How Is Know-Who Accumulated? The Routes by Which Knowledge of Who Knows What Forms, and How It Is Lost

Looking for candles during a blackout

During a blackout, you may not know where the candles are, but you can find them by asking the person you live with. Wegner and colleagues opened their 1991 paper with this scene¹. Even what you do not know yourself is within reach if you know who knows it.

Economics has a name for this knowledge of who knows. Lundvall and Johnson (1994) divided knowledge into know-what, know-why, know-how, and know-who. The OECD (1996) adopted this typology and defined know-who as “information about who knows what and who knows how to do what.” It involves forming special social relationships that make it possible to reach experts and use their knowledge efficiently. The OECD went on to say that know-who is learned in social practice, develops in day-to-day dealings with customers and sub-contractors, and cannot easily be transferred through formal channels of information.

So what in social practice builds up know-who? The OECD document does not name the routes. The routes have been studied in experiments and in the field by research on transactive memory (a system in which group members know who remembers what and divide the work of remembering among themselves) in psychology, and by research on expertise location in organizations. The concept of transactive memory originates in a chapter by Wegner (1987). This note draws on both lineages to lay out the routes by which know-who forms and the conditions under which it breaks down.

Scope and method

The question is how know-who is accumulated in individuals, teams, and organizations. The active mode was academic, covering peer-reviewed papers, scholarly book chapters, and reports by international organizations. Two parallel searches yielded 66 candidates (52 after removing duplicates), and sources whose full text could be retrieved were used to support the main claims. Sources for which only the abstract could be confirmed are cited only for what the abstract says, and the text says so. The retrieval routes and results for every reference are recorded in source/review/know-who-accumulation/fulltext-manifest.json, and the supporting passage for each claim is recorded in the corpus’s verification log.

Where does the directory get its entries?

Wegner and colleagues compared know-who to memory sharing between computers. Because brains are not connected, people cannot read the contents of another person’s memory directly. Instead, each person keeps a rough directory of what the other person’s memory probably holds, and by keeping it up to date can draw on the other’s detailed memories. Accumulating know-who means writing and updating entries in this directory.

Wegner and colleagues named four sources of directory entries.

• Defaults: inferring what a stranger probably knows from social categories such as gender or age.
• Negotiated entries: the partner who agrees to pay the bills becomes the household’s repository of financial information from then on.
• Perceived relative expertise: inferring the other’s areas of expertise from preferences and past activities revealed through self-disclosure (someone who likes the zoo may know about animals).
• Access to information: deferring to the partner because they accessed the information first, for the longest time, or most recently.

These four also differ in the cues used to write an entry. Defaults are usable from the first meeting, but their content comes from the category a person belongs to rather than the person. The other three can only be obtained through repeated interaction. Wegner and colleagues argued that updating the directory beyond the defaults is the key to forming a more advanced transactive memory.

Route 1: Working together

If the directory really builds up through interaction, long-standing pairs and impromptu pairs should differ in how much they can remember. Wegner and colleagues took 118 people who had been dating for at least three months and paired them either with their partner or with an opposite-sex member of another couple, then had each pair memorize items from seven categories. When no division of labor was assigned, natural couples recalled more than impromptu pairs (means of 31.40 vs. 27.64, t(55) = 1.69, p < .05). Natural couples agreed on which partner was more expert in a mean of 5.52 of the seven categories.

The same experiment also produced the opposite result. When a division of labor was imposed from outside (“you take food, you take history”), natural couples performed worse than impromptu pairs. Wegner and colleagues read this as structural interference, a clash between the existing implicit division of labor and the imposed one. Accumulated know-who is itself hard to reorganize.

Group work shows the same direction. According to the abstract of Liang, Moreland, and Argote (1995), three-person groups trained together to assemble transistor radios recalled more of the procedure and built better-quality radios than groups formed after individual training. Analysis of videotapes indicated that the difference arose mainly through the development of transactive memory. Working together writes entries in the directory of who remembers what, alongside learning how to do the task.

Route 2: Asking, narrowing, and escalating

For people with whom they share no experience, people build know-who while searching. McDonald and Ackerman (1998) conducted a five-month field study at a medium-sized software company and divided expertise finding into three stages. Identification lists candidates, selection picks one while taking into account things like the candidate’s workload and not wanting to be seen as a pest, and escalation moves on when that is not enough. Participants used iterative procedures that minimized the number of candidates while keeping a high chance of reaching the needed knowledge.

What matters in selection is not only what the other person knows. The abstract of Borgatti and Cross (2003) states that they modeled the probability of seeking information from someone as a function of knowing what that person knows, valuing it, being able to gain timely access to their thinking, and perceiving that asking would not be too costly, and found support at two research sites (except for the cost variable). The effect of physical proximity on information seeking was mediated by knowing, access, and cost. Know-who works as a directory of what people know, overlaid with information about how easy they are to ask.

Route 3: Referral and brokerage

Gaps in one’s directory can also be filled by borrowing someone else’s. At McDonald and Ackerman’s field site, one employee referred people seeking information to those most likely to have it. They called this role the expertise concierge and described it as a critical identification resource. Part of an organization’s know-who gathers not in each individual but in the people who serve as nodes of referral.

People who stand between groups are positioned to touch the know-who of both sides. Burt (2004) studied the networks of 673 managers who ran the supply chain of one of America’s largest electronics companies. Managers whose networks spanned structural holes (gaps between groups that are not connected to each other) were more likely to express ideas and discuss them with colleagues, have them engaged by senior management, and have them judged valuable. Burt explained this advantage as a “vision of options otherwise unseen” that comes from brokering between groups. A brokerage position makes it easier not only to collect know-who but also to turn it into new combinations.

Route 4: Traces left in artifacts

Know-who can be inferred from traces of work even without direct conversation. At McDonald and Ackerman’s site, source code change histories were used as a cue to identify who knew a module well. But change histories make it hard to tell small changes from large ones, and people who had only touched the code briefly were sometimes mistaken for experts. Artifacts mechanically record what Wegner and colleagues called “access to information,” but not how deep that access was.

Route 5: Seeing other people’s conversations

The fifth route is seeing conversations one is not part of. Leonardi (2015) started from prior research showing that people’s metaknowledge (knowing who knows what and who knows whom) tends to be confined to the people they talk with regularly. At a large financial services firm, he ran a quasi-natural experiment in which only the marketing division received an enterprise social network, compared with a matched operations division. After six months, only the marketing division that used the network improved the accuracy of knowing who knows what by 31% and who knows whom by 88%. The operations division, which did not use it, showed no significant change in either.

Leonardi called this mechanism ambient awareness. Because coworkers’ exchanges with specific partners are visible to others, observers can infer from the content and recipients of messages what their coworkers know and whom they know. If Routes 2 and 3 accumulate know-who by going out to ask, Route 5 accumulates it by overhearing without asking. It can be seen as a mechanism that makes Wegner and colleagues’ “access to information” observable at the scale of an organization.

How accurate is the directory?

Among the sources of directory entries, defaults come from categories rather than the person. This leaves room for bias to enter the directory. Joshi (2014) analyzed peer-evaluation data from more than 60 teams and more than 500 scientists and engineers in multidisciplinary research centers at a university. According to the abstract, whether expertise was recognized depended less on the gender and education of the person being evaluated than on the attributes of the evaluator and the relationship between the two. The directory of who knows what is written through the eyes of the person doing the evaluating.

Lewis’s (2003) 15-item scale has been used to measure know-who. According to the abstract, its validity was tested in 124 laboratory teams, 64 MBA consulting teams, and 27 teams from technology companies. The scale measures three things: the specialization of members’ knowledge, credibility (trust in one another’s knowledge), and coordination in using that knowledge. Ren and Argote (2011) reviewed 76 papers on transactive memory and organized its antecedents and consequences into a framework. Among future research directions they listed the dynamic evolution of transactive memory, virtual teams, and organization-level transactive memory supported by information technology.

Student teams and design teams

Student teams have short histories and weak formal structures. Hu et al. (2026) analyzed 214 undergraduate teams (852 valid responses) from universities in Sichuan Province, China, that took part in a national extracurricular science and technology competition (the Challenge Cup). Coordination was positively related to innovation performance at the 1% level, while specialization and credibility showed positive relationships only at the 10% level. The effect of specialization was stronger when credibility was high (β = 0.184, p = .009) and when coordination was high (β = 0.250, p = .002). A reading that follows is that knowing who is good at what is not enough on its own; it leads to results only when people trust it and have the arrangements to use it.

Evidence on the early stages of design is scarce. Dastmalchi et al. (2021) analyzed the ideation records of a single design team and reported that, of 48 ideas, those presented on platforms visible to everyone, such as a whiteboard, and those stored in the cloud were more likely to be selected. It is a single case, and the authors themselves state that they do not generalize. What it supports is only that whether ideas are presented where everyone can see who holds which idea may play a part in which ideas are kept or dropped.

How it is lost

Because know-who resides in people, the directory develops holes when people leave. Christian et al. (2014) had 78 four-person teams run a command-and-control simulation and removed one member without warning partway through. Teams with a well-developed transactive memory maintained higher performance after losing a member. But when the lost member was a critical one, the benefit of transactive memory shrank. Teams that lost a critical member had difficulty reformulating their plans with the remaining people. The more detailed the directory, the more the absence of the person it pointed to matters.

The structural interference of transactive memory (Route 1) is another pattern of loss. Long-standing pairs performed worse when a division of labor was imposed from outside, because it clashed with their existing one. Accumulated know-who stops working well even without membership change when the way work is divided changes from outside.

What changes when AI joins?

Studies are beginning to treat AI as a member of a transactive memory system. Bienefeld et al. (2023) analyzed a simulation in which 180 ICU physicians and nurses worked with an AI agent. Getting information from the AI agent was positively linked to generating new hypotheses and speaking up with doubts or concerns, but only in higher-performing teams. Getting information from human team members was negatively linked to both, regardless of performance. Even when AI becomes one of the candidates for whom to ask, whether that leads to good outcomes depends on the state of the team.

When AI mediates organizational memory, Route 5 (seeing conversations) can narrow. Lee et al. (2026) had 21 people in four university research labs use CHOIR, a chatbot that supports a lab’s organizational memory, for one month. Of the 95 questions students asked, only 11 were shared with other members. Because questions were asked privately, directors found it harder to see where documentation was lacking. When asking AI returns an answer, information about who is struggling with what does not get written into anyone else’s directory.

In learning settings, how much to trust AI also becomes part of the directory. Islam et al. (2026, preprint) randomly assigned 42 students in an undergraduate course to three usage conditions and deliberately had them use a weaker language model. In the condition where students thought first and then consulted AI, credibility ratings of the AI’s output fell the most. Whether AI gets written into the directory as “a partner who knows everything” or “a partner who needs checking” may change with the order in which it is used.

The five routes side by side

The routes can be distinguished by what they write into the directory.

What all the routes share is that know-who is written only through interaction with others or by observing traces of interaction. This is the property the OECD pointed to when it said know-who is hard to transfer through formal channels. Conversely, when interaction becomes private, when people leave, or when the division of work is changed from outside, updating of the directory stops or drifts.

Gaps in the literature

• Speed of directory updating: No study was found comparing how quickly each route makes the directory accurate. Leonardi (2015) is a six-month comparison, not a comparison of speed across routes.
• Visibility of conversations mediated by AI: The CHOIR results come from four labs over one month. Whether private questions to AI actually lower the accuracy of an organization’s know-who has not been measured.
• Know-who in design teams: Research on the early stages of design is limited to a single case (Dastmalchi et al. 2021). How a directory of who holds which ideas and cues affects design decisions remains open.
• Correcting biased defaults: Within the scope of this search, no longitudinal study was found tracking how far inferences from defaults (categories such as gender) are corrected by accumulated interaction.

Related Notes

• What Social Coordination Costs Arise from Collaborating with Others on Creative Work: treats transactive memory as one of the mechanisms that lower coordination costs. This note covers how that mechanism builds up and breaks down.
• Can Incompetence Be Defined as Raising Other People’s Cognitive Load?: examines whether colleagues’ self-regulation affects one’s own cognitive load, drawing on lines of research including transactive memory. An accurate directory is a condition for widening what can be left to colleagues.

Unverified items

No main claim remains unverified. Sources for which the full text could not be retrieved through three or more routes and only the abstract was confirmed (Lundvall & Johnson 1994; Liang et al. 1995; Borgatti & Cross 2003; Joshi 2014; Lewis 2003; Ren & Argote 2011) are cited within the scope of their abstracts, as stated in the text. For Wegner (1987), neither full text nor abstract could be retrieved, and only its bibliographic record is given as the origin of the concept.

References

All sources accessed 2026-09-28.

Defining know-who

• OECD (1996). The Knowledge-based Economy. OCDE/GD(96)102. Paris: OECD. https://one.oecd.org/document/OCDE/GD%2896%29102/en/pdf
• Lundvall, B.-Å., & Johnson, B. (1994). The Learning Economy. Journal of Industry Studies, 1(2), 23–42. https://doi.org/10.1080/13662719400000002
• Wegner, D. M. (1987). Transactive Memory: A Contemporary Analysis of the Group Mind. In B. Mullen & G. R. Goethals (Eds.), Theories of Group Behavior (pp. 185–208). Springer. https://doi.org/10.1007/978-1-4612-4634-3_9

Routes of accumulation

• Wegner, D. M., Erber, R., & Raymond, P. (1991). Transactive Memory in Close Relationships. Journal of Personality and Social Psychology, 61(6), 923–929. https://doi.org/10.1037/0022-3514.61.6.923
• Liang, D. W., Moreland, R., & Argote, L. (1995). Group Versus Individual Training and Group Performance: The Mediating Role of Transactive Memory. Personality and Social Psychology Bulletin, 21(4), 384–393. https://doi.org/10.1177/0146167295214009
• McDonald, D. W., & Ackerman, M. S. (1998). Just Talk to Me: A Field Study of Expertise Location. Proceedings of the 1998 ACM Conference on Computer Supported Cooperative Work (CSCW ‘98), 315–324. https://doi.org/10.1145/289444.289506
• Borgatti, S. P., & Cross, R. (2003). A Relational View of Information Seeking and Learning in Social Networks. Management Science, 49(4), 432–445. https://doi.org/10.1287/mnsc.49.4.432.14428
• Burt, R. S. (2004). Structural Holes and Good Ideas. American Journal of Sociology, 110(2), 349–399. https://doi.org/10.1086/421787
• Leonardi, P. M. (2015). Ambient Awareness and Knowledge Acquisition: Using Social Media to Learn “Who Knows What” and “Who Knows Whom”. MIS Quarterly, 39(4), 747–762. https://doi.org/10.25300/MISQ/2015/39.4.1

Accuracy and measurement

• Joshi, A. (2014). By Whom and When Is Women’s Expertise Recognized? The Interactive Effects of Gender and Education in Science and Engineering Teams. Administrative Science Quarterly, 59(2), 202–239. https://doi.org/10.1177/0001839214528331
• Lewis, K. (2003). Measuring Transactive Memory Systems in the Field: Scale Development and Validation. Journal of Applied Psychology, 88(4), 587–604. https://doi.org/10.1037/0021-9010.88.4.587
• Ren, Y., & Argote, L. (2011). Transactive Memory Systems 1985–2010: An Integrative Framework of Key Dimensions, Antecedents, and Consequences. Academy of Management Annals, 5(1), 189–229. https://doi.org/10.5465/19416520.2011.590300

Student teams and design teams

• Hu, P., Hu, D., She, M., & Li, Z. (2026). The Impact of Transactive Memory Systems on Innovation Performance in Undergraduate Teams. Frontiers in Psychology, 17, 1720641. https://doi.org/10.3389/fpsyg.2026.1720641
• Dastmalchi, M. R., Balakrishnan, B., & Oprean, D. (2021). Exploring the Role of Transactive Memory Systems in Team Decision-Making during Ideation Phase. Proceedings of the Design Society, 1 (ICED21), 1529–1536. https://doi.org/10.1017/pds.2021.414

How it is lost

• Christian, J. S., Pearsall, M. J., Christian, M. S., & Ellis, A. P. J. (2014). Exploring the Benefits and Boundaries of Transactive Memory Systems in Adapting to Team Member Loss. Group Dynamics: Theory, Research, and Practice, 18(1), 69–86. https://doi.org/10.1037/a0035161

Collaborating with AI

• Bienefeld, N., Kolbe, M., Camen, G., Huser, D., & Buehler, P. K. (2023). Human-AI Teaming: Leveraging Transactive Memory and Speaking Up for Enhanced Team Effectiveness. Frontiers in Psychology, 14, 1208019. https://doi.org/10.3389/fpsyg.2023.1208019
• Lee, S., Abbas, A., Chen, Y., Kim, Y.-H., & Lee, S. W. (2026). CHOIR: A Chatbot-mediated Organizational Memory Leveraging Communication in University Research Labs. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, 1–23. https://doi.org/10.1145/3772318.3791314 (author version https://arxiv.org/abs/2509.20512)
• Islam, M. T., Akgun, M., & Billah, S. (2026). Shaping Credibility Judgments in Human–GenAI Partnership via Weaker LLMs: A Transactive Memory Perspective on AI Literacy. arXiv preprint (peer-reviewed version not confirmed). https://arxiv.org/abs/2603.26522

Footnotes

1. The opening paragraph of Wegner, Erber & Raymond (1991). The same paragraph notes that “such knowledge of one another’s memory areas takes time and practice to develop.” ↩