Collaboration is supposed to be good for organisations. That, at least, is the received wisdom of the last four decades. Flatten the hierarchy, break down the silos, connect the people, and watch performance improve. And yet, something has gone badly wrong. Collaboration — as currently practised in most knowledge-intensive organisations — is not generating the performance gains its advocates promised. It is generating burnout.
The data tell a stark story. According to research conducted by Cross, Rebele, and Grant and published in the Harvard Business Review, the time spent by managers and employees in collaborative activities has increased by more than 50% over two decades, with many knowledge workers now spending upward of 80% of their working day in meetings or responding to colleagues’ requests [1]. Atlassian’s 2024 State of Meetings report found that 78% of workers say they are expected to attend too many meetings, while more than half said they regularly work overtime simply to complete the individual tasks that collaborative demands have displaced [2]. And the critical insight from the HBR research — one that has only sharpened since its 2016 publication — is that this burden is not evenly distributed. Between 20% and 35% of the value-added collaboration in a typical organisation flows through just 3% to 5% of its people.
That concentration is the mechanism that burns people out in collaboration. And understanding it — at a structural, rather than a cultural or motivational level — is the essential first step in doing anything about it.
This article is not an argument against collaboration. Working together to accomplish things that individuals cannot achieve alone has been fundamental to human organisation throughout history [3, 4], and it remains indispensable. The argument is a different one: that the structural conditions under which collaboration now operates in most organisations — the team sizes, the degrees of virtuality, the levels of task complexity, the volume of technology-mediated interaction, and the absence of any serious attempt to manage the human costs of network centrality — make high performance unsustainable and burnout among the most capable people structurally inevitable.
What follows is an explanation of why. It requires some reading. I make no apology for that.
Historical Context: How We Got Here
Most people would say they know what a team is, regardless of whether they have worked on one. The term is so widely used as to be almost meaningless [5]. In an organisational setting, every person is likely to identify with several teams simultaneously. Even in our private lives, group membership gives us a sense of identity and defines who we are.
This should come as no surprise. We are tribal and social by nature. Forming groups to collaborate on tasks we cannot complete independently has been common to humans throughout history [3] — and is just as prevalent in the natural world [6–8]. The study of organisational groups and teams is, as a result, one of the most intensively researched domains in organisational science, despite — or perhaps because of — more than a century of accumulated evidence [9].
Over the last fifty years, in particular, organisations have sought to harness the benefits of collaborative working [10, 11], replacing hierarchical structures with flat, lean designs [12] that require creative thinking and intensive collaborative problem-solving [13] in the face of loosely defined goals, high uncertainty, and complex tasks. Many will remember the shift: from the scientific management era [14] with its high-conflict, carrot-and-stick extrinsic motivational systems, to the humanistic management era [15–24], justified by the argument that highly prescribed work methods had a de-humanising effect and led to motivational loss [25].
The legacy of that transition is the world in which we now work. Today it is hard to find any management literature that does not report how organisations have shifted from individual to team-based working [26–32]. Fortune 1000 firms reported a steady increase in the use of team structures from 20% in 1980 to over 80% by 2000 [33]. The group, not the individual, has become the foundational unit of the organisation [34] — and with it, the foundational unit of collaborative demand.
Current Context: An Intensifying Problem
What was a manageable trend has, in the years since 2016, become a structural crisis. The COVID-19 pandemic dramatically accelerated the shift to remote and virtual working, embedding distributed structures in organisations that had previously relied on colocation to provide the informal coordination that compensated for weak team design. Almost overnight, the ambient social environment of the office — the overheard conversations, the corridor exchanges, the physical proximity that enables rapid, low-cost information transfer — was removed. In its place came an unprecedented volume of scheduled, technology-mediated interaction.
The traditional team form [35, 36] is now so rare in many organisations [37] that it is in danger of extinction. Many teams are so radically changed in their structure and membership that they are barely recognisable as teams at all. This has profound implications for performance management: much of the knowledge we have accumulated about how to compose, develop, and lead teams was built on a model of collocated, bounded, relatively stable groups that simply no longer reflects organisational reality [38].
Understanding where we are, how we got here, and how to function effectively in the virtual social networks [39–43] that have evolved is therefore not an academic exercise. It is an operational necessity.
Structural Drivers
Organisations large and small are confronting a compound set of structural challenges that, individually and especially in combination, place extraordinary demands on collaborative capacity. These include: the growing centrality of the knowledge economy [44]; increasing goal uncertainty and task complexity; volatile global economic conditions [45]; the de-layering of hierarchical structures; de-routinisation and the rise of ‘swarming’ patterns of work [46]; increasingly frequent restructuring and reorganisation; multi-team membership [47]; shared and emergent leadership [48–52]; self-managing teams [53–58]; and the fragmentation resulting from outsourcing, remote working, matrix structures, rising functional and cultural diversity [59–62], and the pluralistic systems [63] that create competing demands on already-scarce human resources [64].
Each of these forces individually increases interaction volume. Together, they create organisational environments in which boundary conditions are ambiguous, colleague engagement is compromised, coordination is disrupted, unproductive conflict is stimulated, and performance failure becomes structurally probable.
Situational Drivers
Compounding these macro forces, the rise of cluster structures and Organisational Communities of Practice [65, 66] — sometimes referred to as Centres of Excellence — has further complicated the social architecture of knowledge work. These structures are representative of the emerging pluralistic organisation [63] and create their own distinctive tensions between performance demands and collaborative capacity.
The uncertainty inherent in knowledge work requires intensive collaboration between interdependent individuals who are often deployed in distributed or dispersed structures [67]. The contrast between collocated groups reporting to designated leaders and the distributed, self-managing workgroups operating under emergent leadership that now characterise much of the knowledge-intensive industry is almost total. The traditional assumptions of team research do not cleanly transfer from one to the other [68–80].
Meanwhile, the expectation that team members can participate in multiple teams simultaneously has become normalised. Estimates suggest that between 65% [81] and 95% [82] of knowledge workers participate in multiple teams simultaneously — a condition that creates role ambiguity, role conflict, and motivational loss at scale. Studies of matrix organisations confirm that the complex decision-making processes inherent in horizontal communication structures result in role conflict, negative attitudes, and moral disengagement, including free-riding and social loafing [83–98].
Virtual Teams: Amplifying the Problem
Virtual teams — defined as groups of people who work interdependently across space, time, and organisational boundaries, using technology to communicate and collaborate [99] — have become the dominant form of organised work in knowledge-intensive sectors. As many as 50% of employees may now be working virtually in some capacity [100], a proportion that has risen sharply since 2020. A pre-pandemic survey found that 80% of organisations expected their use of virtual teams to increase [101]; post-pandemic, that trend has become simply the reality of how work is organised.
Dispersion is a defining and challenging feature of virtual teams, since members often span international boundaries and cultures [102]. But it is not only geographic distance that matters. It is the nature of the interactions that distance forces — technology-mediated communications via video conferencing, email, instant messaging, intranet, and enterprise collaboration platforms. These channels are categorically different from the context-rich, high-bandwidth communication of face-to-face interaction [103], and they impose specific cognitive and social costs that compound the collaboration problem considerably.
Research on virtual team dysfunction has consistently identified a cluster of associated pathologies: low individual commitment, role overload, role ambiguity, absenteeism, attrition, and various forms of moral disengagement [104]. Pronounced attitudinal and behavioural differences among culturally and functionally diverse members further complicate coordination [105]. Some estimates suggest that between 50% [106] and 70% of virtual teams fail to meet their performance objectives — a failure rate that has barely changed despite significant organisational investment in collaboration technologies.
The Meeting Overload Crisis
One consequence of the shift to virtual working that has received belated but increasing research attention is the escalation of meeting load. Data from Flowtrace’s 2024 analysis found that half of all meetings start late, nearly two-thirds lack a clear agenda, and participant lists are routinely bloated [2]. Atlassian’s research confirms the picture: 65% of senior managers say meetings prevent them from completing their own work, and 71% describe their meetings as unproductive and inefficient [2]. Fifty-five per cent of remote workers believe a majority of their meetings could have been handled through asynchronous communication.
A 2024 study conducted in Germany using experience-sampling methodology over 590 workdays found that while acute ‘Zoom fatigue’ has partially habituated since the early pandemic period, the structural burden of meeting volume has not diminished. Workers have adapted to the format while remaining overwhelmed by its frequency [107].
This is an important distinction. The problem is not the technology of video meetings per se. It is the volume of meetings, the lack of discipline around their design and participant lists, and the fact that meeting time is almost universally scheduled without regard for the cognitive cost it imposes on the individuals who attend. Research on meeting recovery by Allen and colleagues [108] confirms that the absence of a transition time between meetings — a near-universal feature of modern digital calendars — significantly elevates burnout risk, as it prevents the attentional recovery required for sustained cognitive performance.
Task Complexity, Interdependence, and the Hidden Cost of Coordination
Task complexity is a function of the degree to which workflows require independent or interdependent working. This distinction is central to understanding why the combination of knowledge work and virtual team structures is particularly costly in human terms.
Simple tasks that can be completed independently are well-suited to virtual working. They require limited synchronous coordination, generate low interaction volume, and place modest demands on shared understanding. Complex tasks — where work flows back and forth between individuals and groups, requiring continuous recombination of outputs and perspectives — are an entirely different matter. They require a high degree of collaboration, coordination, and social integration, all of which depend on synchronous, real-time communication [109].
While technology enables interdependent working within distributed structures, it does so at a cost. Virtually mediated teams have substantially greater difficulty developing shared understandings of complex tasks. The stress of managing uncertainty in low-bandwidth communication environments leads to more frequent conflict, which inhibits the trust-building that effective collaboration requires. The interaction volume associated with managing these dynamics in a virtual context is, as a consequence, significantly higher than in equivalent collocated settings.
There is a cruel irony at work here. The organisations that most need the performance benefits of collaboration — those working on complex, uncertain, high-stakes knowledge problems — are precisely those most severely exposed to the costs of collaborative overload.
The Mathematics of Overload: Bossard’s Law
Delayering and virtualising operations produce a compound effect: larger teams, greater dispersion, and greater diversity. Knowledge work is, by its nature, complex, uncertain, and frequently requires creative problem-solving. This combination is a primary driver of the interaction volume that sits at the heart of the collaborative overload problem. What is less widely appreciated is the exponential relationship between interaction volume and team size — a relationship explained by Bossard’s Law [110].
Bossard’s Law, originating in sociology but directly applicable to organisational network dynamics, describes the nature of interactions between actors in a relational system. It is expressed as:
INTERACTIONS=N(N−1) / 2
Where N = the number of actors in the binary relationship
The implications are straightforward but underestimated. In a team of 2, there is 1 interaction relationship. In a team of 5, there are 10. In a team of 10, there are 45. In a team of 20, there are 190. Each time a team member is added, the growth in interaction relationships is not linear but exponential. It does not take many additions before the interaction volume within a team becomes practically unmanageable.
Research suggests that people can generally sustain an interaction volume of around 100 for a limited period before performance begins to decline. This approximates to a group size of about 15. Yet group structures in matrix organisations and virtual teams routinely exceed this threshold — sometimes by multiples — and they do so without any systematic accounting of the human cost.
The nature of interactions in real organisational settings further compounds the effect. In practice, the relevant equation is not interactions within a single team but interactions between and within groups — a more complex version of Bossard’s formula that produces even more extreme growth curves as the scale of the organisational network increases. When the project unit is the entire organisation — as in a major digital transformation, a post-merger integration, or a cross-functional product launch — the interaction mathematics become genuinely staggering.
A 2020 quantitative study specifically modelling collaborative overload and performance found direct evidence that the relationship between collaboration engagement and individual performance is nonlinear: beyond a threshold, increased collaborative demand produces monotonically declining performance. The authors confirmed that knowledge transfer is not evenly distributed: a small number of key expertise holders bear a disproportionate, and ultimately unsustainable, burden of collaboration [111].
Social Network Theory and the Overloaded Core
Social network theory provides the explanatory mechanism that converts the abstract mathematics of Bossard’s Law into an account of what actually happens to people in organisations. It is a framework that the original HBR analysis [1] drew upon, and one that subsequent research has substantially developed.
Every project in a knowledge-intensive organisation follows a recognisable cycle. It typically begins with an executive sponsor setting an overall goal — usually to solve a problem that is vaguely articulated at the outset. In the beginning, uncertainty is at its maximum: the nature of the problem is unclear, the solution space is undefined, and the required resources are unspecified. The team begins by reaching outward into the organisation’s network, searching for expertise, information, and support — a process that social network researchers call sense-making [112].
As team members acquire information, they share it with others who need it — sense-giving [113]. This cycle of sense-making and sense-giving repeats as the project progresses, with each new team member extending the network further, adding new interaction relationships to the existing load. The goal throughout this process is to build and maintain a Shared Mental Model [114, 115] — a common, aligned understanding of the problem, the solution, and what each member needs to do. In complex, uncertain projects, maintaining that shared understanding requires continuous interaction. Every change in scope, resource, or direction requires another round of sense-making and sense-giving to keep the model current.
Social network theory explains that certain individuals — by virtue of their tenure, experience, knowledge, skills, personality, or organisational position — will become central to the network. These people occupy what researchers call in-group positions. They are well-connected, their social capital is high, they have extensive networks of both strong and weak ties, they attend important meetings, and they are the first to receive and disseminate critical organisational information. As a result, the interaction traffic directed toward these individuals is disproportionately high. They become organisational bottlenecks: indispensable, heavily loaded, and increasingly at risk.
A 2023 study published in Applied Ergonomics provided the most rigorous quantitative evidence to date of the link between network centrality and burnout, using three network centrality metrics (in-degree, closeness, and betweenness) and the Maslach Burnout Inventory. The researchers found a clear, statistically significant association: the more central an individual is to their organisational advice-seeking network, the more severe their burnout symptoms [116]. This is not a correlation that should surprise anyone familiar with the social network literature — but it is one that most organisations continue to ignore entirely in their performance management and wellbeing strategies.
The institutional response to overloaded in-group members is almost universally counterproductive. When a key individual becomes overwhelmed, the organisation’s instinct is to give them more resources — additional team members who can share the load. But according to Bossard’s Law, adding members to an already overloaded team does not reduce the interaction burden on the centre. It increases it, as the new members themselves need onboarding, coordination, and sense-making support before they can contribute independently. The problem compounds.
Hyperconnectivity and the Always-On Culture: A New Dimension of Overload
The shift to virtual and hybrid working has added a dimension to collaborative overload that was less prominent in the literature before 2020: the erosion of temporal and psychological boundaries between work and non-work. In the collocated organisation, the physical act of leaving the workplace provided a default boundary that, however imperfectly, protected recovery time. In the always-connected digital organisation, no such default exists.
European research has documented that teleworkers are significantly more likely to work during what would otherwise be their free time, and substantially more likely to struggle to disconnect from work at the end of the formal working day [117]. This pattern of ‘soft overtime’ — unpaid work driven not by explicit managerial demand but by the social expectation of digital availability — has been identified as a major driver of stress, anxiety, and burnout among knowledge workers. The phenomenon of technostress — the psychophysiological strain associated with the expectation of continuous connectivity — has become a recognised and clinically significant occupational health concern [118].
A 2024 PRISMA-based systematic review examining the effects of digital connectivity on occupational health concluded that while digital tools genuinely improve efficiency and flexibility, they simultaneously increase cognitive overload, workload, and stress [119]. Prolonged digital engagement is associated with mental exhaustion and sleep disturbance — two conditions that directly compromise the higher-order cognitive functions on which knowledge work depends: problem-solving, creative thinking, judgment, and sustained attention.
The political and legislative response to these findings is beginning to emerge. A growing number of jurisdictions have introduced or are considering ‘right to disconnect’ legislation, establishing that employees have a legally protected entitlement to ignore work communications outside their contracted hours. These legislative developments reflect a growing recognition that the always-on culture of digital knowledge work is not a natural or inevitable condition — it is the product of specific organisational choices, and it can be changed by different ones.
A 2024 study on digital workplace information overload by Marsh and colleagues found that it is specifically anxiety about information — the fear of missing something important, of falling behind, of being excluded from a critical decision — rather than the use of digital tools per se that drives the most harmful wellbeing outcomes [120]. This finding has important design implications. The primary lever for reducing harm from hyperconnectivity is not persuading individuals to put down their phones; it is restructuring the organisational information environment so that the fear of missing out is not a rational response to one’s actual situation.
The AI Paradox: When the Cure Makes Things Worse
The obvious contemporary response to collaboration overload is technology. If we are overwhelmed by the volume of meetings, messages, and coordination demands, surely the accelerating capabilities of artificial intelligence can help — by summarising, filtering, automating, and augmenting the collaborative work that consumes so much of the knowledge worker’s day?
The evidence, so far, is sobering. Cal Newport, whose analysis of knowledge work productivity has been among the most influential of the past decade [121], has documented a pattern that recurs across the history of collaborative technology adoption: a new tool promises to reduce the friction of certain tasks, everyone becomes excited about the prospect of more time for high-value work, and the actual result is a net increase in the volume of low-value tasks — because the tool makes those tasks faster without reducing the demand for them or the social expectations that surround them. Newport cites a recent study showing that introducing AI tools increased administrative tasks by more than 90% while reducing deep work effort by almost 10% [122].
We have seen this pattern before. Email was supposed to reduce the need for meetings. Instead, it added a new and parallel layer of interaction demand. Mobile computing was supposed to increase efficiency by enabling work from anywhere. Instead, it eroded the boundary between work and recovery. Online meeting software was supposed to reduce the cost of coordination. Instead, it reduced the friction of scheduling meetings to the point where they proliferated beyond any rational assessment of their value.
The pattern with generative AI tools appears to be following the same trajectory. A 2024 study by Humlum and Vestergaard found that 27% of Danish knowledge workers in exposed occupations were using ChatGPT at work by the end of 2023 [123]. But adoption is outpacing evidence of benefit. Research on Microsoft 365 Copilot — the most widely deployed enterprise AI assistant — suggests that productivity gains are concentrated in specific, narrowly defined tasks, while the broader effects on meeting culture, information demand, and collaborative expectation remain largely unchanged or potentially aggravated [124].
None of this is an argument against AI adoption. It is an argument for the same thing that applies to all collaborative technologies: tools do not, by themselves, change the structural conditions that produce overload. Only organisational design can do that. A tool that helps a knowledge worker process their emails faster does not help them if the cultural expectation is that a faster response time will simply generate more email. The solution must be structural.
Things You Can Do: A Structural Response
There is no easy solution. But there are things — structural, behavioural, and cultural — that leaders and organisations can do to reduce the human cost of collaborative overload. What follows is an updated set of evidence-based principles, grounded in both the foundational theory outlined above and the most recent research findings.
- Treat interaction volume as a managed resource rather than a natural phenomenon. Be explicitly aware of the exponential impact of team size on interaction demand. Keep interacting in groups as small as the task genuinely allows. Resist the instinct to expand teams in response to overload — doing so almost always makes the overload worse, not better.
- Design collaborative demand to match task complexity. Audit the degree of interdependence your work actually requires. Complex, high-uncertainty tasks requiring intensive real-time coordination should involve small, stable, tightly integrated groups. Tasks that can be decomposed into independent workstreams should be structured to enable asynchronous, low-interaction working. Stop defaulting to collaborative structures for work that does not require them.
- Protect your central nodes. Identify the individuals who occupy high-centrality positions in your organisation’s advice-seeking network — they will not always be the people who appear most senior in the hierarchy. Monitor their interaction load explicitly. The 2023 research by Terra and colleagues provides a methodology for doing this using network centrality metrics [116]. Ignoring the burnout risk of in-group overload will reliably lead to the loss of the people on whom the organisation’s knowledge flows most critically.
- Do difficult things face-to-face, or as close to it as you can get. Reserve synchronous, high-bandwidth communication for the work that genuinely requires it: trust-building, conflict resolution, complex problem-solving, and the early stages of sense-making on high-uncertainty projects. Set clear communication norms that protect the more cognitively expensive channels from the low-stakes coordination that has colonised them.
- Build an asynchronous-first default, not an asynchronous-only aspiration. The 2024 research on meeting bridges [125] suggests that well-designed asynchronous information artefacts — structured records of meeting outcomes that enable post-meeting sense-making — can significantly reduce the synchronous meeting load without losing the coordination benefits those meetings are designed to provide. Organisations that have implemented structured asynchronous protocols report reductions in meeting volume without corresponding losses in alignment or output quality.
- Establish and protect a right to disconnect. The evidence for hyperconnectivity is clear: the always-on expectation actively harms the cognitive performance required by knowledge work. Organisations should establish explicit norms — ideally formalised as policy — around availability expectations outside contracted hours. Leaders should model the behaviours they want to see, not sending non-urgent messages outside working hours, not rewarding visible busyness over substantive output, and actively communicating that recovery time is organisationally valued rather than culturally sanctioned as evidence of insufficient commitment.
- Invest in boundary-spanning capability. Coach teams in the practices of Boundary Spanning [126, 127] — the skills and behaviours that enable teams to reach effectively into the wider organisation for the resources they need, while simultaneously protecting themselves from collaborative demands that exceed their capacity. Boundary spanning is one of the most consistently evidence-supported interventions for improving team performance in complex, distributed environments.
- Develop shared mental models early and maintain them efficiently. The sense-making and sense-giving cycles that drive interaction volume are most intense at the outset of projects, when uncertainty is highest, and shared understanding is most incomplete. Investing in structured, high-quality knowledge-sharing processes at the early stages of a project — rather than relying on organic, uncoordinated outreach to drive interaction volume— can significantly reduce the total collaborative cost of achieving alignment.
- Approach AI tools as organisational design decisions, not individual productivity hacks. AI-powered collaboration tools will reduce overload only if their adoption is accompanied by deliberate changes to the meeting culture, information norms, and collaborative expectations that surround them. Simply deploying a tool without changing the structural conditions that produce the overload is, on current evidence, likely to make things worse. Before adopting a new tool, ask not whether it speeds things up but whether it reduces the right things. Deep work — the sustained, focused, cognitively demanding effort from which the highest-value outputs emerge — should be the protected resource. Administrative and shallow collaborative work should be the target of automation [121].
- Think carefully about the personality profiles that are distributed in high-interaction environments. The evidence from personality research suggests that highly collaborative, high-interaction virtual environments place specific and substantial demands on individual resilience, agreeableness, and emotional regulation. Not all personality profiles are well-suited to sustained, high-volume interaction in virtual environments. Team composition decisions should account for this, and support structures should be in place for individuals whose natural profile makes them particularly vulnerable to collaborative overload.
- Set explicit priorities and centralise the management of interdependencies. Much of the unmanaged interaction volume in knowledge-intensive organisations stems from the absence of a clear, authoritative prioritisation framework. When everything is important, everything demands attention, and the interaction burden of managing competing demands falls disproportionately on the most capable and well-connected people. A visible, actively maintained priority framework reduces the number of active interdependencies at any given time and gives in-group members the organisational cover to decline or defer lower-priority collaboration requests.
The insight that runs through all of these principles is the same one that has been evident in the academic literature for two decades, but that organisational practice has been consistently slow to absorb: collaboration is not inherently valuable. It is a means to an end. The end is performance. And performance in knowledge-intensive organisations depends not on maximising the volume of collaborative interaction, but on ensuring that the right interactions happen between the right people at the right time — at a volume that the human beings involved can sustainably bear.