Content uploaded by Jörg Weking
Author content
All content in this area was uploaded by Jörg Weking on Nov 28, 2019
Content may be subject to copyright.
15th International Conference on Wirtschaftsinformatik,
March 08-11, 2020, Potsdam, Germany
When Algorithms Go Shopping: Analyzing Business
Models for Highly Autonomous Consumer Buying Agents
Michael Weber1, Marek Kowalkiewicz2, Jörg Weking1, Markus Böhm1, and
Helmut Krcmar1
1 Technical University of Munich, Chair for Information Systems, Garching, Germany
{mic.weber, joerg.weking, markus.boehm, helmut.krcmar}@tum.de
2 Queensland University of Technology, Chair in Digital Economy, Brisbane, Australia
marek.kowalkiewicz@qut.edu.au
Abstract. Consumer buying agents (CBAs) are software programs that automate
tasks in the consumer buying process (e.g., product search and evaluation).
Recently, CBAs have the ability to nearly automate the whole buying process,
executing transactions with only minimal human involvement. With the rise of
such highly autonomous CBAs, updates to business models (BM) of involved
parties are expected (e.g., adding a sales channel and increasing customer value).
However, our understanding of BMs for highly autonomous CBAs remains
limited. In this work we aim to close this gap. We investigate 23 cases and
develop a BM taxonomy for highly autonomous CBAs. We further encode these
cases into the taxonomy and derive BM patterns. Our work contributes to
research by setting a foundation for the conceptual understanding of BMs for
highly autonomous CBAs. Practitioners can use our taxonomy and patterns for
strategic guidance and to support BM innovation.
Keywords: Consumer Buying Agent, Autonomous Agent, Autonomous
Shopping, Business Model, Taxonomy.
1 Introduction
With the rise of E-commerce, scholars have begun discussing potential use cases for
software agents, which are continuously running, personalized software programs
capable of carrying out actions on their own [1]. Software agents assist and automate
tasks for both businesspeople and consumers. For example, businesses can use agent
technology to automate procurement activities [2, 3] or to deliver a personalized
marketing experience [3, 4]. On the consumer side, agents can support buying decisions
by comparing offerings or making recommendations [1, 5]. Whereas the traditional role
of agents is decision support, a shift toward more autonomous and decisive agents is
occurring, putting humans in the supervising role [2, 3].
A recent trend features connected things able to purchase goods and services on
behalf of consumers [6, 7]. As an example, air filters monitor their lifecycle and printers
detect shortages of ink. When needed, they can place new supply orders without human
intervention. Voice assistants (e.g., ) can process simple shopping
commands to select and purchase products (on human consent). Thereby these things
take on typical tasks of a consumer: analyzing needs, choosing among different
offerings and executing transactions. Because only minimal human interaction and
decision making is required, these things become highly autonomous consumer buying
agents (CBA).
With the advent of highly autonomous CBAs, updates to business models
1
(BM) of
involved parties (e.g., retailers and appliance manufacturers) are expected. First, CBAs
create new sales channels [8]. Second, automatic replenishment and voice shopping
reduce customer efforts and enable less-frictional purchases [6]. The increased
convenience will likely become the standard customer expectation. Therefore, it
becomes a key part of the value proposition. Lastly, key partnerships will probably play
an essential role to ensure that the offerings are actually considered by CBAs [9]. For
involved businesses, it is important to understand the dynamics of such trends to adapt
and innovate their BMs. The process of business model innovation (BMI) is a necessary
activity to sustain competitive advantage in an ever-changing environment [10, 11].
However, past cases have shown that BMI can be difficult to implement [12].
Existing BM research has not yet analyzed highly autonomous CBAs. Therefore,
there is a lack of conceptual understanding on the topic. In this work, we build a BM
taxonomy for highly autonomous CBAs and derive business model patterns (BMP).
We aim to bridge the research gap and provide valuable insights for practitioners
striving for BMI.
The rest of the paper is structured as follows. We first summarize related work on
BMs and CBAs, to provide both context and background. We then continue by
describing our research method. As the result of our study, we present a BM taxonomy
and BM patterns for highly autonomous CBAs. We then discuss our outcomes
regarding both research and practice before we conclude the paper.
2 Related Work
2.1 Business Models
Many definitions for the BM have been proposed [13], however they seem to agree on
a common set of components. According to Gassmann, Frankenberger and Csik [14],
a BM describes the customer segment (who), value proposition (what), value chain
(how), and profit mechanism (why) of a firm. Similar components are used in other
definitions, such as those by Teece [11] and Osterwalder and Pigneur [15].
BMPs can be generalizations of reoccurring BMs with similar characteristics [16,
17]. BMPs can both describe whole BMs or only specific parts [16] (e.g., the profit
mechanism). Practitioners can use BMPs for BMI and collaborative ideation [14].
Various generic BMPs have been presented or collected in different works [14-17].
Apart from general research on BMs, scholars have analyzed BMs in specific
information system (IS) domains (e.g., carsharing [18], fintech [19], industry 4.0 [20],
1
We interpret business models as formal concepts [13]
internet of things (IoT) [21], or product service systems [22]). Such works contribute
to a better understanding of the business side of these technology-driven trends, such
as by creating taxonomies, analyzing important dimensions, or deriving common
patterns. Thus far, studies have not yet targeted CBAs as actors in BMs.
In summary, extant research provides us with generic dimensions and patterns,
which can be used to analyze BMs for CBAs. However, BM literature does not yet
provide an in-depth investigation on CBAs. Therefore, no specialized tools exist.
2.2 Consumer Buying Agents
CBAs
2
can be defined as software programs that automate parts of the consumer buying
process [1]. The consumer buying behavior model (Figure 1) can be used to analyze
agent-automated tasks [1, 23]. The consumer buying behavior model describes all steps
a consumer typically goes through when making a purchase. Agents can thus identify
needs [1, 3], collect and evaluate information on products and merchants [1, 3, 23],
negotiate conditions [1, 3, 23], and potentially execute the purchase [1]. Depending on
the novelty and risk of the purchase, consumers prefer different tasks to be automated
by CBAs [24]. Highly autonomous CBAs have thus far not been addressed by extant
literature. There is a lack of understanding on what tasks they automate and how they
Furthermore, scholars have studied the impact of CBAs. Consumers can save time
and improve decision quality by using recommender [5, 25] or product comparison
agents [25, 26]. On business side, CBAs impact firm performance. It has been shown
[27]. However, the success
also depends on the suppliers and their pricing strategies [28].
To the best of our knowledge, there is only indirect evidence for the use of CBAs in
BMs. CBAs have been discussed in the context of direct selling and third-party
marketplace BMs [25], mediating transactions between buyers and sellers [1, 23]. In
addition, CBAs are recognized as value-added services of firms [27]. Nevertheless,
BMs for CBAs remain an understudied topic, lacking a direct investigation.
Figure 1. Consumer buying behavior process model (Maes, Guttman and Moukas [1])
3 Research Method
Our research follows a four-step approach. First, we define the scope of our research
and determine the criteria for highly autonomous CBAs. Second, we collect cases based
2
We refer to consumer buying agents as all kinds of agents automating tasks in the consumer
buying process. Highly autonomous consumer buying agents represent a subset within this
definition with a certain level of autonomy (see section 3.1 for a detailed definition).
Need
Identification Product
Brokering Merchant
Brokering Negotiation Purchase and
Delivery Service and
Evaluation
on these criteria. Third, we use these cases to develop a BM taxonomy, which allows
to describe the underlying BMs. Finally, we apply the taxonomy to the cases, and we
encode the respective BMs. We derive BMPs via qualitative cluster analysis.
3.1 Definition of Scope
For the scope of our research, we further refine the term highly autonomous consumer
buying agent and set criteria to guide case selection. According to Jennings, Sycara and
Wooldridge [29], an autonomous software agent should be able to act without human
intervention and to exercise control over its own actions. If we apply this definition to
CBAs, a fully autonomous CBA must be able to identify needs, select appropriate goods
or services, and execute the purchase without human intervention. However, consumers
do not necessarily want to give all control to software agents, especially not in novel,
high-risk purchase situations [24]. To ensure practical relevancy, we therefore focus on
a slightly weaker notion of autonomy, referred to as highly autonomous in this paper.
We use the consumer buying behavior process model to define highly autonomous.
We first include all fully autonomous CBAs that automate the whole buying process
(type 1). In addition, we include CBAs that, based on a human command, can select a
matching good or service (out of multiple options) and subsequently execute the
purchase (type 2). Thereby type 2 CBAs automate the whole buying process up to the
purchase, except for the need identification step.
We exclude CBAs that execute the purchase of a fixed good or service based on a
human command (e.g., reorder buttons). In that case, the good or service selection is
skipped. We argue that the CBA does not make any decisions, and therefore has no
control over its own actions. Moreover, we also exclude all CBAs that cannot execute
the purchase step (e.g., recommender, comparison agents), as they are merely
supporting decisions instead of autonomously executing purchases.
3.2 Case Collection
We build our case collection on the method suggested by Larsson [30]. Thus, we first
identified the cases we were already familiar with. Second, we conducted a case search
to identify new cases. The case search consisted of different search strategies and
sources, which helps to reduce case collection bias [30]. Owing the novelty of our
perspective on CBAs, we could not rely on extant literature for case collection. Instead,
we looked at business reports, databases for enterprise information, news articles, and
websites. To find these potential sources, we used web search, app stores, and queried
the database CrunchBase
3
, both by combining different search terms (e.g., bot,
assistant, shopping, purchase, automation, reorder, replenish). Some sources for one
case additionally refer to other cases that we had not identified at that time. For
4
5
app stores to discover
3
https://www.crunchbase.com/ (queried 24.06.2019)
4
https://www.amazon.com/alexa-skills/b/?node=13727921011 (last accessed 01.11.2019)
5
https://assistant.google.com/explore (last accessed 01.11.2019)
other cases with a voice interface. Hence, similar to backward and forward search in
literature reviews, we could identify cases based on the already known ones.
We considered all active cases where highly autonomous CBAs (type 1 or type 2)
were involved and where enough information was available to understand the
underlying BMs. In the case of start-ups, the respective company must have still existed
during our study. For Amazon Dash Replenishment, we only considered one case for a
certain type of appliance to avoid redundancy. For example, we only considered
Kyocera printers, whereas other firms offered a similar service.
We identified 23 cases in total. We stored all data in a central case base [31]. For
. We triangulated the data by synthesizing the findings from
all sources for a case, helping us to build a more profound understanding of the
underlying BM. Such data triangulation helps to increase the construct validity of a case
study [31]. Appendix A lists all cases, references the main sources used, and shows to
what type of highly autonomous CBA the cases adhere (type 1 or type 2).
3.3 Taxonomy Development
We applied the iterative method of Nickerson, Varshney and Muntermann [32] for
taxonomy development. This method has successfully proven itself in several related
IS studies, where BMs in a certain domain were investigated with the development of
a taxonomy [18-20]. Furthermore, it follows a holistic approach, in which well-funded
theoretical knowledge and empirical insights can be combined.
As a first step, we defined meta-characteristics for the taxonomy. Meta-
characteristics are the most comprehensive characteristics of the taxonomy and serve
as the basis for further selection [32]. We used the four dimensions of the widely
accepted BM definition of Gassmann, Frankenberger and Csik [14] as our initial set of
meta-characteristics: customer segment (who), value proposition (what), value chain
(how), and profit mechanism (why). These dimensions are both comprehensive and
abstract enough to serve as suitable meta-characteristics for our BM taxonomy.
Second, we set ending conditions for the taxonomy development. We used the
objective and subjective ending conditions suggested by Nickerson, Varshney and
Muntermann [32].
Third, we iteratively created the taxonomy. During the first iteration, we used the
conceptual-to-empirical approach, building upon related work. In particular, we
considered how CBAs create value for consumers in the buying process as well as how
are used in BMs. We evaluated the identified characteristics using
concrete cases and grouped them into the dimensions target customer, purchase
automation, purchase brokering, market role, and revenue model. During the second
iteration, we applied the empirical-to-conceptual approach, analyzing a subset of cases
with a focus on the value chain and key partners. We identified new characteristics,
added the dimension purchase scope, and split up the dimension market role into
endpoint type, endpoint belonging, and provisioning. At this moment, the taxonomy
now considered fine-grained differences in the value chain. During the third iteration,
using the empirical-to-conceptual approach again, we analyzed a larger subset of cases
from a technical viewpoint. We identified new characteristics, added the dimensions
agent stage, agent deployment, and agent interface, and split up revenue model into
ongoing revenue and upfront revenue. During the fourth iteration, we chose the
empirical-to-conceptual approach once more, this time analyzing all cases using our
taxonomy. The analysis and comparison of the cases did not require to add or modify
any of the characteristics or dimensions. All other ending conditions were met. We
stopped the process and the resulting taxonomy can be applied to all cases.
3.4 Derivation of Business Model Patterns
The BMPs were derived following a qualitative analysis approach. First, we encoded
the cases in a matrix (see Appendix A). Each row of the matrix represents a case. Each
column of the matrix represents a dimension in the BM taxonomy. With the help of the
matrix, we performed a qualitative cluster analysis [31] to derive BMPs. At the highest
level, we identified three overarching BMPs. Within these three overarching BMPs, we
further distinguished eight sub-patterns.
4 Results
4.1 Business Model Taxonomy
The BM taxonomy builds upon the four meta-characteristics: customer segment, value
proposition, value chain, and profit mechanism. Table 1 shows 34 characteristics along
12 dimensions. The taxonomy describes BMs for highly autonomous CBAs. It covers
concrete CBA instances and their supporting ecosystem. As an example, BMs for agent
platform providers can also be described. In this case, some dimensions of the
taxonomy might be intentionally left blank, because they depend on the concrete
instantiation of an agent. For instance, an agent platform might not prescribe whether
the agent sells its own products or third-party products. Each dimension is briefly
explained below.
Businesses can both directly target consumer (i.e., business-to-consumer (B2C))
with agent instances, or they can target other businesses (i.e., business-to-business
(B2B)) with agent-enabling services (target customer). Businesses can offer a ready-
to-use agent for consumers, agent blueprints, or extendible agent platforms for other
businesses (agent stage). Agents can either fully automate purchases by identifying
needs or be triggered by human commands (purchase automation). Agents might
consider different types of products and different merchants for their decision.
However, some agents might do no brokering at all and just purchase a fixed offering
(purchase brokering). Agents can purchase a single offering, offerings in a certain
domain, or even across multiple domains (e.g. booking hotels and public transport)
(purchase scope). Agents can be deployed in mobile or web applications, on an agent
platform, or via connected devices such as printers and dishwashers (agent
deployment). Agents can be controlled through a graphical user interface, through a
conversational interface (e.g., voice or chat), or with no interface at all (e.g., for fully
automatic purchases). In this case, the human might monitor the agent in some way
(agent interface). Agents can either purchase from their own business or from external
business partners (endpoint belonging). For the choice of offering, agents can consider
a single vendor that provides the offering, a set of vendors, or a marketplace (endpoint
type). Businesses can provide purchased goods or services on their own, rely on third
parties, or follow a mixed approach (provisioning). Businesses can generate ongoing
revenue through direct sales, transaction-based commission fees, subscriptions, or
generate no revenue at all. In this case, they might profit in other ways (e.g., through
platform network effects) (ongoing revenue) [33]. Businesses might require consumers
to purchase a physical device before an agent can be used (e.g., in the case of
appliances) (up-front revenue).
Table 1. Business model taxonomy for highly autonomous consumer buying agents
Meta-charact.
Dimension
Characteristics (No. of cases)
Customer Seg.
Target Customer
B2C (18)
B2B (5)
Value
Proposition
Agent Stage
Agent Application (18)
Agent Blueprint (1)
Agent Platform (4)
Purchase Automation
Automated (13)
Human-triggered (10)
Purchase Brokering
Product (2)
Merchant (3)
Both (4)
None (11)
Purchase Scope
Single G/S (14)
Domain (2)
Cross-Domain (4)
Value Chain
Agent Deployment
Web/Mobile (4)
Agent Platform (7)
IoT Device (12)
Agent Interface
GUI (1)
Conversational (10)
None/Monitoring (12)
Endpoint Belonging
Own Endpoint (6)
External Endpoint (13)
Endpoint Type
Single Vendor (4)
Set of Vendors (5)
Marketplace (10)
Provisioning
Own G/S (7)
3rd Party G/S (8)
Both (3)
Profit
Mechanism
Ongoing Revenue
Sales (7)
Commission (12)
Subscription (2)
None (1)
Up-front Revenue
Sales of Device (10)
None (13)
4.2 Business Model Patterns
We identify three overarching BMPs for highly autonomous CBAs and eight sub-
patterns. At the highest level, we differentiate between agents for direct sales (pattern
1), agents as mediators (pattern 2), and agent enablement (pattern 3). For agents for
direct sales, a business uses an agent to sell their own goods or services to a consumer.
For agents as mediators, an intermediary uses an agent to mediate transactions between
a consumer and one or many businesses. In contrast to that, agent enablement is not
concerned with the actual usage of an agent. Rather, it enables third parties to use
agents. Table 2 illustrates and summarizes the three overarching BMPs with their
respective sub-patterns. Below the table, we describe the eight sub-patterns within the
three overarching BMPs in more detail.
Table 2. Business model patterns for highly autonomous consumer buying agents
Pattern & Sub-patterns (No. of cases)
Illustration
Pattern 1: Agents for Direct Sales
• IoT-enabled Auto-Selling (6)
• Conversational Selling (2)
Pattern 2: Agents as Mediators
• IoT-enabled Auto-Mediation (3)
• Agent-enabled Mediation (5)
• Agent-supported Mediation (2)
Pattern 3: Agent Enablement
• Agent Platform (2)
• IoT-Integration Platform (2)
• Solution Provision (1)
Legend
Pattern 1: Agents for Direct Sales. Businesses deploy agents to directly sell goods
and services to their customer. The agent becomes an additional sales channel that
automates the purchasing process for the customer. We identify two sub-patterns for
direct sales:
IoT-enabled Auto-Selling. Connected appliances or products can track their status and
detect the need for a replacement. In case of an upcoming need, an order can be placed
automatically without human intervention. For example, HP, Inc. offers instant ink
replenishment for their printers. Filtrete uses Amazon Dash Replenishment to let their
smart filters replace themselves.
Conversational Selling. Businesses provide a conversational interface to order goods
or services via chat or voice. The agent enables simple purchases and can assist the
customer in the decision process. For example, Virgin Train built a voice app on
Amazon Alexa for ticket bookings and information. Walmart allows its users shop via
voice using the Google Assistant, learning their preferences from past orders.
Pattern 2: Agents as Mediators. Intermediaries deploy agents to mediate between
buyers (consumers) and sellers (businesses). The agent assists consumers in their
purchasing decisions. We identify three sub-patterns for mediators:
IoT-enabled Auto-Mediation. As for IoT-enabled Auto-Selling, connected appliances
and products automatically detect needs and can reorder replenishables. However, the
agent owner does not sell its own goods. Instead, the agent mediates between the
consumer and a third party. For example, GE Appliances uses Amazon Dash
Replenishment on their connected dishwashers to automatically reorder dishwasher
pods. WePlenish developed a smart container for a set of compatible coffee pods that
can automatically restock via Amazon Dash Replenishment.
Agent-enabled Mediation. The agent integrates multiple goods or service providers and
connects them with potential consumers. The agent provides an individualized service
for customers (e.g., shopping inspiration, best price comparison, or process
automation). Through this mediation, a marketplace is created that is enabled by the
agent. For example, Myia is a bot that automates switching to the cheapest energy
provider in the U.K. In India, Niki.ai is a chatbot that assists with various transactions,
such as mobile recharges, public transport tickets, hotel bookings, or electricity bills.
Agent-supported Mediation. Here the intermediary is already running a BM that
connects buyers and sellers. The agent is used to support mediation by providing a
convenient interface with automation to the customer. In contrast to Agent-enabled
Mediation, the agent is not essential for the mediation model. Amazon uses Alexa to
enable its users to shop at their online marketplace by using simple voice commands.
Google follows a similar approach with the Google Assistant and Google Express
marketplace.
Pattern 3: Agent Enablement. Enablers provide services for other businesses to
enable their agents. In contrast to the previous patterns, this BMP does not involve a
concrete agent instance. We identify three sub-patterns for enablers:
Agent Platform. Agent enabler provide a platform for other businesses to develop and
deploy agents. The agent enabler can monetize their services and potentially benefit
from network effects on the platform. Third-party businesses can benefit from technical
expertise and the reach of the platform. For example, Amazon and Google offer an app
store for their personal voice assistants, where third parties can publish agents with a
voice interface.
IoT-Integration Platform. Agent enabler provide a platform for IoT-enabled auto-
replenishments. Third parties connect their appliances and products to the platform and
thereby instantiate an agent. The agent enabler can monetize from the ongoing
transactions. For example, Amazon offers Amazon Dash Replenishment, a service for
automated replenishments on the Amazon marketplace. U.K. start-up Pantri builds a
holistic subscription platform that integrates appliances for automated replenishments.
Solution Provision. Agent enabler provide third parties with services to create and
manage their agents. Third parties can be provided with a non-technical interface,
configurable agent blueprints, and the management of multi-platform deployments. As
an example, Blutag enables retail voice applications for Amazon Alexa and Google
Assistant.
5 Discussion
Highly autonomous CBAs have the potential to change the ways consumers buy and
the ways businesses sell products and services. However, BMs for highly autonomous
CBAs remain understudied. Therefore, we developed a BM taxonomy and derived
BMPs for highly autonomous CBAs.
The three overarching BMPs are agents for direct sales, agents as mediators, and
agent enablement. Because of their high level of generalization, we can find similar
concepts in existing BMPs (e.g., direct selling, online brokerage, or value chain service
provider [16]). Other patterns cover the idea of agency, where a business represents the
interests of a buyer and profits from successful transactions (e.g., agent models or
search agent [16]). The differentiating aspect of our proposed BMPs is the algorithmic
agent, which nearly autonomously executes transactions on behalf of the consumer.
These highly autonomous CBAs provide superior added value and become an essential
part of the overall value chain.
The agent creates value by either completely automating purchases or by brokering
the best-fitting offer. Complete automation seems to be especially suited for low-risk,
frequent purchases for which the purchase need can be physically measured (e.g., IoT-
enabled auto-selling and IoT-enabled auto-mediation). This then becomes quite similar
to a typical subscription model [16]. However, in case of full automation, we can expect
regulators stepping in to protect consumers (e.g., forcing the inclusion of multiple
offerings [7], or forbidding [34]). The brokering of offerings can be
used to select the best-fitting offer for the consumer. The agent can broker the offerings
of established sellers (conversational selling, agent-supported-mediation). However,
the agent can also create a novel marketplace via the brokering activity (agent-enabled
mediation). This seems to be particularly suited for start-ups. In cases where CBAs
broker from multiple offerings, we do not see full automation. Whether consumers will
ever completely distribute complex purchasing decisions to agents remains open.
Moreover, the agent becomes an essential part of value chain, placing itself as a
mediator between businesses and consumers. On business side, CBAs represent an
additional sales channel. On consumer side, CBAs can offer novel ways of interacting,
such as conversational interfaces. Because conversational interfaces provide limited
interaction possibilities, they seem to align well with highly autonomous agents.
Furthermore, we find that agent enablers play an important role in the ecosystem.
Out of 23 cases, 11 cases built on enablers and five were enablers themselves. Agent
enablers serve with technological expertise in a rather complex field. In the case of
platforms, enablers provide a user base and reach across multiple channels. We
identified Amazon and Google as tech giants leading the field. However, start-ups like
Pantri and Blutag reveal niche opportunities in the market.
We thus contribute to the research with, to the best of our knowledge, the first
investigation of BMs for CBAs. Therefore, our taxonomy and proposed patterns set a
foundation for the conceptual understanding of BMs for CBAs. Furthermore, by
specifically setting the scope on highly autonomous CBAs, we raise the research
CBAs.
Our work also contributes to practice. We built a taxonomy and a structured set of
patterns that can be systematically applied. The results provide practitioners strategic
guidance by giving insights to the BMs of a previously understudied trend. Practitioners
can use the taxonomy and the patterns to analyze and innovate their BMs.
Our study comes with limitations. First, because highly autonomous CBAs represent
a recent trend, we could only analyze 23 cases in our study. Nonetheless, our results
provide valuable insights for both research and practice. Additionally, the BM
taxonomy and the respective BMPs are extendable, such that future research can build
upon our results when more cases emerge. Second, as Nickerson, Varshney and
Muntermann [32] argued, taxonomies are never perfect, but should rather be useful
with regards to their purpose. We argue for the usefulness of our taxonomy, because it
allows us to describe and compare BM cases that lead to the derivation of BMPs.
Looking at future research, the emergence of novel types of highly autonomous
CBAs provides a fruitful avenue for research. Building upon our study, additional cases
can lead to a more extensive BM taxonomy and potentially to the discovery of
additional BMPs. Quantitative methods could be used to further validate the results.
Furthermore, researchers could analyze BMs of less autonomous subsets of CBAs and
compare their results with ours to get a more holistic view of the field. Moreover, as
with future research could investigate the impact of highly
autonomous CBAs on consumers, firms, and supply chains (e.g., through simulation),
as well as evaluate certain design features (e.g., explaining automated actions).
6 Conclusion
Software agents are now capable of near autonomous behavior, purchasing goods and
services on behalf of consumers. These highly autonomous CBAs are expected to
influence BMs by providing added value, additional sales channels, and new key
partnerships. It is important for businesses to understand such trends to adapt and
innovate their BMs in an ever-changing competitive environment. However, until now,
the literature has not investigated BMs for highly autonomous CBAs. Our goal was to
close this research gap with this paper. We first refined the term highly autonomous
consumer buying agent. Based on this understanding, we collected 23 cases. We used
the method by Nickerson, Varshney and Muntermann [32] to develop a BM taxonomy
for highly autonomous CBAs. We encoded the cases using the BM taxonomy and
derived three overarching BMPs and eight sub-patterns. We found out that businesses
could use agents to either directly sell products or to mediate between buyers and
sellers. Agents provide added value to the consumer buying process and become
important actors in the value chain. Furthermore, we identified agent enablers as
important players in the business ecosystem, enabling highly autonomous CBAs.
We contribute to the literature by setting the foundation for the conceptual
understanding of BMs for highly autonomous CBAs. Practitioners can use our BM
taxonomy and BMPs as strategic guidance to support BMI. Future research can build
upon our work, especially after more cases of highly autonomous CBAs have emerged.
7 Acknowledgements
The authors would like to thank all anonymous reviewers and the editors for their
helpful comments and suggestions. This research is funded by the German Research
Foundation (Deutsche Forschungsgemeinschaft
Research Center 768: Managing cycles in innovation processes Integrated
development of product servic
Center for Very Large Business Applications (CVLBA)@TUM.
References
1. Maes, P., Guttman, R.H., Moukas, A.G.: Agents that buy and sell. Communications of the
ACM 42, 81-82 (1999)
2. Nissen, M.E., Sengupta, K.: Incorporating software agents into supply chains: Experimental
investigation with a procurement task. MIS Quarterly 30, 145-166 (2006)
3. Yu, B., Vahidov, R., Saade, R.: Agents and e-commerce: Beyond automation. In: 21st
Americas Conference on Information Systems Puerto Rico. (2015)
4. Kumar, V., Dixit, A., Javalgi, R.G., Dass, M.: Research framework, strategies, and
applications of intelligent agent technologies (IATs) in marketing. Journal of the Academy
of Marketing Science 44, 24-45 (2016)
5. Xiao, B., Benbasat, I.: E-commerce product recommendation agents: use, characteristics,
and impact. MIS Quarterly 31, 137-209 (2007)
6. Gartner, https://www.gartner.com/smarterwithgartner/top-10-trends-in-digital-commerce
(Accessed: 30.05.2019)
7. Kowalkiewicz, M., Rosemann, M., Dootson, P.: Retail 5.0: Check-Out The Future. Business
Report, QUT PwC Chair in Digital Economy (2017)
8. Gregory, J.: The Internet of Things: Revolutionizing the Retail Industry. Business Report,
Accenture Strategy (2015)
9. Dawar, N.: Marketing in the Age of Alexa. Harvard Business Review May-June 2018 Issue,
80-86 (2018)
10. Sorescu, A., Frambach, R., Singh, J., Rangaswamy, A., Bridges, C.: Innovations in Retail
Business Models. Journal of Retailing 87, 3-16 (2011)
11. Teece, D.J.: Business models, business strategy and innovation. Long Range Planning 43,
172-194 (2010)
12. Chesbrough, H.: Business model innovation: Opportunities and barriers. Long Range
Planning 43, 354-363 (2010)
13. Massa, L., Tucci, C.L., Afuah, A.: A critical assessment of business model research.
Academy of Management Annals 11, 73-104 (2017)
14. Gassmann, O., Frankenberger, K., Csik, M.: The business model navigator: 55 models that
will revolutionise your business. Pearson, Harlow, England (2014)
15. Osterwalder, A., Pigneur, Y.: Business model generation: A handbook for visionaries, game
changers, and challengers. John Wiley & Sons, Inc., Hoboken, New Jersey (2010)
16. Remane, G., Hanelt, A., Tesch, J.F., Kolbe, L.M.: The business model pattern databasea
tool for systematic business model innovation. International Journal of Innovation
Management 21, 1750004 (2017)
17. Weking, J., Hein, A., Böhm, M., Krcmar, H.: A hierarchical taxonomy of business model
patterns. Electronic Markets In Press (2018)
18. Remané, G., Nickerson, R.C., Hanelt, A., Tesch, J.F., Kolbe, L.M.: A taxonomy of
carsharing business models. In: 37th International Conference on Information Systems
Dublin, Ireland. (2016)
19. Eickhoff, M., Muntermann, J., Weinrich, T.: What do FinTechs actually do? A Taxonomy
of FinTech Business Models. In: 38th International Conference on Information Systems
Seoul, South Korea. (2017)
20. Weking, J., Stöcker, M., Kowalkiewicz, M., Böhm, M., Krcmar, H.: Archetypes for industry
4.0 business model innovations. In: 24th Americas Conference on Information Systems New
Orleans, LA, USA. (2018)
21. Dijkman, R.M., Sprenkels, B., Peeters, T., Janssen, A.: Business models for the Internet of
Things. International Journal of Information Management 35, 672-678 (2015)
22. Weking, J., Brosig, C., Böhm, M., Hein, A., Krcmar, H.: Business Model Innovation
Strategies for Product Service Systems An Explorative Study in the Manufacturing
Industry. 26th European Conference on Information Systems Portsmouth, UK (2018)
23. He, M., Jennings, N.R., Leung, H.F.: On agent-mediated electronic commerce. IEEE
Transactions on Knowledge and Data Engineering 15, 985-1003 (2003)
24. Sproule, S., Archer, N.: A buyer behaviour framework for the development and design of
software agents in e-commerce. Internet Research 10, 396-405 (2000)
25. Häubl, G., Trifts, V.: Consumer decision making in online shopping environments: The
effects of interactive decision aids. Marketing Science 19, 4-21 (2000)
26. Hostler, R.E., Yoon, V.Y., Guimaraes, T.: Assessing the impact of internet agent on end
users' performance. Decision Support Systems 41, 313-323 (2005)
27. Pathak, B., Garfinkel, R., Gopal, R.D., Venkatesan, R., Yin, F.: Empirical Analysis of the
Impact of Recommender Systems on Sales. Journal of Management Information Systems
27, 159-188 (2010)
28. Li, L.S., Chen, J.Q., Raghunathan, S.: Recommender System Rethink: Implications for an
Electronic Marketplace with Competing Manufacturers. Information Systems Research 29,
1003-1023 (2018)
29. Jennings, N.R., Sycara, K., Wooldridge, M.: A roadmap of agent research and development.
Autonomous agents and multi-agent systems 1, 7-38 (1998)
30. Larsson, R.: Case Survey Methodology: Quantitative Analysis of Patterns across Case
Studies. The Academy of Management Journal 36, 1515-1546 (1993)
31. Yin, R.K.: Case study research and applications: Design and methods. Sage, Thousand Oaks
(2014)
32. Nickerson, R.C., Varshney, U., Muntermann, J.: A method for taxonomy development and
its application in information systems. European Journal of Information Systems 22, 336-
359 (2013)
33. Hein, A., Weking, J., Schreieck, M., Wiesche, M., Böhm, M., Krcmar, H.: Value co-creation
practices in business-to-business platform ecosystems. Electronic Markets 29, 503518
(2019)
34. Reuters, https://www.reuters.com/article/us-amazon-com-germany-court/court-says-
amazon-dash-buttons-violate-german-law-idUSKCN1P42HW (Accessed: 13.09.2019)
35. Google, https://support.google.com/googlenest/answer/7505831 (Accessed: 09.08.2019)
36. Amazon, https://www.amazon.com/b?ie=UTF8&node=16924225011 (Accessed:
09.08.2019)
37. Kyocera, https://www.kyoceradocumentsolutions.eu/en/products/software/amazon-dash-
replenishment.html (Accessed: 09.08.2019)
38. GE Appliances, https://www.geappliances.com/ge/connected-appliances/dishwashers.htm
(Accessed: 09.08.2019)
39. Filtrete, https://www.filtrete.com/3M/en_US/filtrete/products/smart-filter-technology/
(Accessed: 09.08.2019)
40. Petcube, https://petcube.com/news/amazon-alexa-drs/ (Accessed: 09.08.2019)
41. Illy, https://www.illy.com/en-us/shop/machines/espresso-machines/francis-francis-
y5/coffee-machines-y5-espresso-and-coffee-bluetooth-us-P.html (Accessed: 09.08.2019)
42. WePlenish, https://weplenish.com (Accessed: 09.08.2019)
43. Oral-B, https://www.amazon.com/stores/page/0A9082ED-0A78-440B-AB29-
644DBDCAC6C4 (Accessed: 09.08.2019)
44. HP, https://instantink.hpconnected.com/uk/en/l/ (Accessed: 09.08.2019)
45. Xerox, https://www.xerox.com/printer-supplies/metered-supplies-replenishment/enus.html
(Accessed: 09.08.2019)
46. Pantri, https://pantri.net/ (Accessed: 09.08.2019)
47. Garbi, https://garbican.com/ (Accessed: 09.08.2019)
48. Activaite, https://www.getmealiq.com/ (Accessed: 09.08.2019)
49. Blutag, https://blu.ai (Accessed: 09.08.2019)
50. The Drum, https://www.thedrum.com/news/2018/09/12/virgin-trains-voice-strategy-
shaping-up-five-months-after-launching-first-amazon (Accessed: 09.08.2019)
51. MyWave, https://myia.ai/ (Accessed: 09.08.2019)
52. Niki, https://niki.ai (Accessed: 09.08.2019)
53. TechCrunch, https://techcrunch.com/2019/04/02/walmart-partners-with-google-on-voice-
enabled-grocery-shopping/ (Accessed: 09.08.2019)
54. LookAfterMyBills, https://lookaftermybills.com/how-it-works/ (Accessed: 09.08.2019)
55. Amazon, https://developer.amazon.com/alexa (Accessed: 09.08.2019)
56. Google, https://developers.google.com/actions/ (Accessed: 09.08.2019)
57. Amazon, https://developer.amazon.com/dash-replenishment-service (Accessed:
09.08.2019)
ID
Case Title
Ref.
Type
Target
Customer
Agent
Stage
Purchase
Automation
Purchase
Brokering
Purchase
Scope
Agent
Deployment
Agent
Interface
Endpoint
Belonging
Endpoint
Type
Provisioning
Ongoing
Revenue
Upfront Revenue
1
Google Assistant Voice
Shopping
[35]
2
B2C
App
Human-
triggered
Both
Cross
Domain
Platform
Conver-
sational
Own
Marketplace
Both
Commission
None
2
Amazon Alexa Voice
Shopping
[36]
2
B2C
App
Human-
triggered
Both
Cross
Domain
Platform
Conver-
sational
Own
Marketplace
Both
Commission
None
3
Kyocera Printer
[37]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
Own G/S
Sales
Sales of Device
4
GE Appliances Dishwasher
[38]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
3rd Party G/S
Commission
Sales of Device
5
Filtrete Smart Filter
[39]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
Own G/S
Sales
Sales of Device
6
Petcube Bites Treat Dispenser
[40]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
3rd Party G/S
Commission
Sales of Device
7
Illy Coffee Machine
[41]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
Own G/S
Sales
Sales of Device
8
WePlenish Smart Container
[42]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
3rd Party G/S
Commission
Sales of Device
9
Oral-B Electronic Toothbrush
[43]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
External
Marketplace
Own G/S
Sales
Sales of Device
10
HP Instant Ink
[44]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
Own
Single Vendor
Own G/S
Subscription
Sales of Device
11
Xerox Metered Supplies
[45]
1
B2C
App
Automated
None
Single G/S
IoT Device
None
Own
Single Vendor
Own G/S
Sales
Sales of Device
12
Pantri Platform
[46]
1
B2B
Platform
Automated
None
Single G/S
IoT Device
None
-
-
-
-
None
13
Garbican Smart Trash Bin
[47]
2
B2C
App
Human-
triggered
Merchant
Domain
IoT Device
None
External
Set of Vendors
3rd Party G/S
Commissions
Sales of Device
14
MealIQ Meal Planer
[48]
2
B2C
App
Human-
triggered
Both
Domain
Web/Mobile
GUI
External
Set of Vendors
3rd Party G/S
Commission
None
15
Blutag Retail Voice Apps
[49]
2
B2B
Blueprint
Human-
triggered
-
-
Platform
Conver-
sational
-
-
-
Subscription
None
16
Virgin Train on Amazon
Alexa
[50]
2
B2C
App
Human-
triggered
Product
Single G/S
Platform
Conver-
sational
Own
Single Vendor
Own G/S
Direct Sales
None
17
Myia Energy Bot
[51]
1
B2C
App
Automated
Merchant
Single G/S
Web/Mobile
Conver-
sational
External
Set of Vendors
3rd Party G/S
Commission
None
18
Niki Chatbot
[52]
2
B2C
App
Human-
triggered
Both
Cross
Domain
Web/Mobile
Conver-
sational
External
Set of Vendors
3rd Party G/S
Commission
None
19
Walmart on Google Assistant
[53]
2
B2C
App
Human-
triggered
Product
Cross
Domain
Platform
Conver-
sational
Own
Single Vendor
Both
Direct Sales
None
20
LookAfterMyBills
[54]
1
B2C
App
Automated
Merchant
Single G/S
Web/Mobile
Conver-
sational
External
Set of Vendors
3rd Party G/S
Commission
None
21
Amazon Alexa Platform
[55]
2
B2B
Platform
Human-
triggered
-
-
Platform
Conver-
sational
-
-
-
Commission
None
22
Google Assistant Platform
[56]
2
B2B
Platform
Human-
triggered
-
-
Platform
Conver-
sational
-
-
-
None
None
23
Amazon Dash Replenishment
Platform
[57]
1
B2B
Platform
Automated
None
Single G/S
IoT Device
None
External
Marketplace
-
Commission
None
Appendix A Case List and Taxonomy Encodings
