Conference PaperPDF Available

Knowledge Broker Bots in Enterprise Social Media: An Exploratory Study

Authors:

Abstract

Enterprise social media (ESM) platforms are a central hub for team collaboration. While they can effectively facilitate communication among distributed individuals and teams, promoting knowledge sharing remains a major challenge. ESM users are often unaware of others’ knowledge and therefore are unable to seek experts or share knowledge with those who need it. A potential solution could be the use of knowledge broker bots that automatically connect knowledge seekers with knowledge providers to facilitate knowledge sharing. However, given the focus of existing literature on the human element of knowledge brokering, our understanding of the use and impact of such bots on knowledge sharing in ESM is limited. Therefore, we conducted a two-month, exploratory study with five student teams on Slack. Our findings provide initial insights into how users interact with a knowledge broker bot and how the bot establishes connections between users as a critical conduit to successful knowledge brokering.
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 1
Knowledge Broker Bots in Enterprise Social Media:
An Exploratory Study
Ulrich Gnewuch
Karlsruhe Institute of Technology
ulrich.gnewuch@kit.edu
Wietske Van Osch
HEC Montréal
Michigan State University
wietske.van-osch@hec.ca
Constantinos K. Coursaris
HEC Montréal
constantinos.coursaris@hec.ca
ABSTRACT
Enterprise social media (ESM) platforms are a central hub
for team collaboration. While they can effectively facilitate
communication among distributed individuals and teams,
promoting knowledge sharing remains a major challenge.
ESM users are often unaware of others knowledge and
therefore are unable to seek experts or share knowledge with
those who need it. A potential solution could be the use of
knowledge broker bots that automatically connect
knowledge seekers with knowledge providers to facilitate
knowledge sharing. However, given the focus of existing
literature on the human element of knowledge brokering, our
understanding of the use and impact of such bots on
knowledge sharing in ESM is limited. Therefore, we
conducted a two-month, exploratory study with five student
teams on Slack. Our findings provide initial insights into
how users interact with a knowledge broker bot and how the
bot establishes connections between users as a critical
conduit to successful knowledge brokering.
Keywords
Bots, enterprise social media, knowledge brokering, team
collaboration, exploratory study, Slack.
INTRODUCTION
Enterprise social media (ESM) platforms, such as Slack or
Microsoft Teams, have become a central hub for virtual team
collaboration in both organizational and educational
settings. While such platforms have been found to facilitate
day-to-day communication among distributed individuals
and teams, promoting other important collaboration
activities, such as knowledge sharing, remains a major
challenge. A key barrier to knowledge sharing is that people
often do not know what others know and therefore are
unable to seek experts or share knowledge with those who
need it (Leonardi, 2014). Although ESM platforms provide
users with some visibility into others’ skills and areas of
expertise (e.g., based on their profiles as well as messages in
public channels), manually searching for, identifying, and
contacting potential experts who might share their
knowledge is tedious (Leonardi, 2014).
Since bots (short for software robots) have become an
integral part of many ESM platforms in recent years, it may
be possible to address this challenge by using bots to
automatically connect ESM users who seek knowledge with
other users who possess the requisite knowledge. These bots,
hereafter referred to as knowledge broker bots, are designed
to establish connections between knowledge seekers and
knowledge providers with the aim of facilitating knowledge
sharing. In contrast to knowledge distributors and
knowledge integrators, the role of a knowledge broker is not
to create, modify, or disseminate knowledge (Drew et al.,
2014). Therefore, unlike other bots in social networks or
online communities, knowledge broker bots do not directly
engage in conversations or transfer information between
users, but rather establish a connection between those who
seek and those who possess the requisite knowledge.
Although several knowledge broker bots were developed for
major ESM platforms in recent years (e.g., Slack’s
“Whocan” or WhoQui, Microsoft Teams’ “Who”), our
understanding of their use and impact is limited. While there
is a growing body of research on bots in social networks
(e.g., Salge et al., 2022), online communities (e.g., Safadi et
al., 2021), and open source software projects (e.g., Hukal et
al., 2019), little attention has been paid to bots designed to
support team collaboration in ESM (Seering et al., 2019).
At the same time, the ESM literature has explored the topic
of boundary spanning (i.e., establishing and maintaining
communication links to external resources) and knowledge
brokering (Van Osch & Steinfield, 2018). However, the
focus has been on the role of inherent affordances of ESM
such as visibilityon boundary spanning, where boundary
spanning is viewed as an intrinsic human activity. The focus
on knowledge broker bots shifts the perspective on boundary
spanning from a human process facilitated by technology to
a process that is inherently sociotechnical.
Against this backdrop, the objective of this exploratory
research is to investigate the use and impact of knowledge
broker bots in ESM and expand our understanding of
knowledge brokering as a sociotechnical process. More
specifically, we address the following research questions: (1)
How do users interact with a knowledge broker bot when
they seek knowledge or when they are contacted by the bot
to provide knowledge to others? (2) How can a knowledge
broker bot establish new connections between knowledge
seekers and providers to facilitate knowledge sharing?
To address these questions, we conducted an exploratory
study in the context of a two-month student team project
with 21 users on Slack. Based on the analysis of digital traces
of user interactions with the bot, communication behaviors
on Slack, and a posteriori qualitative feedback, we provide
initial insights into how users interact with a knowledge
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 2
broker bot and how the bot establishes connections between
users as a critical conduit to successful knowledge
brokering. While our preliminary findings suggest that the
bot was successful in connecting knowledge seekers with
knowledge providers, we also identified several challenges
and opportunities related to the use and the design of
knowledge broker bots in ESM. With our exploratory
findings, we aim to contribute to IS research on human-bot
interaction by generating a deeper understanding of the use
and impact of knowledge broker bots. Furthermore,
disentangling the role of knowledge broker bots helps to
extend the ESM literature on knowledge brokering and
boundary spanning by adding a fundamental sociotechnical
perspective. For practitioners, our study highlights
opportunities and challenges of using bots to facilitate
knowledge sharing among distributed individuals and teams.
RELATED WORK AND THEORETICAL FOUNDATION
Bots
Software robots, or bots for short, are fully automated
software programs that perform a variety of tasks on behalf
of their developers and users (Safadi et al., 2021). They are
omnipresent in social networks (e.g., Twitter), online
communities (e.g., Reddit), customer service, and open
source software projects (e.g., on GitHub) (Gnewuch et al.,
2022; Hukal et al., 2019; Salge et al., 2022). Bots can be
viewed as a class of agentic IS artifacts as they can react to
certain stimuli or action triggers (e.g., a new tweet or an
update in a GitHub repository) and carry out actions
autonomously (Baird & Maruping, 2021; Salge et al., 2022).
In recent years, bots have also become an integral part of
ESM platforms such as Slack or Microsoft Teams. Existing
bots are primarily designed to automate repetitive tasks (e.g.,
meeting organization), provide real-time information (e.g.,
notifications about GitHub activities), and facilitate
communication and collaboration among individuals and
teams (e.g., onboarding, project management). In contrast to
bots for dyadic, one-on-one interactions or broadcasting bots
in social networks, these bots are primarily designed to act
as non-human community members that support
collaboration between the other (human) members of a
community or team (Seering et al., 2019).
Enterprise Social Media
Enterprise social media (ESM) are web-based platforms that
enable users to effectively communicate with each other,
network, organize, leverage information available on the
platform, and collaborate (Leonardi et al., 2013). Most
organizations use some form of ESM and the number of
ESM users has drastically increased during the COVID-19
pandemic as employees were forced to work from home, a
work practice that is forecasted to continue to a large extent
even after the pandemic. Further, ESM platforms are
increasingly used in educational settings to facilitate
communication and collaboration among students.
Users of ESM platforms are able to communicate with other
users through text-based synchronous and asynchronous
communication. The communication can take place in
private chat rooms or in public spaces, often called channels.
Users can create and join channels as well as contribute to
and consume content. Channels can be open to the entire
organization or closed (i.e., involve only invited participants,
such members of a specific unit, team, or project) (Van Osch
& Steinfield, 2018).
ESM platforms aim to create an environment where users
can effectively share knowledge. By participating in ESM,
users can learn at least two kinds of knowledge (Leonardi et
al., 2013): instrumental knowledge (i.e., knowledge about
how to do something) and metaknowledge (i.e., knowledge
about who knows what and who knows whom).
Metaknowledge is crucial because it is an antecedent to the
transfer of instrumental knowledge (Leonardi et al., 2013).
In other words, before users can acquire instrumental
knowledge from others, they need to know where that
knowledge can be found (metaknowledge). However, users
often do not know what others know and therefore are
unable to find experts with the requisite knowledge
(Leonardi, 2014). Although ESM helps users acquire
metaknowledge (e.g., based on reading what others post or
comment), manually searching for, identifying, and
contacting potential experts who might share their
knowledge is tedious (Leonardi, 2014).
Knowledge Brokering and Boundary Spanning
Knowledge brokering can be understood as the process of
connecting knowledge seekers with knowledge providers
(Haas, 2015; Hargadon, 2002). In contrast to knowledge
distributors and knowledge integrators, knowledge brokers
do not create, modify, or disseminate knowledge themselves
(Drew et al., 2014). Instead, knowledge brokers establish
connections between seekers and providers of knowledge in
order to facilitate the flow of information from those who
possess knowledge to those who need it (Drew et al., 2014).
The concept of knowledge brokering is closely related to that
of boundary spanning, which refers to establishing and
maintaining communication links to external resources
(Tushman & Scanlan, 1981). Some studies have suggested
that the critical difference between knowledge brokers and
boundary spanners is the fact that the former focus on
knowledge sharing between disconnected individuals,
whereas boundary spanners aim to connect individuals or
teams to external knowledge (Haas, 2015). Nonetheless,
both refer to the fundamental process of connecting to
critical resources that exist outside the immediate network
(whether this is composed of individuals or teams/groups).
Furthermore, similar to boundary spanners, knowledge
brokers need to acquire metaknowledge of who knows what
before they are able to make meaningful connections.
In the past, knowledge brokering has been a central human
activity. Many organizations even have formal knowledge
broker roles for people whose job is specifically to know
what others know and help make connections (Hargadon,
2002). Similarly, the literature on boundary spanning in
ESM has explored various boundary-spanning activities
representation, information search, and coordinationas
inherent human processes occurring in the context of ESM.
Furthermore, it has explored the role of ESM affordances,
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 3
such as visibility, in facilitating these human processes of
boundary spanning (Van Osch & Steinfield, 2018), but a
fundamental sociotechnical understanding is lacking.
Recent developments in artificial intelligence (AI) have
enabled technology to take on the role of a knowledge
broker, for example, in the form of knowledge broker bots in
ESM. In the context of this study, we define knowledge
broker bots as fully automated software programs that
acquire metaknowledge of who knows what and connect
knowledge seekers with knowledge providers to facilitate
the flow of information from those with knowledge to those
who need it for a particular purpose. Figure 1 illustrates the
concept of a knowledge broker bot, and this study focuses
on exploring its role in transforming knowledge brokering
into a fundamental sociotechnical process.
Figure 1. Knowledge Broker Bot
METHOD
Research Context
Our exploratory study was carried out at a large European
university in the context of a master’s human-computer
interaction (HCI) course attended by 21 students with a
background in information systems and industrial
engineering and management. In the course, students were
assigned to one of five teams and completed a two-month
design project in partnership with a multinational European
energy provider. The overall goal and topic of the project
was to create innovative design solutions for supporting the
human resources lifecycle of remote employees. Each team
worked on a different phase of the lifecycle (e.g., attracting
and hiring, onboarding, working remotely, offboarding).
Students were assigned to a team based on their individual
background, skills, and interests. As two students dropped
out in the first week after the project started, one team had
three members, while the other teams comprised four to five
members. During the project, there were bi-weekly meetings
with the instructor team and two employees of the partner
company. The teams held three presentations over the course
of the project to collect feedback. All official meetings
except for the final presentation session took place virtually.
Slack was the primary platform for team communication and
collaboration used in the project. The use of Slack was
mandatory for students and they were informed that all
project-related communication had to take place in Slack
and not via email or other platforms. We created public
channels for each team (i.e., visible to everyone), a general
1
https://slack.com/apps/A019V8JUJ9X-whocan
channel primarily used for messages from the instructor
team, and a specific questions channel where students could
post and answer questions. We also invited the employees of
the partner company to join our Slack workspace so that they
could be contacted directly inside the platform. Finally, our
Slack workspace included a knowledge broker bot (see
below). At the beginning of the project, we explained the
bot’s features to students and encouraged them to use the bot
for seeking help from others for a specific task.
Knowledge Broker Bot “Whocan”
To select a knowledge broker bot for use in our study, we
carefully reviewed all existing bots on Slack. We identified
only two bots that matched our definition of a knowledge
broker bot, namely “Whocan”
1
and “WhoQui”
2
. After
testing and analyzing both bots, we decided to use Whocan
because it offered a richer set of features related to the role
of a knowledge broker. We contacted the developers of
Whocan, explained the purpose of our study, and they agreed
to share log data of bot interactions with us after the project.
Consistent with our definition of a knowledge broker bot,
Whocan is described as a bot that helps users find another
user with certain knowledge (called skill) by asking around
for them. It also “keeps track of who is good at what and
connects experts to those who need them. Whocan has two
main features. First, for Whocan to be able to identify
knowledge providers (called experts), users can use the
command /skills [skill1, skill2, …] to manually set and
update their areas of knowledge. Whocan stores this
information, which is considered its metaknowledge (i.e.,
who knows what), in a database. Second, users who need
help can make a knowledge request to Whocan using the
command /whocan [request] feature (see Figure 2). In
addition, Whocan actively monitors conversations in public
Slack channels and reacts to posted questions by offering its
help via private chat to the user who posted the question.
Figure 2. Knowledge Request from a User to Whocan
When Whocan receives or detects a knowledge request, it
browses its database (i.e., its metaknowledge) and sends out
help requests to five users. To identify potential knowledge
providers, Whocan uses a keyword matching algorithm that
compares the content of the request with the skills in its
database. If no matches are found, Whocan randomly selects
five users to contact each day or until the request is cancelled
by the user. Potential knowledge providers are contacted via
private chat using the message On behalf of a team member,
I’m looking for someone who can [request]. Can you please
help out or know who can? (see Figure 3). Contacted users
2
https://slack.com/apps/AUBPY2EAC-whoqui
Knowledge
Seeker
Knowledge
Provider(s)
?
Knowledge Sharing
Knowledge
Broker Bot
New Connection
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 4
can either agree (“Yes, I can help”) or disagree to help (“No,
sorry”), forward the bot to someone else (“I know who can
help”), or snooze the bot. When a user agrees to help,
Whocan connects both users by opening a new group chat
with both of them and repeats the knowledge request to start
the conversation (see Figure 4).
Figure 3. Help Request from Whocan to Potential Experts
If contacted users forward Whocan to someone else, a
window opens where the user can select another user in the
Slack workspace who is then contacted by the bot as well.
Finally, after two users have been connected in a new group
chat, the user who made the knowledge request can rate the
expertise of the knowledge provider. This information is
used to update Whocan’s database and therefore taken into
account when selecting potential knowledge providers for
similar requests in the future. Over time, Whocan is
therefore able to continuously improve its metaknowledge.
Participants
Participants in our study were the 21 students who attended
the master’s HCI course. Participants were mostly male
(85.7%) with a mean age of 23.8 years. Two-thirds studied
information systems, while the others studied industrial
engineering and management. Most participants were in
their second year of study (M = 2.35) and all of them had
some experience in the areas of software development, UX
design, and user research. We explained the study procedure
to all participants and assured them that their data would be
anonymized for the analysis, treated confidentially, and not
used for grading purposes. All participants provided their
informed consent before participating in the study.
Data Collection and Analysis
In our study, we collected, preprocessed, and analyzed
different, complementary types of data. First, we collected
data on user interactions with the bot (e.g., making
knowledge requests, reacting to help requests from the bot)
as well as on the bot’s activity in the background (e.g.,
updating skills in the database). After the project, the
developers of the bot exported this data for us. The data was
in JSON format and included digital traces of all bot-related
activities. We preprocessed this dataset in order to build two
clean datasets in CSV format, one for skill updates and one
for knowledge requests. Second, we collected all messages
from both public channels and private conversations on
Slack. After the project, we exported this data from our Slack
workspace. The export came in the form of a zip file
containing many directories (channels and conversations)
and JSON files with the message history broken into dates.
We collated the exported message histories for each channel
or conversation, extracted the message content and
metadata, and converted them into a clean CSV format for
further analysis. In the final step, we linked this dataset to
the bot dataset and anonymized the users’ identity by
replacing their original usernames with randomly generated
ids. Third, we also collected participants’ feedback on the
bot and their interaction with it in a short survey with open-
ended questions at the end of the project.
Given the exploratory nature of our study, our analysis
focused on providing initial insights into the use and impact
of the knowledge broker bot in the ESM platform Slack.
Based on the data of user interactions with the bot,
communication behavior on Slack, and qualitative feedback
from users, we performed the following analyses. First, we
explored if and how users contributed information about
their own knowledge areas to the bot’s metaknowledge (who
knows what) and rated the expertise of others after they were
connected through the bot. Second, we analyzed how users
interacted with the bot when seeking knowledge or when
contacted by the bot to help out others. Third, we examined
if the bot was able to establish new connections between
users, as a potential indicator of its ability to facilitate
knowledge sharing. In each step of the analysis, we
complemented our results with qualitative insights from a
content analysis of users’ feedback in the survey.
PRELIMINARY FINDINGS
Throughout the project, all teams used our ESM platform
Slack to communicate and collaborate. In total, 1,491
messages were sent over the two-month period, suggesting
that a great deal of project-related communication (e.g.,
about developing prototypes, preparing presentations,
organizing meetings) took place inside the platform. Most
messages were sent in the five team channels. Although
everyone, including members from other teams, could see
these messages, we observed that users rarely interacted with
others outside their own team in public. For example, the
“questions channel was never used, suggesting that users
preferred private or intra-team communication when asking
questions. While all teams regularly posted their
presentations in the general channel (as we had asked them
to do), we did not observe users or teams who actively shared
their knowledge with others (e.g., by posting the results of
their interviews with employees of the partner company).
However, our analysis indicated that many users interacted
with the knowledge broker bot during the two months. These
user interactions include 207 manual updates of knowledge
areas (“skills”), 83 knowledge requests to the bot, and 40
successful connections between knowledge seekers and
providers. In the following, we present the results of our
analysis of these interactions and the bot’s overall impact.
Contributing and Acquiring Metaknowledge
To be able to connect knowledge seekers and knowledge
providers in ESM, a knowledge broker bot needs to acquire
metaknowledge of who knows what. Ideally, the bot would
automatically extract users’ areas of knowledge from
existing conversations, profile pages, or other organizational
databases. However, since our project started from scratch,
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 5
there was no existing data to feed the bot and therefore we
asked all users at the beginning of the project to use the
/skill”-command to manually provide information about
their knowledge areas to the bot.
Our first analysis shows that all 21 users manually entered
information about their knowledge areas to help the bot
acquire metaknowledge. Their input ranged from single
words (e.g., “R”, “excel”, “surveys”) to full sentences (e.g.,
i can develop your backend structure”). Frequently
mentioned knowledge areas were programming languages
(e.g., Python, JavaScript), tools (PowerPoint, Figma),
methods (e.g., Scrum, project management), and topics (e.g.,
entrepreneurship, security). In total, users manually
provided information about 207 knowledge areas to the bot
(M = 9.86 per user). Most knowledge areas (71%) were
entered in the first two weeks of the project (i.e., during the
initial 25% of the project timeline); only a few users
manually updated them later during the project. The number
of unique knowledge areas entered was 112 and on average,
a particular area was entered by 1.84 users showing the
relative uniqueness and complementarity of knowledge
areas. Thirty-seven knowledge areas were entered more than
once, with some popular ones entered more than ten times
(e.g., Python). In general, these results suggest that users
were willing to adopt the knowledge broker bot and help it
learn about them in order to build up its metaknowledge.
However, the results also indicate some overlap of
knowledge areas between users, at least initially, which
could be a result of the relative homogeneity of our student
participants.
In addition to the knowledge areas manually provided by
users, the bot also updated its metaknowledge automatically
when a knowledge seeker positively evaluated a knowledge
provider after they had been connected by the bot. In total,
there were 210 automatic updates of knowledge areas for 17
users, including 84 unique areas. Out of these 84 knowledge
areas, 48 areas were new because they had not been entered
manually before. This result suggests that the bot was able
to increase its metaknowledge ‘volume’ (by 101%; based on
210 updates to the existing 207) and breadth (by 43%;
based on 48 new additions to the 112 unique areas) over time
by learning from successful connections.
Knowledge Requests to the Bot
In our second analysis, we examined how users interacted
with the bot when they sought knowledge. During the two-
month period, users made a total of 83 knowledge requests
to the bot, corresponding to ~4 requests per user and 1
request per user per fortnight. These requests were usually
short sentences or questions with an average of 13.22 words.
As the following examples show, users knowledge requests
varied in their level of detail and degree of specificity:
Who can set up a database in AWS?
Who can set up Prometheus and Grafana or similar
services to monitor our deployed Kubernetes service?
I need someone who can analyze surveys
The feedback provided by users indicated that they liked the
simplicity of making knowledge requests. They generally
found it very easy to post Whocan requestsand mentioned
that “requests can be created quickly and easily”. However,
some users made multiple requests in a short time span and
were subsequently blocked by the bot’s spam protection: I
asked several questions but was blocked after the first one
for half an hour. Taken together, these results suggest that
the bot was easy to use, but sometimes wrongly interpreted
messages as spam resulting in user frustration and lost time.
While mechanisms need to be in place to prevent spam,
classifying whether or not a request is spam is not a trivial
task and false positives can lead to frustration among users.
Bot Help Requests to Potential Knowledge Providers
In our third analysis, we examined how users interacted with
the bot when they were identified as a potential knowledge
provider and contacted by the bot to help out another user.
To recap, after receiving a knowledge request, the bot asked
around on the ESM platform by sending out help requests to
potential knowledge providers (see Figure 3). Based on the
83 knowledge requests made by users in the two-month
period, the bot sent out a total of 994 help requests (M =
11.97 per knowledge request). As explained earlier, the bot
used a keyword matching algorithm to identify potential
knowledge providers by comparing the content of the
knowledge request with the list of knowledge areas stored in
its database. If no matches were found, the bot randomly
selected five users to contact each day or until the request
was cancelled by the user. In total, 40 requests (48.19%)
were successful in that a potential knowledge provider who
was contacted by the bot agreed to help the knowledge
seeker with her or his request. The remaining requests were
unsuccessful either because no one agreed to help (22.89%),
they were cancelled by the knowledge seeker before
someone was found (13.25%), or blocked as spam (15.66%).
On average, it took 11.37 bot requests and approximately
two days for a successful request to be answered. Further,
we found that seven requests were forwarded by a user to
someone else. Four of these requests were then answered by
the referred user (57.14%).
In general, the majority of user feedback about the bot’s
ability to connect knowledge seekers and providers was
positive. Most users felt that the bot was quite fast in
connecting [them] with experts”. Some only needed 4-5
hours to find the first expert who [they were] looking for.
However, consistent with our quantitative results above,
some users complained that their knowledge requests had
not been successful. One user stated that half of [his]
searches were never answered by any of the other students
and suggested that the bot should give more incentive to
answer other people’s searches”. These results suggest that
the bot was able to connect knowledge seekers and providers
for many, but not all, knowledge requests. There are several
possible explanations of why some knowledge requests were
unsuccessful. One explanation could be that those who were
contacted did not possess the requisite knowledge or did not
want to help, perhaps due to a request’s (poor) timing, a
recipient’s lack of time or, ultimately, motivation. As the
examples above show, some requests were quite unspecific
and may have inhibited others from agreeing to help. Finally,
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 6
it could be that the bot did not contact a user with the
requisite knowledge before the request was cancelled.
Users also commented on the volume of the bot’s help
requests. One user wondered why he was never contacted by
the bot: “There were no suggestions for me, whom I could
help.[…] What for did I enter my skills in the beginning?”.
However, other users complained about being spammed by
the bot: I got too many notifications from the bot, which is
annoying”. These results highlight the importance of the
bot’s ability to identify and select potential knowledge
providers. Ideally, only those users who possess the requisite
knowledge would be contacted. However, randomly
contacting users is not an entirely bad idea because users
might have acquired new knowledge recently or will
forward the bot to someone else who might be able to help.
It all depends on the volume of the bot’s help requests
because users will become annoyed at some point when they
are constantly contacted by the bot with requests that may
not be relevant to their expertise. Given that we only had 21
users on our platform, many were contacted multiple times.
However, nobody blocked or “snoozed” the bot, even
though this was possible. Again, an improved identification
of users’ knowledge areas based on inferences from existing
knowledge areas or monitoring public conversations, as well
as an intended ‘load management’ of requests across
available users, might help to reduce the number of unrelated
requests and keep users motivated to interact with the bot.
Establishing Connections between Knowledge Seekers
and Knowledge Providers
In our final analysis, we specifically examined the set of
successful knowledge requests where the bot was able to
establish a connection between a knowledge seeker and
provider. We initially planned to focus our analysis on the
conversations between knowledge seekers and providers in
the group chat opened by the bot, but we realized that most
conversations were rather short as users often set up a video
call or exchanged phone numbers to discuss the problem at
hand using a different communication channel. Therefore,
we could not assess whether and how much knowledge was
actually shared between the two users. However, as a
potential indicator for the value of the connections made by
the bot, we analyzed how knowledge seekers rated
knowledge providers. These ratings were requested by the
bot two days after it established the connection. Our analysis
shows that in 32 of the 40 successful requests, the knowledge
seeker confirmed the knowledge provider’s expertise in at
least one knowledge area. This result suggests that most
users (i.e., 80%) were able to get help through the
connections established by the bot and that the bot was able
to facilitate knowledge sharing among users.
In addition, we analyzed the bot’s impact on connections
between users on the ESM platform. Specifically, we
examined whether the bot was able to establish connections
between users who had not been in contact before (e.g.,
because they were members of different teams), as a
potential indicator of its ability to facilitate knowledge
sharing outside users’ existing network of peers. To do so,
we visualized our data in a simple graph, in which nodes
represent users and edges represent connections between
them. The left graph in Figure 6 displays intra-team
connections between members of the same team (black) and
cross-team connections between users who had a private
conversation or participated in the same group chat (gray).
The graph shows that while two teams appear well-
connected (team 1 and 3) and one member of team 4 engaged
with several other teams, there was not much contact across
teams in general, with only 15 cross-team connections in
total. In comparison, the right graph in Figure 6 displays
intra-team connections (black) and new cross-team
connections made by the bot through a successful
knowledge request (red). This graph shows that the bot
generated 38 new cross-team connections as opposed to the
only 15 cross-team connections that already existed (i.e., an
increase by 253%), suggesting that the bot was able to
facilitate exchangesi.e., through brokeragebetween
members of different teams. In summary, these results
suggest that the connections through the bot were quite
helpful because they connected otherwise unconnected
members from different teams who might have never
interacted with one another if not for the bot.
Figure 6. Network of Existing Connections between Users (left)
and New Connections established by the Bot (right)
DISCUSSION
In this exploratory research, we investigated the use and
impact of a knowledge broker bot in ESM. The preliminary
findings of our two-month study with five student teams
suggest that the bot was successful in connecting knowledge
seekers with knowledge providers on Slack. While our data
does not allow us to evaluate how much knowledge was
shared, there is some indication that the connections
established by the bot helped knowledge seekers find
someone with the requisite expertise and facilitated
collaboration between members of different teams. In
addition, our findings provide initial insights into how users
interact with a knowledge broker bot when they contribute
information about their own knowledge areas to the bot’s
metaknowledge, when they formulate knowledge requests,
and when they are contacted by the bot to help out others.
With our exploratory findings, we add to ESM literature and
the emerging stream of IS research on human-bot interaction
by revealing opportunities and challenges related to the use
of knowledge broker bots in ESM. First, extant research
shows that acquiring and updating metaknowledge of who
knows what in an organization is a major challenge for
humans (Leonardi, 2014). Our findings suggest that
knowledge broker bots face a similar challenge, particularly
when they are introduced in a new environment without
existing data to use for “mining” users’ areas of knowledge
(e.g., profile pages or existing conversations). Although we
Gnewuch et al. Knowledge Broker Bots in Enterprise Social Media
Proceedings of the Twenty-First Annual Pre-ICIS Workshop on HCI Research in MIS, Copenhagen, Denmark, December 11, 2022 7
find that users are willing to actively contribute to the bot’s
metaknowledge, this willingness decreases over time and
probably some kind of recurring reminder would be
necessary (e.g., the bot could ask users from time to time if
their knowledge areas have changed). Moreover, ESM
literature suggests that proclaimed expertise is a worse
indicator of actual expertise than the content of users’
messages (Leonardi, 2015). Thus, the bot’s ability to learn
from existing conversations and the successful connections
it made between knowledge seekers and providers is crucial,
not only to update its metaknowledge but also to avoid
spamming users with knowledge requests that may not
actually be adequately fulfilled considering their expertise.
This unique ability of the bot has the potential to improve
opportunities for successful knowledge brokering by relying
on more “objective” indicators of expertise. Another
promising opportunity could be to give the bot access to
messages in private conversations or channels rather than
focusing on public spaces to identify experts. Existing
research has shown that knowledge creation and sharing
often happens in private spaces (Van Osch & Bulgurcu,
2020). Although ESM are mostly implemented with the
purpose of facilitating unlimited knowledge sharing, if
knowledge and expertise largely reside in private spaces, it
inherently inhibits access from those outside these spaces.
Opening them to other users would undermine the inherent
benefits of privacy for knowledge sharing. Thus, knowledge
broker bots could be an optimal tool for identifying relevant
expertise in private spaces without revealing confidential
information to outsiders or undermining privacy.
Our study has two main limitations. First, our study was
conducted in the context of a student team project. Although
students worked together on a real-world, two-month project
in collaboration with a company, this environment is not
fully representative of the actual working environment in an
organization. Another limitation is that our data did not
allow us to analyze how much knowledge was shared after
the bot had established a connection between a knowledge
seeker and provider due to the fact that users typically
leveraged other communication modalities for the actual
knowledge exchange. Future research could combine the
types of data that we used with other data collection
approaches (e.g., surveys) to gain a more holistic
understanding of the impact of a knowledge broker bot.
ACKNOWLEDGEMENTS
The authors thank all study participants, the employees of
the partner company, and the developers of the Whocan bot
for their support. This work was supported by a Mitacs
Globalink Research Award.
REFERENCES
1. Baird, A., & Maruping, L. M. (2021). The Next
Generation of Research on IS Use: A Theoretical
Framework of Delegation to and from Agentic IS
Artifacts. MIS Quarterly, 45(1), 315341.
2. Drew, H., Ritchie, F., & King, A. (2014). How do
knowledge brokers work? Implications for policy and
practice in the case of WERS. International Journal of
Technology Management & Sustainable Development,
13(3), 205218.
3. Gnewuch, U., Morana, S., Adam, M. T. P., & Maedche,
A. (2022). Opposing Effects of Response Time in
HumanChatbot Interaction. Business & Information
Systems Engineering, 139.
4. Haas, A. (2015). Crowding at the frontier: boundary
spanners, gatekeepers and knowledge brokers. Journal
of Knowledge Management, 19(5), 10291047.
5. Hargadon, A. B. (2002). Brokering knowledge: Linking
learning and innovation. Research in Organizational
Behavior, 24, 4185.
6. Hukal, P., Berente, N., Germonprez, M., & Schecter, A.
(2019). Bots Coordinating Work in Open Source
Software Projects. Computer, 52(9), 5260.
7. Leonardi, P. M. (2014). Social Media, Knowledge
Sharing, and Innovation: Toward a Theory of
Communication Visibility. Information Systems
Research, 25(4), 796816.
8. 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), 747762.
9. Leonardi, P. M., Huysman, M., & Steinfield, C. (2013).
Enterprise Social Media: Definition, History, and
Prospects for the Study of Social Technologies in
Organizations. Journal of Computer-Mediated
Communication, 19(1), 119.
10. Safadi, H., Lalor, J. P., & Berente, N. (2021). The Effect
of Bots on Human Interaction in Online Communities.
Proceedings of the 42nd International Conference on
Information Systems (ICIS 2021).
11. Salge, C. A. de L., Karahanna, E., & Thatcher, J. B.
(2022). Algorithmic Processes of Social Alertness and
Social Transmission: How Bots Disseminate
Information on Twitter. MIS Quarterly, 46(1), 229260.
12. Seering, J., Luria, M., Kaufman, G., & Hammer, J.
(2019). Beyond Dyadic Interactions: Considering
Chatbots as Community Members. Proceedings of the
2019 CHI Conference on Human Factors in Computing
Systems, 113.
13. Tushman, M. L., & Scanlan, T. J. (1981). Boundary
Spanning Individuals: Their Role in Information
Transfer and Their Antecedents. Academy of
Management Journal, 24(2), 289305.
14. Van Osch, W., & Bulgurcu, B. (2020). Idea Generation
in Enterprise Social Media: Open versus Closed Groups
and Their Network Structures. Journal of Management
Information Systems, 37(4), 904932.
15. Van Osch, W., & Steinfield, C. W. (2018). Strategic
Visibility in Enterprise Social Media: Implications for
Network Formation and Boundary Spanning. Journal of
Management Information Systems, 35(2), 647682.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Research has shown that employing social cues (e.g., name, human-like avatar) in chatbot design enhances users’ social presence perceptions and their chatbot usage intentions. However, the picture is less clear for the social cue of chatbot response time. While some researchers argue that instant responses make chatbots appear unhuman-like, others suggest that delayed responses are perceived less positively. Drawing on social response theory and expectancy violations theory, this study investigates whether users’ prior experience with chatbots clarifies the inconsistencies in the literature. In a lab experiment ( N = 202), participants interacted with a chatbot that responded either instantly or with a delay. The results reveal that a delayed response time has opposing effects on social presence and usage intentions and shed light on the differences between novice users and experienced users – that is, those who have not interacted with a chatbot before vs. those who have. This study contributes to information systems literature by identifying prior experience as a key moderating factor that shapes users’ social responses to chatbots and by reconciling inconsistencies in the literature regarding the role of chatbot response time. For practitioners, this study points out a drawback of the widely adopted “one-design-fits-all” approach to chatbot design.
Article
Full-text available
Social network theory has produced conflicting results regarding the link between different social network structures-bridging versus bonding-and idea generation. To address this conundrum, we conduct a naturally occurring quasi-experiment of 126 open and 108 closed groups within an Enterprise Social Media (ESM) system of a multinational enterprise. Our findings show that idea generation occurs when the type of social network structure-bridging or bonding is matched to a group's openness or closedness, respectively. We further show that the reverse is counterproductive: when closed groups display bridging ties and open groups display bonding ties, idea generation is significantly undermined. Theoretically, these findings clarify the conditions and mechanisms by which both bridging and bonding can result in idea generation and provide a deeper understanding of the use of ESM for idea generation. Practically, our findings provide valuable and actionable insights regarding the use of ESM for idea generation in groups.
Conference Paper
Full-text available
Chatbots have grown as a space for research and development in recent years due both to the realization of their commercial potential and to advancements in language processing that have facilitated more natural conversations. However, nearly all chatbots to date have been designed for dyadic, one-on-one communication with users. In this paper we present a comprehensive review of research on chatbots supplemented by a review of commercial and independent chatbots. We argue that chatbots' social roles and conversational capabilities beyond dyadic interactions have been underexplored, and that expansion into this design space could support richer social interactions in online communities and help address the longstanding challenges of maintaining, moderating, and growing these communities. In order to identify opportunities beyond dyadic interactions, we used research-through-design methods to generate more than 400 concepts for new social chatbots, and we present seven categories that emerged from analysis of these ideas.
Article
Full-text available
Effective workgroups engage in team boundary spanning, that is, using communication ties as conduits to critical external resources. The proliferation of enterprise social media (ESM) and the associated increase in visibility of people, content, and interactions, has resulted in a widespread assumption that unlimited visibility improves boundary spanning. Consequently, the ESM literature has generally ignored the sentry functions of teams and failed to examine the possible strategic nature of visibility choices by ESM groups. Using log and content data from 655 ESM-based workgroups at a multinational enterprise, we contribute a deeper understanding of the distinct ways that ESM visibility—bounded or unbounded—is leveraged strategically to evoke diverse network structures, which in turn have implications for distinct boundary-spanning activities. Practically, these findings show that ESM present a unique opportunity for workgroups to simultaneously sustain multiple virtual spaces—with varying levels of visibility—through which they can manage their diverse boundary-spanning goals.
Article
The argument proffered in this paper is that use of enterprise social networking technologies can increase the accuracy of people's metaknowledge (knowledge of "who knows what" and "who knows whom") at work. The results of a quasi-natural field experiment in which only one of two matched-sample groups within a large financial services firm was given access to the enterprise social networking technology for six months revealed that by making people's communications with specific partners visible to others in the organization, the technology enabled observers to become aware of the communications occurring amongst their coworkers and to make inferences about what and whom those coworkers knew based on the contents of the messages they sent and to whom they were sent. Consequently only individuals in the group that used the social networking technology for six months improved the accuracy of their metaknowledge (a 31% improvement in knowledge of who knows what and an 88% improvement in knowledge of who knows whom). There were no improvements in the other group over the same time period. Based on these findings, how technologically enabled "ambient awareness" - awareness of ambient communications occurring amongst others in the organization - can be an important antecedent for knowledge acquisition is discussed.
Article
Purpose – This paper aims to contribute to defining the concepts of boundary spanner, gatekeeper and knowledge broker. Design/methodology/approach – A review of the literature covering more than 100 sources. Findings – A review of past research leads to proposing a set of new definitions and also to the detection of six research avenues. Originality/value – The ability of organizations to recognize, source and integrate key information or knowledge is important for their strategy, innovation and performance over time. Three types of individuals have information gathering and knowledge dissemination roles at the frontier of organizations and groups: boundary spanners, gatekeepers and knowledge brokers. Although research on these individuals is well-developed, we found that in practice, the definitions of the concepts overlap and still need a clarification. So far, no systematic comparison of these roles has been undertaken.
Article
This paper offers a theory of communication visibility based on a field study of the implementation of a new enterprise social networking site in a large financial services organization. The emerging theory suggests that once invisible communication occurring between others in the organization becomes visible for third parties, those third parties could improve their metaknowledge (i.e., knowledge of who knows what and who knows whom). Communication visibility, in this case made possible by the enterprise social networking site, leads to enhanced awareness of who knows what and whom through two interrelated mechanisms: message transparency and network translucence. Seeing the contents of other's messages helps third-party observers make inferences about coworkers' knowledge. Tangentially, seeing the structure of coworkers' communication networks helps third-party observers make inferences about those with whom coworkers regularly communicate. The emerging theory further suggests that enhanced metaknowledge can lead to more innovative products and services and less knowledge duplication if employees learn to work in new ways. By learning vicariously rather than through experience, workers can more effectively recombine existing ideas into new ideas and avoid duplicating work. Moreover, they can begin to proactively aggregate information perceived daily rather than engaging in reactive search after confronting a problem. I discuss the important implications of this emerging theory of communication visibility for work in the knowledge economy.
A resource-based view of organizations suggests that internal knowledge is amongst the most important sources of competitive advantage (Hendry and Pettigrew 1990; Leonard-Barton 1995). Regardless of whether these ideas are exaggerated, it is not necessary to buy into them all to accept the significance of knowledge as a resource. Nevertheless, the fact remains that knowledge is complex, multi-faceted, intangible, often tacit and specialised (Davenport and Prusak; Ahmed et al. 2002). Consequently, it is difficult to manage exchange. This has implications for narrowing the gap between academic research and policy impact and developing policy-relevant academic research and to improve a flow of information between academics and policy-makers. One way of improving the exchange and flow of knowledge is to use knowledge brokers (KBs), who bridge the gap between creators and users of knowledge (Lavis 2006). The function of a KB may vary according to context. In the case of specialised knowledge, Stuiver, et al (2013) have gone as far as to suggest that KBs are part of a creative process, where knowledge-broking may be a way of codifying, translating and, finally, sharing, difficult to disseminate knowledge. This idea has been expanded upon in the literature with KBs engaged in the process of the creation of new knowledge (Oldham and McLean 1997; Stuiver et al. 2013). This paper builds upon a study examining how the Work and Employment Relations Survey (WERS) data informs government policy, in order to establish the significant role which KBs play in bridging the gap between research, knowledge and policy-making. Our findings indicate that WERS is widely used and feeds into a wide range of non-academic policy outputs through the involvement of particular agents, who play the role of KBs at various points in the dissemination and use of WERS data.