat complexity remain plausible primary drivers of brain size. If ecology is the true selective pressure, the inferential chain linking neocortex volume to a social group limit loses its foundation, and the ~150 figure becomes a correlation without a confirmed mechanism.

Individual Variation Challenges Universal Layers

Perhaps the most practically significant recent challenge comes from a 2025 peer-reviewed study in PLOS ONE by Li, Lee, Stahl, and Bayer. The researchers found substantial individual differences in how people allocate relational energy across social layers, suggesting that Dunbar's concentric circles are not fixed universal structures but vary meaningfully from person to person. This finding matters for organizational design because it implies that a single template derived from the population average will systematically misfit a large proportion of any real team. Some individuals can sustain a genuine inner circle of ten; others are relationally saturated at four. Treating the layers as biologically fixed ignores this variance entirely.

The Right Conclusion: Heuristic, Not Hard Law

None of these critiques demolish the observation that human social groups consistently cluster near a recognizable upper boundary. The ethnographic and historical evidence for that clustering is real. What the critiques do dismantle is the stronger claim that 150 is a precise, universal, biologically ordained limit. Recognizing the difference matters enormously for practice. Organizations that treat 150 as an inviolable law will build brittle structures, applying a rule at the wrong scale for their specific culture, function, or workforce. Organizations that treat it as a well-grounded heuristic, one that signals when relational friction is likely to increase without predicting the exact moment it will, retain the insight while preserving the flexibility to respond to what their own data shows.

The Layer Collapse Nobody Talks About

Most discussions of Dunbar's number ask a single directional question: how large can a group grow before cohesion breaks down? That framing treats the 150-person ceiling as a lid pressed downward onto organizational design, a constraint on how many people a leader can effectively manage. The question is legitimate, but it is only half the picture. Reading the layers upward, from the apex of a hierarchy toward the outermost circles rather than from the base inward, reveals a complementary and far more uncomfortable problem. As a leader climbs, the innermost layer is not preserved while the outer layers erode. The innermost layer collapses first.

The Asymmetry That Hollows the Center

Dunbar's four nested circles distribute roughly 40% of a person's total relational energy among their five closest relationships, with a further 20% absorbed by the next ten, meaning 60% of all social investment concentrates in just fifteen people. That distribution assumes something critical: the energy exchange is reciprocal. Both parties invest, both parties receive. Positional authority disrupts that reciprocity at its foundation.

When a person holds formal authority over others, the relational dynamic shifts structurally. The people below a senior leader have strong institutional incentives to manage the relationship with care: to present well, to signal alignment, to avoid delivering unwelcome information too bluntly. They are not being dishonest, they are being rational. The result is that a VP or CEO receives a continuous stream of socially calibrated signals rather than genuine peer-level intimacy. The five-person inner circle appears populated. The intimacy that defines membership in it has quietly been replaced by managed performance. The layer erodes from the inside before the leader notices it is gone.

The Apex Alone Condition

This produces what might be called the apex-alone condition: a leader who commands span-of-control over hundreds of people while simultaneously belonging to the innermost Dunbar circle of effectively zero. The number of people they influence grows with every promotion. The number of people who can offer them unfiltered challenge, genuine care, or honest peer-level correction shrinks toward none. Position and intimacy move in opposite directions along the same career trajectory.

The problem is not visible in org charts, headcount reports, or engagement surveys, all of which measure relational breadth rather than relational depth. A leader who scores well on 360-degree feedback may still be entirely isolated at the innermost layer, not because the feedback is dishonest, but because the feedback mechanism itself reflects the managed performance dynamic rather than genuine intimacy.

Structural Problem, Not Personal Failing

Leadership loneliness is frequently treated as a coaching problem, a matter of emotional intelligence, self-disclosure, or work-life balance. That framing misidentifies the mechanism. Individual differences in how people allocate energy to relationships are real and documented, but they do not explain why loneliness concentrates so predictably at organizational apex positions regardless of the leader's personality, warmth, or social skill. The predictability is the signal. Predictable outcomes point to structural causes, not individual ones.

The inner-circle collapse is not addressed by offsite retreats that simulate intimacy for seventy-two hours before everyone returns to their reporting lines. It is not solved by executive coaches who offer a confidential relationship but not a peer one. It is not remedied by expanding span-of-control training, which addresses the downward-facing question and leaves the upward-facing one entirely untouched.

The Cone Model's structural response to this condition is specific and deliberate: place two people at the apex, not one. A genuine deputy with real delegated authority is not succession planning. It is the mechanism by which the leader regains access to the innermost Dunbar layer that positional authority systematically destroys. The cure matches the diagnosis because both are structural.

Oscillating Cognitive Load Across Organizational Altitude

The standard Dunbar ceiling argument is built on a single directional logic: keep groups below 150 and relational management remains tractable; push beyond it and cohesion fractures. That framing has genuine explanatory power at the macro level, and the three decades of corroborating evidence are not trivial. But it is a static model applied to a dynamic problem. It treats cognitive load as a threshold to avoid rather than a variable that shifts shape and character as leaders move through different altitudes within an organization. For practitioners designing real structures, this is a significant analytical gap.

The Fractal Span Pattern

Consider what actually happens to span of control as authority ascends through a well-structured organization. A team leader manages approximately ten people, operating comfortably within Dunbar's innermost layers of five and fifteen. Their supervisor manages roughly five such leaders. The director above manages two or three supervisors. Then, at the next power-of-ten threshold, something structurally interesting occurs: the pattern resets, and a new circle of approximately ten direct reports becomes the norm again. Emerging practitioner literature on organizational phases and the Dunbar threshold captures this fractal span-reset pattern clearly, noting that it recurs across organizational layers rather than following a simple linear trajectory.

This oscillating pattern is not visible in the conventional ceiling framing. The conventional model predicts a smooth accumulation of relational complexity as headcount grows. The fractal reality is more nuanced: span narrows dramatically at intermediate altitudes, then resets to a broader circle at each new organizational tier.

Compression Is Not Relief

The critical insight is that moving from a span of ten to a span of two does not reduce cognitive load; it transforms it. A leader narrowing from ten direct relationships to two experiences a compression of relational diversity, not a holiday from relational work. The quality of connection required at that compressed altitude intensifies. The stakes of each individual relationship rise. And, crucially, the structural conditions that generated peer support at the wider layer have thinned or disappeared entirely.

This matters because Dunbar's nested layers, particularly the fifteen-person and fifty-person circles, are not merely containers for relationships. As Dunbar himself has argued, these middle layers function as the organic social infrastructure of trust, reciprocity, and mutual accountability within communities. When a leader ascends to a span of two or three, they are no longer embedded in those generative layers. The relational ecosystem that sustained them at lower altitudes has contracted to almost nothing.

When Peer-Generated Needs Go Unmet

The Cone Model identifies fourteen needs that people require in order to flourish at work, and these needs are not met by the leader from above. They include Belonging, Safety, Trust, Friendship, Accountability, Encouragement, and nine others; all of them are generated laterally, peer to peer, within functioning relational layers. These needs are structurally dependent on the fifteen- and fifty-person Dunbar layers operating correctly. When altitude compresses those layers, the fourteen peer-generated needs do not simply diminish gradually. They lose their structural source. The leader is not failing to provide them; the architecture is no longer producing them at all.

This reframes what leadership loneliness actually is. It is not a personality deficiency. It is a predictable structural consequence of ascending past the relational layers that generate peer support, without any compensating design put in their place.

Redesigning the Question

Recognizing cognitive load as oscillating rather than linear changes the design question organizations should be asking. The question is not "what is the maximum team size before things break?" That question focuses attention on a single ceiling and produces a single rule. The more precise and productive question is: what relational structures need to be deliberately built at each altitude so that load remains sustainable and peer-generated needs continue to be met as spans narrow and reset?

A 2025 study published in PLOS ONE adds a further layer of complexity, finding significant individual variation in how leaders allocate relational energy across their social circles. Some concentrate deeply in the innermost layer of five; others distribute more broadly. This means the oscillating load experienced at altitude will not compress identically for every leader, and structural design must account for that variability rather than assuming a universal response.

The design implication is architectural: organizations that understand altitude-based oscillation can deliberately build what the structure stops generating, whether through structured peer cohorts at senior levels, formal deputy relationships at the apex, or other mechanisms that restore the relational conditions the narrowing span has removed.

The Structural Response to Inner-Circle Collapse

The most direct structural response to inner-circle collapse at the apex is deceptively simple: place two people there rather than one. Not a chief of staff whose authority evaporates at the door of the boardroom. Not a succession-planning candidate who is being groomed rather than trusted. A genuine deputy with real delegated authority, shared exposure to apex-level cognitive load, and the standing to push back when the leader is wrong. This is the structural cure that the Cone Model advances, and it is grounded in a precise reading of Dunbar's layered framework, not merely in leadership intuition.

The Deputy Relationship and the Innermost Layer

Dunbar's innermost circle of approximately five people is defined by deep trust, genuine reciprocity, and shared emotional exposure. Research on relational energy allocation suggests that people invest roughly 40% of their total social effort in these five closest relationships, with a further 20% concentrated in the next ten. That means 60% of a person's relational capacity is bound up in just fifteen people. If the innermost layer is empty or populated only with subordinates and stakeholders rather than genuine peers, no outer layer compensates. The energy concentration is too steep and the relational character of each layer too qualitatively distinct for substitution to work.

A genuine deputy relationship sits squarely within this innermost layer and is the only relationship available to a senior leader that is simultaneously structurally peer-like and organizationally sanctioned. A board member operates at a different altitude with different accountability. A functional peer in another division carries different pressures and different information. Only a deputy, sharing the same strategic exposure, the same organizational consequences, and the same daily cognitive load, can occupy the innermost layer in a way that is both authentic and structurally supported. This is why the Cone Model places two people at the apex, not as a redundancy measure, but as the structural cure for the isolation that positional altitude otherwise makes inevitable.

Why Headcount Ceilings Are the Wrong Diagnostic Tool

The 2025 PLOS ONE research by Li, Lee, Stahl, and Bayer adds an important dimension to this argument. Their finding that relational energy allocation varies significantly by individual challenges the premise underlying most Dunbar-derived organizational rules: that a single number, whether 5, 15, or 150, applies uniformly across leaders. Some individuals sustain richer inner circles at higher altitude. Others reach a functional inner-circle size of near-zero far earlier than their org chart would suggest. Rigid span-of-control rules derived from a universal figure cannot distinguish between these cases.

This is precisely the limitation that Dunbar himself has acknowledged over 30 years of scrutiny: the layers describe population-level tendencies, not individual prescriptions. For organizational designers, the implication is that headcount ceilings are a blunt instrument. What leaders actually need is a diagnostic lens on their own relational structure as it currently exists, not a ceiling imposed from population averages.

Surfacing What Org Charts Cannot See

Standard organizational charts map reporting lines, spans of control, and positional hierarchy. They do not map relational density, inner-circle occupancy, or the lived experience of isolation at altitude. A leader can appear structurally supported on paper while their actual innermost layer is functionally empty. This gap is invisible in conventional analysis and is therefore consistently unaddressed.

The Cone Model's free Proximity and Lone-Wolf Check assessment is designed precisely to surface this gap. It identifies whether a leader's actual relational inner circle is populated or effectively vacant, a question that no org chart, performance review, or engagement survey is built to answer. The diagnostic shift matters: from describing what the structure looks like to revealing how it functions for the individual leader inside it.

Distributing Relational Load Structurally

Beyond the deputy relationship, the Cone Model identifies fourteen peer-generated needs, including Belonging, Trust, Friendship, Correction, and Accountability, that must be met within the 15- and 50-person Dunbar layers by peers, not delivered from the apex downward. Seven of the twenty-one needs people require to flourish at work legitimately belong to the leader: Direction, Vision, Mission, Mentoring, Reward, Advocacy, and Autonomy. The remaining fourteen are structurally dependent on the horizontal peer relationships that populate the inner and intermediate layers.

Attempting to deliver all twenty-one needs from a single apex is both cognitively impossible and structurally incoherent. It overloads the leader, hollows out the peer layers, and guarantees that the fourteen peer-generated needs go unmet regardless of the leader's effort or intention. The practical implication for organizational designers is to stop treating Dunbar's nested layers as an external constraint on headcount and start treating them as a relational architecture requiring conscious design, active maintenance, and deliberate protection at every altitude within the organization.

Conclusion: From Cognitive Heuristic to Structural Design Tool

Dunbar's number is a robust cognitive heuristic with convergent support from neuroscience, military history, and operational design, but its most consequential implications have been consistently under-read. The layers do not simply impose a ceiling on outward growth; they collapse inward under altitude. The innermost circle of approximately five people is the most structurally vulnerable layer in any organisation, and the leader at the apex bears the full weight of that vulnerability without the peer-generated support that sustains everyone below them.

Reframing the nested layers as dynamic relational architecture, rather than a static headcount ceiling, changes the practical questions leaders ask. The issue is not merely how large a team can grow before cohesion fractures; it is whether the 15- and 50-person layers are actively generating the fourteen peer-to-peer needs such as belonging, trust, accountability, and care, or whether those needs are silently migrating upward and depleting the leader's capacity to function sustainably.

The immediate actionable step is concrete: audit the actual population of your innermost five-person circle, assess whether a genuine deputy relationship exists with real delegated authority, and identify which relational needs your structure is genuinely meeting versus leaving exposed. The free Proximity and Lone-Wolf Check at theconemodel.com surfaces exactly this picture in under ten minutes, converting Dunbar's cognitive heuristic into a structural design intervention you can act on today.

Buy Now on Amazon Take the free 360° Assessment About the Book