When the Pipeline Is Full but the Knowledge Doesn't Move

Organizations have spent considerable energy over the past three decades solving the wrong half of the knowledge transfer problem. The architecture of most knowledge management initiatives rests on a diagnostic premise: that the bottleneck lies with the sender. Firms have built communities of practice, cross-functional forums, and enterprise collaboration platforms premised on the idea that knowledge would flow freely if only individuals and units could be persuaded to share it. The incentive systems, the cultural interventions, and the technology investments all point in the same direction toward the source.

The evidence increasingly suggests the diagnosis is incomplete. In a landmark study of 122 best-practice transfers across eight large companies, Gabriel Szulanski found that the primary barriers to internal knowledge transfer were not motivational but cognitive: specifically, the recipient's absorptive capacity deficit, causal ambiguity in the knowledge itself, and the quality of the source-recipient relationship.¹ Motivational resistance - what practitioners often call the "not-invented-here" syndrome - ranked substantially lower. The implication should have been disorienting for management practice. It was not, perhaps because it pointed toward problems that are harder to solve with incentives and culture-change programs than with better sharing mechanisms.

What Szulanski identified at the firm level warrants more granular examination at the level of the receiving team itself. This piece argues that the most underexamined constraint on organizational knowledge utilization is neither motivational resistance nor prior knowledge deficit. It is what might be called the integration cost problem: the real-time cognitive work a receiving team must perform to translate incoming knowledge from a foreign problem context into its own decision-relevant vocabulary, and the organizational failure to recognize that work or support it.

Figure 1. Where Knowledge Transfer Breaks Down. The integration cost barrier (right) operates even in willing, capable recipients - independently of the motivational barrier (left) that dominates current management practice.

A Structural Friction That Persists Even in Willing Recipients

When Cohen and Levinthal introduced absorptive capacity, they argued that a firm's ability to recognize and exploit external knowledge is largely a function of prior related knowledge.² Zahra and George subsequently refined this framework by distinguishing between potential absorptive capacity, i.e., the ability to acquire and assimilate external knowledge, and realized absorptive capacity, i.e., the ability to transform and exploit it.³ The gap between these two, they argued, is where organizational value routinely fails to materialize. That gap is precisely what the integration cost problem describes, but at a level of resolution the absorptive capacity framework does not reach.

When a team receives knowledge from a different organizational unit, it does not arrive as neutral information awaiting merit-based evaluation. It arrives shaped by the problem representations, interpretive conventions, and decision vocabulary of its originating context. Before any integration can occur, the receiving team must perform active translation work -- recasting the incoming knowledge into categories legible within their own cognitive and operational framework. Hargadon and Sutton, in their study of the design firm IDEO, observed something structurally similar: that organizations need dedicated brokering routines not because knowledge is guarded or hidden, but because connecting solutions from one domain to problems in another is genuine cognitive work.⁴ The same logic applies inside organizations.

Critically, this translation burden persists in well-motivated, well-resourced recipients. It is not a subspecies of the not-invented-here syndrome, which Antons and Piller have carefully characterized as a negative attitude toward externally sourced knowledge rooted in identity threat, in-group favoritism, and perceived challenge to epistemic authority.⁵ A team may hold no such attitudes and still fail to integrate cross-boundary knowledge under operating conditions, simply because the translation cost is real and the organizational environment neither recognizes nor allocates resources to bearing it.

This distinction matters considerably for practice. The interventions that address motivational resistance, namely incentive redesign, leadership signaling, community-of-practice investment, do not reduce the translation burden on a receiving team that is already motivated to integrate but lacks the bandwidth, the shared vocabulary, or the structural permission to do so. The managerial levers point in different directions.

An Unexpected Mechanism in AI Adoption

An entirely separate stream of inquiry on artificial intelligence introduces a possibility worth taking seriously, even if the causal chain is not yet established empirically. Recent work by Burton and colleagues, published in Nature Human Behaviour, identifies large language models as a potentially significant force reshaping how information is aggregated and transmitted in organizations and societies.⁶ Their focus is on collective intelligence at scale on how AI-mediated information environments alter deliberation, consensus formation, and problem-solving in groups. But their framework surfaces a micro-level behavioral dynamic that deserves independent examination.

Consider the routine cognitive experience of a professional who works regularly with AI-generated outputs: she receives synthesized analyses, proposed framings, and draft solutions that originate entirely outside her own reasoning process. She then evaluates these on their merits, decides what to retain, and integrates selected content into her work. The social threat architecture that normally complicates cross-boundary human knowledge reception is structurally absent. The AI system holds no departmental allegiance, makes no career claim, and poses no competitive threat to the receiving team's epistemic standing. The professional can engage with its outputs as information in a relatively pure form.

The behavioral implication is important. If regular AI interaction provides repeated, low-stakes rehearsal of the specific cognitive move that integration requires -- decoupling epistemic assessment from social signaling, asking "is this useful?" rather than "what does accepting this imply about us?" -- then organizations with high AI adoption may be inadvertently developing more receptive workforces. The practice of merit-based evaluation, exercised routinely under socially neutral conditions, may lower the activation threshold for the same evaluation behavior in more socially fraught cross-boundary contexts.

This is a hypothesis, not a finding, and it carries a significant countervailing risk. Messeri and Crockett have documented how AI use can generate an illusion of explanatory depth: users who rely on AI to synthesize or frame a problem may mistake the AI's apparent coherence for genuine understanding of their own.⁷ If AI interaction becomes passive rather than active, if it trains reliance rather than critical engagement, the receptivity spillover disappears and what remains is a different kind of cognitive degradation. The mechanism described here operates only under conditions of engaged, evaluative AI use, and distinguishing those conditions empirically from unreflective reliance is a non-trivial research challenge.

Figure 2. The AI Receptivity Spillover Hypothesis. Regular, critical engagement with AI-generated outputs may train merit-based evaluation habits that transfer to cross-boundary human knowledge reception. The mechanism is conditional on engaged (not passive) AI use.

The Research Agenda and the Practical Implication

Three questions follow from this analysis that have not been adequately addressed by existing work. First, can integration cost be operationalized and measured at the team level, independently of prior knowledge stocks and motivational attitudes? The absorptive capacity literature has strong firm-level constructs; the team-level translation burden has no comparable measurement tradition. Second, does the pattern of AI use within teams, specifically, whether interaction is evaluative or delegative, moderate the relationship between AI adoption and cross-silo knowledge integration? The difference between practicing critical evaluation and offloading cognitive effort to a language model may be the pivotal variable in assessing AI's organizational learning effects. Third, what organizational structures reduce integration cost directly? Dedicated translation time in project cycles, integrator roles whose mandate is reformulating incoming knowledge into a team's working vocabulary, and performance metrics that capture what a unit incorporates rather than only what it produces - these are interventions that address the receiving side of the transfer equation and have received almost no systematic empirical attention. The pipeline metaphor that dominates knowledge management thinking is misleading in a specific way. It treats knowledge as a substance that moves, accumulates, and can be unblocked. What actually moves is symbol systems that must be reinterpreted on arrival. How that reinterpretation happens, and under what conditions it fails, is where the work lies.

References

1. Szulanski, G. (1996). Exploring internal stickiness: Impediments to the transfer of best practice within the firm. Strategic Management Journal, 17(S2), 27-43. 

   
   

2. Cohen, W.M. and Levinthal, D.A. (1990). Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly, 35(1), 128-152.

   
   

3. Zahra, S.A. and George, G. (2002). Absorptive capacity: A review, reconceptualization, and extension. Academy of Management Review, 27(2), 185-203.

   
   

4. Hargadon, A.B. and Sutton, R.I. (1997). Technology brokering and innovation in a product development firm. Administrative Science Quarterly, 42(4), 716-749.

   
   

5. Antons, D. and Piller, F.T. (2015). Opening the black box of "Not Invented Here": Attitudes, decision biases, and behavioral consequences. Academy of Management Perspectives, 29(2), 193-217.

   
   

6. Burton, J.W. et al. (2024). How large language models can reshape collective intelligence. Nature Human Behaviour, 8(9), 1643-1655.

   
   

7. Messeri, L. and Crockett, M.J. (2024). Artificial intelligence and illusions of understanding in scientific research. Nature, 627, 49-58.

Anil Kshatriya is an Assistant Professor of Accounting and Management Control at ESSEC Business School (Cergy, France). He holds a PhD in Accounting from the University of Amsterdam. His research sits at the intersection of accounting, economics, and behavioral science, with a focus on how management control systems shape knowledge-sharing, transparency, and decision-making in organizations.