Conference PaperPDF Available

Towards a Generation of Artificially Intelligent Strategy Tools: The SWOT Bot

Authors:

Abstract and Figures

Strategy tools are widely used to inform the complex and unstructured decision-making of firms. Although software has evolved to support strategy analysis, such digital strategy tools still require heavy manual work especially on the data input and processing levels, making their use time-intensive, costly, and susceptible to biases. This design research presents the ‘SWOT Bot’, a digital strategy tool that exploits recent advances in natural language processing (NLP) to perform a SWOT (strengths, weaknesses, opportunities, threats) analysis. Our artifact uses a feed reader, an NLP pipeline, and a visual interface to automatically extract information from a text corpus (e.g., analyst reports) and present it to the user. We argue that the SWOT Bot reduces time and adds objectivity to strategy analyses while allowing the human-in-the-loop to focus on value-adding tasks. Besides providing a functioning prototype, our work provides three general design principles for the development of next-generation digital strategy tools.
Content may be subject to copyright.
Association for Information Systems Association for Information Systems
AIS Electronic Library (AISeL) AIS Electronic Library (AISeL)
ECIS 2022 Research-in-Progress Papers ECIS 2022 Proceedings
6-18-2022
Towards a Generation of Arti<cially Intelligent Strategy Tools: The Towards a Generation of Arti<cially Intelligent Strategy Tools: The
SWOT Bot SWOT Bot
Christian Au
Mainz University of Applied Sciences
, christian.au@hs-mainz.de
Till J. Winkler
University of Hagen
, till.winkler@fernuni-hagen.de
Herbert Paul
Mainz University of Applied Sciences
, herbert.paul@hs-mainz.de
Follow this and additional works at: https://aisel.aisnet.org/ecis2022_rip
Recommended Citation Recommended Citation
Au, Christian; Winkler, Till J.; and Paul, Herbert, "Towards a Generation of Arti<cially Intelligent Strategy
Tools: The SWOT Bot" (2022).
ECIS 2022 Research-in-Progress Papers
. 63.
https://aisel.aisnet.org/ecis2022_rip/63
This material is brought to you by the ECIS 2022 Proceedings at AIS Electronic Library (AISeL). It has been
accepted for inclusion in ECIS 2022 Research-in-Progress Papers by an authorized administrator of AIS Electronic
Library (AISeL). For more information, please contact elibrary@aisnet.org.
Thirtieth European Conference on Information Systems (ECIS 2022), Timișoara, Romania 1
TOWARDS A GENERATION OF ARTIFICIALLY
INTELLIGENT STRATEGY TOOLS: THE SWOT BOT
Research in Progress
Christian Au, Mainz University of Applied Sciences, Mainz, Germany, christian.au@hs-mainz.de
Till J. Winkler, Hagen University, Hagen, Germany, till.winkler@fernuni-hagen.de and Copenhagen
Business School, Frederiksberg, Denmark, winkler@cbs.dk
Herbert Paul, Mainz University of Applied Sciences, Mainz, Germany, herbert.paul@hs-mainz.de
Abstract
Strategy tools are widely used to inform the complex and unstructured decision making of firms.
Although software has evolved to support strategy analysis, such digital strategy tools still require heavy
manual work especially on the data input and processing levels, making their use time-intensive, costly,
and susceptible to biases. This design research presents the SWOT Bot, a digital strategy tool that
exploits recent advances in natural language processing (NLP) to perform a SWOT (strengths,
weaknesses, opportunities, threats) analysis. Our artifact uses a feed reader, an NLP pipeline, and a
visual interface to automatically extract information from a text corpus (e.g., analyst reports) and
present it to the user. We argue that the SWOT Bot reduces time and adds objectivity to strategy analyses
while allowing the human-in-the-loop to focus on value-adding tasks. Besides providing a functioning
prototype, our work provides three general design principles for the development of next-generation
digital strategy tools.
Keywords: Decision support systems, Strategy tools, Strategic analysis, SWOT, Natural language
processing, Design science
Strategy tools—such as a SWOT, Porter’s Five Forces, or the PEST modelare widely used among
managers, analysts, and consultants in today’s business practice (e.g., Jarzabkowski and Kaplan, 2014;
Wright et al., 2013; Vaara and Whittington, 2012 and or Clark, 1997). While the term ‘tool’ generally
refers to anything that helps perform a specific task, management tools cover any model, concept,
framework, method, or technique that helps to solve a managerial problem (Hakala and Vuorinen, 2020;
Jarzabkowski and Kaplan, 2014; Knott, 2008). Strategy tools are a subset of management tools that deal
with strategic challenges, such as understanding a company’s positioning and developing a new
competitive strategy. Strategy tools are cognitive as well as material artifacts (Paroutis et al., 2015) that
shape the mental models of strategists and hence affect both content and process of strategy work
(Vuorinen et al., 2018). The application of strategy tools in practice is based on quantitative and
qualitative inputs, where qualitative information seems to dominate.
Although the use of strategy tools has evolved from pen and paper drawings to digital templates and
dedicated software suites over the past decades (Ain et al., 2019; Arnott et al., 2017), our initial analysis
found that digital tools for strategy available on the market lack support on the input, processing, and
output levels. Contemporary digital strategy tools heavily rely on human tasks because strategic analysis
is a complex process that integrates knowledge and data from heterogeneous and often unstructured
sources (Hakala and Vuorinen, 2020; Jarzabkowski and Kaplan, 2015). While human involvement in
strategy processes matters (Reeves and Ueda, 2016), it also comes with certain costs (Grant, 2016).
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 2
There are substantial internal costs, e.g., for staff in strategy departments performing such knowledge
integration tasks as well as external costs for strategy consultants (Paroutis et al., 2016; Gray, 2018).
Albeit costly, the analysis through humans puts limits to the amount of available information that can
be processed while at the same time making this process susceptible to human biases, such as
confirmation biases (Nickerson, 1998).
The advent of artificially intelligent text processing through natural language processing (NLP)
techniques, especially models based on the transformer architecture, promises new opportunities for
decision support systems dealing with large amounts of unstructured data (Devlin et al. 2018; Wolf et
al., 2020). This applies equally to strategy analysis, which is the problem domain we address in this
research. We therefore pursue the question: How can we increase efficiency, and reduce human bias, in
the use of strategy tools?
Taking a design science approach, we iteratively developed and prototyped the ‘SWOT Bot’, a digital
strategy tool that automatically performs a SWOT (strengths, weaknesses, opportunities, threats)
analysis. We argue that SWOT is a good instance for our problem domain since it is one of the most
frequently used strategy tools in management (Jarzabkowski et al., 2009; Schneemann, 2019). Our
design artefact consists of a feed reader (automatically retrieving, e.g., external analyst reports, annual
reports, etc.), an NLP pipeline applying a pre-trained language model, and a visual interface aggregating
and presenting the outputs.
Our development followed three design principles of (1) automatic data updates and signal extraction,
(2) automatic synthesis of information, and (3) interactive data exploration and curation. Besides
showcasing a functioning prototype in this paper, we argue these principles as generalizable to the
development of next generation digital strategy tools, beyond our case of a SWOT. We believe a digital
strategy tool such as our SWOT Bot will be of use particularly for corporate and business units that
monitor competitor strategies on a regular basis as well as for consultancies that offer such strategic
analyses as a service. A plan for the experimental evaluation of our design artifact with a sample of
students and an evaluation in practice is provided.
Digital strategy tools can be related to broader classes of information systems (IS). They represent a
class of decision support systems (DSS), because they focus on supporting and improving managerial
decision making (Arnott and Pervan, 2016). They can also be seen as knowledge management systems
(KMS) since they “... support and enhance … processes of knowledge creation, storage/retrieval,
transfer, and application” (Alavi and Leidner, 2001, p. 114). Digital strategy tools combine strategic
management tools, that is abstract knowledge artifacts, with software and thus create digital material
artifacts (Gregor and Jones, 2007). We use the input-processing-output model of software systems
analysis (Boell and Cecez-Kecmanov, 2012) to display, based on our judgement, the degree to which
components of past generations of (digital) strategy tools have been supported by technology. Figure 1
illustrates this evolution over the past decades. The nine common technical components of KMS are
based on Shim (2002) and Saito (2007).
Traditionally, the application of strategy tools has relied on analogue material artifacts such as flipcharts,
whiteboards, post-its and spreadsheets (Jarzabkowski et al., 2013). Since the rise of the desktop PC in
the 1980s, practitioners have increasingly used Office software, such as Powerpoint and Excel, to
document their results and conduct analyses (Kaplan, 2011). The focus of these tools has been to move
from paper to digital files for the authoring and storing of the results.
Around the mid 1990s, multiple commercial vendors started to offer templates to implement specific
straegy tools in these office applications. These Excel- and Powerpoint-templates not only help users
document the results, but also present them in visually appealing ways and share them with others. Some
templates embed the tool in workflows that allow to re-use the results at later stages for further analyses
(e.g., by linking content from several Excel sheets). In the last years, online whiteboard-tools have
emerged that provide users with visual templates for analytic and creative tasks (e.g., Miro).
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 3
The latest generation of digital strategy tools are part of dedicated software suites, that not only
document the results, but also support simple workflows. Some tools, for example, link strategy analysis
to strategic planning by supporting the definition of key performance indicators, cascading them in the
organization, and monitoring them. Due to their complexity, these tools are usually standalone
applications (i.e., not built on top of Office applications). Most commercial tools focus on performance
management issues, but there are some smaller vendors that offer software for the complete strategic
planning process (i.e., from the design of vision and mission statements to environmental scanning).
Despite this evolution, the use of contemporary digital strategy tools comes with at least two important
shortcomings. First, while they facilitate the visual representation on the output levels and offer some
workflow supports on the processing level, they still require manual work to search, store and analyse
primary information sources on the input and processing levels (see Figure 1). For example, a user
performing a SWOT, PEST, or Five Forces analysis will typically search for information sources such
as annual reports, external analyses, and news reports. The analysis of these information sources and
their transformation into a synthesis of the dimensions of the specific strategy tool (e.g., the strengths,
weaknesses, opportunities and threats of the SWOT) is a cognitive task that requires significant time
and intellectual capability, depending also on the quantity and quality of the available sources and the
experience of the analyst in performing this task.
Second, the cognitive processing, and the visualization or authoring on the output levels, inevitably
introduce human biases due to the previous knowledge and individual interpretations of the analyst,
such as confirmation biases (Nickerson, 1998). Confirmation biases are widely recognized as a tendency
“...for people to seek information and cues that confirm the tentatively held hypothesis or belief, and not
seek (or discount) those that support an opposite conclusion or belief” (Wickens and Hollands 2000, p.
312). Such biases can have multiple adverse effects in the domain of strategy analysis. Jarzabkowski
and Kaplan (2015) point out that tool users have interpretive flexibility and tend to adapt the results
according to their interpretations to support their favoured views. Phadermrod et al. (2019), for example,
observed in brainstorming sessions that the SWOT analysis generated subjective judgements. Different
interpretations can lead to different strategic conclusions, which pose a potential threat in strategy
making (Fisher et al., 2020). While some level of interpretation, validation, and refinement through
human involvement may be desirable, researchers have, in fact, ever since called for more objectivity
in strategic decision-making processes. Hill and Westbrook (1997, p. 51) demanded for strategy
practitioners an “... obligation to verify statements and opinions with data and analyses.”
In sum, while strategy tools have become more digitally supported over the past decades, the current
generation of digital strategy tools falls short in assisting the analyst in the input and processing tasks,
which in turn makes their work effortful and susceptible to human biases. Vuorinen et al. (2018) find
that, until today, strategy work appears to have remained untouched by the technological developments
Figure 1. Evolution of digital strategy tools (authors’ representation)
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 4
and the opportunities presented by crowdsourcing, big data analysis, and artificial intelligence. Against
this backdrop, we put up three design requirements for next-generation digital strategy tools:
(1) Input: Digital strategy tools should support automated data updates and signal extraction. This can
be achieved through access to internal and external data sources with the help of APIs, automatic API-
based data updates, collaborative identification and sharing of relevant information.
(2) Processing: Digital strategy tools should support automatic synthesis of information. This involves
the automatic extraction of relevant information in large texts, the synthesis or summarization of relevant
text fragments, recognition of entities in the text and their ontological linkages.
(3) Output: Digital strategy tools should enable interactive data exploration and curation. This allows
users to understand complex information faster and trace relations between evidences. In addition, users
should be able to quickly manipulate data points in order to add relevant information or overwrite
algorithmic-based classifications and summarizations if these are inaccurate.
We followed a design science approach to address these three design requirements through a prototype
artifact. Design science provides a methodological framework for problem-solving oriented research in
IS that “... focuses on creating and evaluating innovative IT artifacts that enable organizations to address
important information-related tasks” (Hevner et al. 2004, p. 98). We focused on the SWOT analysis
since it is one of the most widely used tools in strategy analysis (Jarzabkowski et al., 2009; Schneemann,
2019). In addition, due to the moderate complexity of this tool with its four relatively intuitive
dimensions, the SWOT provides a suitable example as a proof of concept. The development of our
artifact, the SWOT Bot, was iterative and informed by theory (the knowledge base) and practice (the
real world environment). It proceeded in three major stages.
The primary motivation can be traced back to the individual experiences of the authors of this paper
all of whom have backgrounds in strategy consultingof how arduous, effortful, and at the same time
imprecise the use of strategy tools is in practice. Based on this experience, one of the authors started
building simple prototypes using structured data (e.g., from financial data providers such as Bloomberg
and Factset) and the software Tableau. In a practice project with the market intelligence department of
a large industrial company the authors learned, however, that the real pain for most analysts is the
extraction of relevant insights from large amounts of qualitative data (such as news articles, annual
reports, internal reports, etc.) rather than the evaluation of quantitative data.
Inspired by the idea to leverage natural language processing (NLP) for this problem, we evaluated, based
on a small corpus of articles, which technology might work best for automatic summarization and
information retrieval. We implemented a basic NLP pipeline to extract texts from documents, split them
into smaller paragraphs and calculate sentence embeddings. We tested several models (specifically
GPT-3 and roBERTa as BERT-derivate) with the help of standardized questions. Comparing the output
of the models showed very encouraging results: 80% of answers extracted with roBERTa correctly
identified by the model as crucial pieces of information from the articles. The manual extraction of
information was performed by students.
In a third stage, we integrated the NLP pipeline and the models into a larger prototype and developed a
front-end. Specifically, we built a news feed that would facilitate the retrieval of relevant data inputs
based on key word search and connected this to the NLP component. The news feed is based on the App
Feedly (2021) and allows configuring news streams based on press reports (filtered by keywords) or
internal documents via RSS feeds. We also designed and implemented in collaboration with a UI expert
a user interface that would allow business practitioners to quickly browse through extracted answers, to
change the relevance assigned by the computer, and to add their own assessments.
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 5
We built a functional prototype to demonstrate the potential of NLP-supported digital strategy tools.
Technically, the SWOT Bot leverages pre-trained transformer models to support analysts for question
answering and topic clustering tasks. In following, we describe the three core components of the
technical architecture of the SWOT Bot in more details, as displayed in Figure 2.
The Feed Reader
The feed reader allows analysts to receive a constant flow of events and news related to the topic they
are planning to analyze. Analysts can easily configure a tailored news feed for a SWOT based on
keywords (e.g., “BMW” or “automotive industry”) and specific news sources they consider relevant.
Most media outlets provide an RSS feed that can be integrated in a feed reader to receive updates on
articles published (e.g., the RSS feed of The Economist, 2022). The feed reader pushes new articles that
are relevant to the feed. Users can browse their timeline to receive updates and bookmark articles they
consider relevant easily. Articles that are highlighted by users are pushed to an RSS feed that forms the
corpus of the SWOT and can be processed by the NLP pipeline. This can also be done cooperatively by
multiple analysts.
There are plenty of existing commercial implementations of feed readers that can be used to configure
a feed and receive updates. In our prototype, we use Feedly (2021), one of the most popular feed reader
apps according to the download statistics of the Apples App Store and Android Play. Feedly provides
both an API and an RSS feed, which we use to connect to the corpus created and retrieve the data for
our NLP pipeline.
NLP Pipeline
The corpus that analysts create with the help of the feed reader is used in a subsequent step as an input
for the SWOT. The SWOT Bot automatically extracts the relevant information from the corpus by
deploying an NLP model that is optimized towards question-answering-tasks. To this end, we use a
BERT-based transformer architecture that builds on the latest advances in natural language technology
(Wolf et al., 2020). Transformer-based pre-trained models are made available as open-source libraries
by the research community and can be used to quickly build domain-specific applications. We
specifically use roberta-base-squad2 as model for our tasks (Liu et al., 2019). RoBERTa is a BERT-
based model that has been fine-tuned to the task of information retrieval in question-answering-systems.
Figure 2. Software components of the SWOT Bot (authors representation)
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 6
It achieves state-of-the-art performance in the SQuAD 2.0 leaderboard that compares the accuracy of
natural language models for reading comprehension tasks (cf. Rajpurkar et al., 2016). For the purpose
of this prototype, we did not fine-tune Roberta to a specific language domain (such as strategy analysis).
As future research, we plan to both train the language model to the strategy analysis domain and to the
downstream task of question answering for SWOT-specific questions.
In our prototype we also use the platform Haystack Hub to store and pre-process documents with the
help of roBERTa. Haystack is designed “... to be a very practical, down-to-earth open source NLP
framework” (Haystack, 2021). It allows users to integrate several components relevant in a NLP pipeline
from the pre-processing of documents, the implementation of a document-store and the usage of the
latest transformer-based models such as roBERTa to retrieve information. Haystack Hub provides an
API both to feed new documents in the document store as well as information-retrieval.
Our NLP pipeline works the following way: If a new document is highlighted in the feed reader the
document is passed to a pre-processor which tokenizes the document and calculates sentence
embeddings (for more details on how Transformer architectures work see Wolf et al., 2020). Resulting
sentence embeddings are put in the document store. Relevant evidences from the new documents are
extracted by passing standardized questions in the context of a SWOT to a reader-retriever component.
The questions emulate an analyst, who wants to extract relevant information from documents. They are
instantiated by filling variables X (company for which strengths and weaknesses) and Y (industry that
is analyzed with respect to opportunities and threats). The questions are:
- “What are strengths of X, especially distinct capabilities and resources?”
- “What are X's weaknesses, specifically where does it have inferior capabilities or resources?”
- “What are opportunities for Y, especially trends that support demand?”
- “What are threats for Y, especially damaging trends?”
The reader returns a list of relevant answers for each question with a scoring that indicates the likelihood
with which this a correct answer. All answers are stored in another database (implemented in
Elasticsearch) in order to use them in the next step.
Figure 3. Clustered answers under topics and summarization (Screenshot)
Dot-Connector
In the last step, all answers extracted automatically in the NLP pipeline are aggregated and presented to
the user with the help of a visual interface, the Dot-Connector. Before the Dot-Connector visualizes the
answers, it analyzes topics in the answers with the help of the sentence BERT architecture (SBERT).
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 7
SBERT is a modification of the BERT architecture. It achieves superior results in similarity comparison
and clustering of documents compared to traditional BERT architectures by using siamese and triplet
network structures (Reimers et al., 2020). Instead of classic clustering approaches, such as a latent
Dirichlet allocation (LDA), which rely on bag-of-words models to cluster sentences (Blei et al., 2003),
SBERT models allow us to calculate sentence embeddings for all evidences. Sentence embeddings
consider the context of the words and thereby represent the meaning of sentences more accurately
(Grootendorst, 2022).
Topics for the SWOT Bot are generated in three steps closely following the approach in Grootendorst
(2022): First, for all evidences from the NLP pipeline sentence embeddings are calculated with the help
of a state-of-the-art SBERTmodel (Hugging Face, 2022a). Second, the resulting embeddings are fed to
an UMAP algorithm to reduce the dimensionality of the embeddings (McInnes et al., 2018) and
subsequently clustered with the help of HDBSCAN (McInnes et al., 2017). Third, topic representations
are generated by summarizing the clustered sentences with the help of the distilbart model, fine-tuned
on the CNN news corpus (Hugging Face, 2022b).We also add named entities to each answer with the
help of Google’s NLP API (Google Cloud, 2021).
The Dot-Connector is built in Svelte a Javascript framework. It presents for each dimension of a SWOT
the following aspects, see Figure 3: (1) A summary of all answers identified, generated with the help of
another fine-tuned transformer model (bart-large-CNN model developed by Facebook, see Hugging
Face, 2022c), (2) Clusters of similar answers are identified and grouped together (using sBERT), and
(3) all entities extracted from the answers. Analysts can use this interactive user interface to change the
automatically assigned relevance of answers and sources in order to overwrite the relevance of answers
calculated by the NLP pipeline.
In line with the guidelines of Hevner et al. (2004) for design research, we plan to assess the SWOT Bot
in terms of its utility, quality, and efficacy. We aim to do so in two stages, first in a controlled experiment
with students and second through an evaluation in practice. The experiment (first stage) aims to test the
time savings and the reduction of results biases through the use of the SWOT Bot. To this end, we plan
to acquire part-time and full-time students in a Master-level strategy course. The subjects will be asked
to develop a SWOT for two comparably complex company cases (e.g., two different automotive
companies) in two separate runs. The text corpuses for each case will be provided. As tools, the subjects
may use either (1) pen & paper or (2) the SWOT Bot and measure the times they need for this task.
Since each subject assesses two company cases, we will assess whether the subjects were systematically
faster using the SWOT Bot compared to using pen & paper.
Group 1: Pen & paper
Group 2: SWOT Bot
A. Control: No prior
information given
A1
A2
Expected
differences
in the durations
of the analyses
B. Treatment: Biased
pre-information given
B1
B2
Expected differences in the assessments
Table 1. Planned experimental design and group setup
In addition, half of each group will be manipulated through a bias (B: treatment groups) while the other
half will not receive this bias (A: control groups). The bias will consist in the students reading a
manipulated industry report (e.g., a text about the decline of the automotive industry) which is not part
of the text corpus. This manipulated report plays down the strengths and opportunities and plays up the
weaknesses and threats of the industry (or vice versa). After each run, the subjects will assess their
individual analysis results in terms of the perceived degrees of strengths, weaknesses, opportunities and
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 8
threats of their company on given questionnaire scales. The comparison of these assessments between
treatment (B) and control groups (A) will allow us to assess whether and to what extent the treatment
groups were potentially biased by this (false) contextual knowledge. Our hypothesis is that the extent of
this bias is larger between the pen & papger groups (A1, B1) and smaller between the SWOT Bot groups
(A2, B2). Table 1 summarizes the planned experimental design and group split up.
In a second stage, we also plan to evaluate the SWOT Bot with a sample of expert practitioners, who
work in strategy consultancies or strategy departments of larger firms and who are experienced in
performing strategy analyses of companies. Analogous to the student groups, these subjects wiill be
asked to perform a SWOT analysis on two comparable cases, one using pen & paper and the other using
the SWOT Bot. Although the number of observations might not suffice for statistical analyses, we will
measure the time for this task to estimate the relative time savings in practice. However, the focus of
this second-state evaluation will be on qualitative insights and the fit of our artifact for practice. To this
end, we plan to interview the analyists before and after the use of the SWOT Bot and observe their work
processes.Through the interviews, we hope to gain more insights into other potential advantages and
disadvantages of our design artifact. For example, subjects might feel that the use of the artificially
intelligent bot for the synthesis reduced their creativity and learning about the company case. We might
also learn about the skills and training required for effectively using the tool from the interviewees. Or,
they might provide additional clues where the synthesis of the bot has systematic omissions or
shortcomings. In the spirit of an iterative artifact design in action (Sein et al., 2011), we thereby hope to
gain additional insights for the further improvement of our design artifact.
We are confident to be able to present the evaluation results along with a live version of the prototype
at the poster session of the ECIS 2022 conference in Timisoara. We hope to gain from the interaction
with other conference participants inputs for the further development and dissemination of our work.
In our view, there are at least four promising avenues for future research emerging from our work that
other researchers may find interesting to discuss with us. First, while we demonstrated the
implementation of the three design requirements for a SWOT analysis and argued our approach to be
applicable to other strategy tools, such as PEST or Porter‘s Five Forces, it will be necessary to expand
the use of our artifact to other uses to underline its generalizability for qualitative strategy tools. Second,
for all of these strategy tools, more labelled training data (e.g., company reports with labelled
SWOT/PEST/Porter categories) should prove useful to fine-tune the automated synthesis and allow the
language model to better learn the specific strategy analysis domain. Third, while the SWOT Bot is text-
based, we see a potential for integrating number with text data to increase the quality of the results.
Next-generation digital strategy tools could, for example, retrieve quantitative firm data from sources
such as Bloomberg and Factset, in addition to our news feed, to bring more measurable indicators into
the analysis. Fourth, group collaboration features could be included in artificially intelligent strategy
tools such as our SWOT Bot to increase the value added in a collaborative work context.
In conclusion, while our prototype demonstrates that natual language processing has great potential
within the context of digital strategy tools to support human analysis, more research is necessary to
advance the knowledge in this field. We hope our work inspires other IS researchers to join this effort.
Ain, N., Vaia, G., DeLone, W. H. and Waheed, M. (2019). “Two decades of research on business
intelligence system adoption, utilization and success A systematic literature review,” Decision
Support Systems 125, 113113.
Alavi, M. and Leidner, D. E. (2001). Knowledge management and knowledge management systems:
Conceptual foundations and research issues,” MIS Quarterly 25 (1), 107-136.
Arnott, D. and Pervan, G. (2008). “Eight key issues for the decision support systems discipline,”
Decision Support Systems 44 (3), 657-672.
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 9
Arnott, D. and Pervan, G. (2016). “A critical analysis of decision support systems research revisited:
The rise of design science,” in: Willcocks, L. P., Sauer, C. and Lacity M. C. (eds.) Enacting Research
Methods in Information Systems, 43-103, London: Palgrave Macmillan.
Arnott, D., Lizama, F. and Song, Y. (2017). Patterns of business intelligence systems use in
organizations, Decision Support Systems 97, 58-68.
Blei, D. M., Ng, A. Y. and Jordan, M. I. (2003). Latent Dirichlet allocation. Journal of Machine
Learning Research, 3 (Jan), 9931022.
Boell, S. and Cecez-Kecmanov, D. (2012). “Conceptualizing information systems: From 'Input-
Processing-Output' devices to sociomaterial apparatuses,” European Conference on Information
Systems (ECIS) 2012 Proceedings. 20. URL: https://aisel.aisnet.org/ecis2012/20 (visited on March
28, 2022).
Burke, G. T. and Wolf, C. (2021). The process affordances of strategy toolmaking when addressing
wicked problems,” Journal of Management Studies 58 (2), 359-388.
Clark, D. (1997). “Strategic management tool usage: A comparative study,” Strategic Change 6 (7),
417-427.
Devlin, J., Chang, M.-W., Lee, K. and Toutanova, K. (2019). Bert: Pre-training of deep bidirectional
transformers for language understanding. arXiv preprint. URL: https://arxiv.org/abs/1810.04805
(visited on March 28, 2022).
Feedly (2021). Tracking feedback across the web without having to read everything. URL:
https://feedly.com/ (visited on Mach 28, 2022).
Fisher, G., Wisneski, J. E. and Bakker, R. M. (2020). Strategy in 3D: Essential tools to diagnose, decide,
and deliver. Oxford: Oxford University Press.
Haystack (2021). deepset-ai/haystack. URL: https://github.com/deepset-ai/haystack/ (visited on March
28, 2022).
Google Cloud (11.09.2021). Cloud natural language. Analyzing entities. URL:
https://cloud.google.com/natural-language/docs/analyzing-entities?hl=en. (visited on March 28,
2022).
Grant, R. M. (2016). Contemporary strategy analysis, 9th Edition. Chichester: John Wiley & Sons.
Gray, J. (2018). Boom time for Germany's management consultants, Handelsblatt. URL:
https://www.handelsblatt.com/english/companies/double-digit-growth-boom-time-for-germanys-
management-consultants/23694984.html?ticket=ST-4284732-42ZYldmCgHYq3p1l1apL-
cas01.example.org (visited on March 28, 2022).
Gregor, S. and Jones, D. (2007). “The anatomy of a design theory,” Journal of the Association for
Information Systems 8 (5), 312-335.
Grootendorst, M. (2022). BERTopic: Neural topic modeling with a class-based TF-IDF procedure.
arXiv preprint. URL: https://arxiv.org/abs/2203.05794 (visited on March 28, 2022).
Hakala, H. and Vuorinen, T. (2020). Tools for Strategy: A starter kit for academics and practitioners.
Cambridge: Cambridge University Press.
Hevner, A. R., March, S. T., Park, J. and Ram, S. (2004). “Design science in information systems
research,” MIS Quarterly 28 (1), 75-105.
Hill, T. and Westbrook, R. (1997). “SWOT analysis: It’s time for a product recall,” Long Range
Planning 30 (1), 46-52.
Hugging Face (2022a). all-mpnet-base-v2. URL: https://huggingface.co/sentence-transformers/all-
mpnet-base-v2 (visited on March 28, 2022).
Hugging Face (2022b). sshleifer/distilbart-cnn-6-6. URL: https://huggingface.co/sshleifer/distilbart-
cnn-6-6 (visited on March 28, 2022).
Hugging Face (2022c). facebook/bart-large-cnn. URL: https://huggingface.co/facebook/bart-large-cnn
(visited on March 28, 2022).
Jarzabkowski, P. and Kaplan, S. (2015). “Strategy tools-in-use: A framework for understanding
´technologies of rationality´ in practice,” Strategic Management Journal 36 (4), 537-558.
Jarzabkowski, P., Oliveira, B. and Giulietti, M. (2009). Building a strategy toolkit: Lessons from
business: Executive briefings. London: Advanced Institute of Management Research.
Towards Artificially Intelligent Strategy Tools
Thirtieth European Conference on Information Systems (ECIS 2022), Timisoara, Romania 10
Jarzabkowski, P., Spee, A. P. and Smets, M. (2013). “Material artifacts: Practices for doing strategy
with ‘stuff’,” European Management Journal 31 (1), 41-54.
Kaplan, S. (2011). “Strategy and Powerpoint: an inquiry into the epistemic culture and machinery of
strategy making, Organization Science 22 (2), 320-346.
Knott, P. (2008). “Strategy tools: who really uses them?Journal of Business Strategy 29 (5), 26-31.
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L. and
Stoyanov, V. (2019). Roberta: A robustly optimized bert pretraining approach. arXiv preprint. URL:
https://arxiv.org/abs/1907.11692 (visited on November 15, 2021).
McInnes, L., Healy, J. and Astels, S. (2017). hdbscan: Hierarchical density based clustering. The Journal
of Open Source Software 2 (11), 205.
McInnes, L., Healy, J., Saul, N. and Großberger, L. (2018). UMAP: Uniform Manifold Approximation
and Projection. Journal of Open Source Software 3 (29).
Nickerson, R. S. (1998). “Confirmation bias: A ubiquitous phenomenon in many guises,” Review of
General Psychology 2 (2), 175-220.
Paroutis, S., Franco, L. A. and Papadopoulos, T. (2015). “Visual interactions with strategy tools:
producing strategic knowledge in workshops: Visual interactions with strategy tools,” British
Journal of Management 26 (1), 48-66.
Paroutis, S., Angwin, D. and Heracleous, L. T. (2016). Practicing strategy: Text & cases, 2nd Edition.
Los Angeles: SAGE.
Phadermrod, B., Crowder, R. M. and Wills, G. B. (2019). “Importance-Performance analysis based
SWOT analysis,” International Journal of Information Management 44, 194-203.
Rajpurkar, P., Zhang, J., Lopyrev K. and Liang, P. (2016). SQuAD: 100,000+ questions for machine
comprehension of text. URL: https://arxiv.org/abs/1606.05250 (visited on March 28, 2022).
Reeves, M. and Ueda, D. (2016). Designing the machines that will design strategy. Harvard Business
Review. URL: https://hbr.org/2016/04/welcoming-the-chief-strategy-robot (visited on March 28,
2022).
Reimers, N. and Gurevych, I. (2019). Sentence-bert: Sentence embeddings using siamese bert-networks.
arXiv preprint. URL: https://arxiv.org/abs/1908.10084 (visited on March 28, 2022).
Saito A, Umemoto, K. and Ikeda, M. (2007). “Strategy-based ontology of knowledge management
technologies," Journal of Knowledge Management 11 (1), 97-114.
Schneemann, P. (2019). Different perspectives on strategizing-theory and practical use of strategy tools.
PhD thesis, London Southbank University.
Sein, M. K., Henfridsson, O., Purao, S., Rossi, M. and Lindgren, R. (2011). “Action design research,”
MIS Quarterly 35 (1), 37-56.
Shim, J.P., Warkentin, M., Courtney, J. F., Power, D. J., Sharda, R. and Carlsson, C. (2002). “Past,
present, and future of decision support technology,” Decision Support Systems 33 (2), 111-126.
Spee, A. P. and Jarzabkowski, P. (2009). “Strategy tools as boundary objects,” Strategic Organization
7 (2), 223-232.
The Economist (2022). RSS Feed. URL: https://www.economist.com/rss (visited on March 28, 2022).
Vaara, E. and Whittington, R. (2012). “Strategy-as-Practice: Taking social practices seriously,”
Academy of Management Annals 6 (1), 285-336.
Vuorinen, T., Hakala, H., Kohtamäki, M. and Uusitalo, K. (2018). “Mapping the landscape of strategy
tools: A review on strategy tools published in leading journals within the past 25 years,” Long Range
Planning 51 (4), 586-605.
Wickens, C. D. and Hollands, J. G. (2000). Engineering Psychology and Human Performance, 3rd
Edition. Upper Saddle River, NJ: Prentice-Hall.
Wolf, T., Chaumond, J., Debut, L., Sanh, V., Delangue, C., Moi, A., Cistac, P., Funtowicz, M., Davison,
J., and Shleifer, S. (2020). “Transformers: State-of-the-art natural language processing,” Proceedings
of the 2020 Conference on Empirical Methods in Natural Language Processing: System
Demonstrations, 38-45.
Wright, R. P., Paroutis, S. E. and Blettner, D. P. (2013). “How useful are the strategic tools we teach in
Business Schools?” Journal of Management Studies 50 (1), 92-125.
Article
Full-text available
This "reflections from practice" piece explores some of the implications of emerging, artificially intelligent tools for the futures and foresight prac-academic community. The authors provide background on these emerging, artificially intelligent tools, and explore, with special emphasis on scenarios, a specific tool named "Chat Generative Pre-trained Transformer" (hereafter, ChatGPT). The authors examine the utility of scenarios generated by artificial intelligence (AI) and explore whether or not the futures and foresight prac-academic community should selectively embrace advances in AI to assist in the generation of scenarios. In particular, the authors will consider (1) the utility of using scenarios generated completely by AI, (2) whether what is produced, in fact, constitute scenarios, based on conventional definitions, and (3) assess the utility of using AI to assist in the production of scenarios. At this point in time, artificially intelligent tools can now generate numerous scenarios on seemingly any topic at essentially zero cost to the user. Still, the authors insist that the utility of those scenarios is largely predicated on the user's ability to coax the appropriate "raw material" from the artificially intelligent bot, which implicates, the authors contend, that such bots can usefully provide base material for the development of scenarios but are unlikely to fully eclipse scenarists in the production of scenarios. Additionally, the authors recommend that the futures and foresight prac-academic community pay especially close attention to artificially intelligent tools for novel insights with regard to the differences in human cognition and, in this case, the logic of large language model outputs.
Article
Full-text available
Studies have examined how managers use strategy tools, but we know much less about how managers create strategy tools de novo. We undertook an ethnographic study of a business facing a wicked problem and investigated the sociomaterial practice of collective toolmaking. We identify how strategy toolmaking oscillates between different problem domains and reveal how this manifests process affordances, which are ‘unintended’ by‐products of the toolmaking process. Counterintuitively, by intentionally making a strategic tool, actors unintentionally create a sociomaterial springboard for 'spin‐off strategizing' and ‘the discovery of latent ambiguities’, generating strategic value beyond the tool produced. These insights illuminate how the practice of collective toolmaking can stimulate wayfinding, indirectly helping managers to respond to wicked problems, characterized by high degrees of complexity, ambiguity, and indeterminacy.
Article
Full-text available
We present a new reading comprehension dataset, SQuAD, consisting of 100,000+ questions posed by crowdworkers on a set of Wikipedia articles, where the answer to each question is a segment of text from the corresponding reading passage. We analyze the dataset in both manual and automatic ways to understand the types of reasoning required to answer the questions, leaning heavily on dependency and constituency trees. We built a strong logistic regression model, which achieves an F1 score of 51.0%, a significant improvement over a simple baseline (20%). However, human performance (86.8%) is much higher, indicating that the dataset presents a good challenge problem for future research.
Book
Cambridge Core - Strategic Management - Tools for Strategy - by Henri Hakala & Tero Vuorinen This book discusses the concept and applications of strategy tools. Strategy tools are frameworks, techniques, and methods that help individuals and organizations to create their strategies. After a brief overview of different ideas on strategy and strategic thinking, we move on to define and discuss what strategy tools are and elaborate on the promise and perils of using them to implement strategic management. We review the most commonly used, classic tools and techniques but also less well-known tools of the strategy trade, as proposed by scholars writing in the leading strategy journals. We conclude by offering suggestions on how to improve strategic design and the effectiveness of the resultant strategy through the selective use of the most appropriate tools. Overall, this Element provides a quick overview of the tools that are available to those tasked with creating organizational strategies and making strategic decisions.
Book
This book provides a tools-based approach to strategic management. The central framework rests on three pillars that constitute the essence of strategy: to diagnose, to decide, and to deliver. Within this framework a suite of strategic management tools is offered, which include both the classics and the more nascent frameworks used to strategize. The first part of the book offers a brief introduction to the essentials of strategic management and unpacks the “3D” framework of strategy. The second part of the book revolves around explaining the purpose, underlying theory, core idea, depiction, process, value created, and risks and limitations of each tool. Hands-on advice is emphasized. The book also offers case illustrations that offer concrete examples of how the tools can be applied. The concluding chapter summarizes the key insights on a high level and offers concluding thoughts on how the tools can be combined.
Article
This article taps into the question of the materialized forms of theorizing in strategy: the strategy tools presented in publications over the past 25 years. This study conducts a systematic search and review of 482 published abstracts and 88 full text articles introducing tools to aid strategizing. The contribution of this study builds on the theoretical classification framework and review of strategy tools to illustrate what might be termed the toolbox of strategy from the publications in leading management journals. The review suggests that the landscape of strategy tools is surprisingly traditional and that contemporary developments in strategic thinking have not yet been transformed into usable tools. Furthermore, the study also provides some recommendations for the developers of new strategy tools in terms of topics and methodological considerations.
Article
Business intelligence (BI) is often used as the umbrella term for large-scale decision support systems (DSS) in organizations. BI is currently the largest area of IT investment in organizations and has been rated as the top technology priority by CIOs worldwide for many years. The most important use patterns in decision support are concerned with the type of decision to be supported and the type of manager that makes the decision. The seminal Gorry and Scott Morton MIS/DSS framework remains the most popular framework to describe these use patterns. It is widely believed that DSS theory like this framework can be transferred to BI. This paper investigates BI systems use patterns using the Gorry and Scott Morton framework and contemporary decision-making theory from behavioral economics. The paper presents secondary case study research that analyzes eight BI systems and 86 decisions supported by these systems. Based on the results of the case studies a framework to describe BI use patterns is developed. The framework provides both a theoretical and empirically based foundation for the development of high quality BI theory. It also provides a guide for developing organizational strategy for BI provision. The framework shows that enterprise and smaller functional BI systems exist together in an organization to support different decisions and different decision makers. The framework shows that personal DSS theory cannot be applied to BI systems without specific empirical support.