See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/366919723

Implications of AI-based robo-advisory for private banking investment

advisory

Article in Journal of Electronic Business & Digital Economics · January 2023

DOI: 10.1108/JEBDE-09-2022-0037

CITATIONS READS

17 573

3 authors, including:

Christian Dietzmann Rainer Alt

University of Leipzig University of Leipzig
6 PUBLICATIONS 61 CITATIONS 559 PUBLICATIONS 5,264 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Rainer Alt on 02 March 2023.

The user has requested enhancement of the downloaded file.

 Implications of AI-based Robo-Advisory

Completed Research Paper

Introduction
Robo- sector. The term
"Fintech" represents the abbreviation of "financial technology" and describes both innovative IT-based
financial solutions as well as companies offering such solutions (Jung, Dorner, Glaser, et al., 2018). RA is
considered as a business model archetype that represents a decision support system from a technological
perspective and a personal assistant from a product offering perspective (Eickhoff et al., 2017). With RA
offering low fees and minimum deposit requirements, private banking (PB) client advisors are facing
increased competition (von Martens and Schildbach, 2019). Besides, the number of global RA users is
expected to be half a billion by 2025 accompanied by changes in client behavior and expectations (Statista,
2021a). Artificial intelligence (AI) may not only boost RA adoption as well as their assets under
management, but AI-based RA application types could also decrease transaction costs and support
investment advisors to understand client needs (Renaux et al., 2019; Brodski, 2019). Wealth managers and
private banks need to develop concepts integrating AI-based solutions within their value chain since
especially young, technology-savvy clients expect digital services as well as personalized offerings and
advisor contact (Kothari and Berry, 2021; Sachse et al., 2012). According to a study by Accenture (2018),
63% of the high-net-worth individuals (HNWIs) are highly interested in individual journeys across digital
channels and more than half of HNWIs would leave their private bank if no integrated and seamless
channels are available. Since the technological possibilities driven by AI indicate the PB client advisor no
longer to be the main client contact and PB profitability is decreasing, private banks need to redesign
investment advisory implementing AI-based solutions such as intelligent RA and find a balance between
machine- and human-centered client interaction (Moulliet et al., 2016). While retail banks already started
integrating intelligent systems, private banks lag the development and are still evaluating or testing the
integration of AI applications in the investment advisory process (Forbes Insights, 2019). The academic
literature is mainly concerned with the design of AI-based RA (e.g. Jung, Dorner, Weinhardt, et al. (2018))
and behavioral investigations (e.g. (Wang, 2020)) with a research gap regarding the processual redesign
and business impact of AI-based RA application types, especially on PB investment advisory. The process
coverage by AI-based systems as well as the application potentials resulting from media discontinuities in
the process have hardly been considered so far ([blinded for review]). Hence, the present research paper
examines two research questions (RQs):
RQ1: What is the processual impact of AI-based robo-advisory on private banking investment advisory?
RQ2: What are the implications for AI-based private banking investment advisory?
The RQs are addressed by deploying process-related and AI-based RA characteristics on the private banking
investment advisory process developed by Nueesch et al. (2016). The two main goals of the study are to
derive knowledge about (1) the process influence of AI-based characteristics and (2) process redesign
integrating AI. The remainder of the paper is as follows: The next section lays the foundations of AI-based
RA, the AI application types, and existing research. Section 3 presents the methodology and characterizes
the PB investment advisory process, section 4 contains the iterative development of the process redesign
and the evaluation. Section 5 discusses the implications, limitations, and further research opportunities.
Section 6 concludes the paper.

1
 Implications of AI-based Robo-Advisory

Existing research on Robo-Advisory

Emergence of AI-based Robo-Advisory
The origins of the RA business model date back to the year 2008, when US Fintech companies Wealthfront
and Betterment were founded (Fisch et al., 2019) and developed innovative solutions for advisory processes,
(Alt and Puschmann, 2012).
With the robo-
Germany since 2014, being the second largest European Fintech market with 25 robo-advisors following
the UK (Alt et al., 2018; von Martens and Schildbach, 2019). The emergence of robo-supported business
models was fueled by both a loss of confidence in actively managed funds following the financial crisis in
2008 and new technical possibilities (Poterba and Shoven, 2002). In contrast to traditional banks, RA
providers offer capital market access to formerly unbanked customers via ETF investments starting from
one Euro as minimum investment or monthly savings rate (Hoelscher and Nelde, 2018). The business
models focus on the young and technology-savvy generation with low, but potentially growing incomes,
which in 2016 was only considered serious competition by six to seven percent of 134 German asset
managers participating in a study (Sironi, 2016; Webersinke, 2017). This impression changed with RA
assets under management in Germany having more than tenfold from EUR 756 million in 2017 to EUR
8.068 billion in 2020 and are expected EUR 30.515 billion in 2025 (Statista, 2021b). The estimates are also
driv
RA -as-a-
service (SaaS
since the RA simply offers its complete service under the name of the contracting partner. The banks risk
of this strategy is a possible migration of customers to RA, the advantages are low implementation and
maintenance costs. In the two other cooperation forms, RAs act as pure IT service providers implementing
the bank's products and processes digitally. Thus, the bank's investment advisor remains with these two
solutions for queries based on product knowledge. The two latter strategies differ by the aspect that in the
platform solution, the RA can be assigned the tasks of processing and custody account management,
whereas in the SaaS solution all non-technical tasks remain with the bank (Schabicki et al., 2020).
Investment advisory digitization triggers transformation along two dimensions. The first one is the physical
dimension, i.e., the face-to-face customer-bank communication getting replaced by digital devices such as
smartphones and tablets (Cocca, 2016; Nueesch et al., 2016). In the second dimension, novel technologies
transform and automate advisory content creation, thereby shifting the focus from human- to algorithm-

represents a converging online and offline channel combination (Nüesch et al., 2015). Although human
advisory still dominates customer relationships in this archetype, clients may change between digital
channels like e-mails, video telephony, or chats. From an evolutionary viewpoint, RA archetype contains
four generations with the first representing a simple tool displaying investment options based on an online
questionnaire. The second generation comprise investment platforms enabling customers to invest in
ready-made investment funds. Asset managers execute the purchase orders, prepare the asset allocations,
and monitor the entire process. The third evolutionary stage is characterized by rule-based algorithms
automatically proposing investments and execute portfolio adjustments. RA 4.0 includes self-learning
algorithms being capable of investing and portfolio rebalancing driven by advances in AI (Moulliet,
Majonek, et al., 2016; Perrin and Roncalli, 2020). Intelligent conversational agents (CA) guide users
through business processes while other AI components like predictive Analytics (PA) or intelligent
customer segmentation (CS) have an impact on the quality and depth of advisory content (Law and Chung,
2020; Beck, 2020).

AI-based Robo-Advisory Application Types

AI-based RA can be decomposed into AI application types, which have significantly driven the evolution.
An application type refers to a cluster of software solving similar problems within a specific task
environment such as recognizing speech, playing chess, or driving cars (Russell and Norvig, 2016; Boobier,
2020). An advanced RA is composed of many such application types to address the high decision-making
complexity in investment advisory. Based on the three types of AI presented by Davenport and Ronanki
(2018), three AI-based RA application types could be identified: Intelligent CA (Day et al., 2018; Hildebrand

2
 Implications of AI-based Robo-Advisory

and Bergner, 2021), intelligent CS (Tertilt and Scholz, 2018; Xue et al., 2018) and PA (Day and Lin, 2019;
Gu et al., 2019). The characteristics of each are the following.
Intelligent conversational agents: The connection of AI and intelligent conversational systems processing
natural language enables automated interactions between banks and clients via text-based chatbots and
audio-based voice assistants (Day et al., 2018; Liao et al., 2018). This form of communication is primarily
used for customer inquiries relieving the cost-intensive front office (Crosman, 2018). According to the
Financial Stability Board (Schindler et al., 2017) the range of tasks performed by chatbots and voice
assistants in the financial industry is still modest and limited to routine and simple activities. However, due
to recent technological advances in machine learning, a generation of intelligent CAs is evolving (Li et al.,
2019; Yan, 2018). A significant advantage over other interaction channels arises primarily from the high
degree of mobility. A CA can typically be implemented across multiple devices and applications so that the
customer can interact with the bank regardless of time and place. For example, the financial services
provider Wells Fargo uses Facebook's Messenger service to automatically answer simple customer queries
(Burnett, 2017).
Intelligent customer segmentation: Banks are increasingly replacing rigid and generalized segmentation
approaches with individual profiling addressing the immense complexity and multi-layered nature of
customer needs as proposed by Khadivizand et al. (Khadivizand et al., 2020). With the help of AI-based CS,
RAs automatically divide the market and its clientele into meaningful (micro) segments by assessing risk
preferences or even analyzing their behavior such as answer switching while filling out a digital
questionnaire (Tertilt and Scholz, 2018; Tsiptsis and Chorianopoulos, 2009; McDonald, 2019). Through
this process, buyer groups may be identified to determine their specific characteristics and to predict the
individual risk (Beck, 2020). Kilic et al. (2017) find that client profiling should not necessarily be carried
out purely digitally, but that joint profiling is much more accepted by clients. If the intelligent system
correctly identifies the customer segment based on the joint profiling, risk-accurate investment solutions
can be offered, underlining that intelligent financial innovations are advantageous for both customers and
bank risk-taking ([blinded for review]).
Predictive Analytics: PA is an AI application type that helps identifying patterns from a vast amount of data
to make predictions about business-relevant content (Joshi et al., 2018). In doing so, algorithms can
identify parametric and non-parametric relationships that are mostly imperceptible to humans (LeCun et
al., 2015). In many cases, the datasets available to organizations are too large for human experts and
therefore, AI-based data processing is effectively supporting the generation of market or stock price
predictions (Day et al., 2018; Althelaya et al., 2018). The underlying algorithms often process input data
such as macroeconomic data and company-related financial information (Wang et al., 2019), social network
relationship data (Xue et al., 2018) or Twitter sentiment (Oliveira et al., 2017). A benefit of PA is the absence
of subjective influences, mitigated by the high degree of diversified information and AI model selection,
e.g., in sentiment-based predictions.

Robo-Advisory in Information Systems Research

RA has already received some attention in the information systems literature. Jung et al. (Jung, Dorner,
Glaser, et al., 2018)
by RA research. Most scholars focus on interface design in connection with (1) transparency and trust (Jung,
Dorner, Weinhardt, et al., 2018; Kilic et al., 2015; Kilic et al., 2017; Mesbah et al., 2019; Rühr, 2020), (2)
design approaches such as anthropomorphism (Morana et al., 2020; Adam et al., 2019), nudging (Jung and
Weinhardt, 2018) as well as user control (Rühr et al., 2019), and (3) the change from display- to voice-based
RA (Ostern et al., 2020)
which scholars investigate adoption determinants and barriers (Wang, 2020; Bruckes et al., 2019), trust-
influencing factors (Guo et al., 2019) -advice (Tauchert and Mesbah, 2019), RA
user characteristics (Woodyard and Grable, 2018; Fulk et al., 2018) and user attitudes encouraging and
discouraging RA adoption (Belanche et al., 2019). In addition, a third stream contains IS literature focusing
on the business process impact and process redesign of and by RA as well as its effect on human-based
financial advisory. The results of a qualitative case study by Coombs and Redman (Coombs and Redman,
2018) imply to augment the human advisor with RA since financial investment is an emotional process and
the RA Jung et al. (2019) ascertain that a lack of emotionality may
improve financial decision-making like professional investors, who are regularly outperforming retail

3
 Implications of AI-based Robo-Advisory

investors (Kinniry and DiJoseph, 2014). Furthermore, RA may be vulnerable to conflicts of interest because
of their affiliation with brokers, clearing firms etc. which in turn may lead to higher prices (Fein, 2015). It
has also been observed that RAs feature difficulties in the evaluation of client risk tolerance and provision
of highly personalized services, which is important in PB (Jung et al., 2019). Nueesch et al. (2014) address
the aspect of human augmentation with tablet-based applications in the PB investment advisory process.
The findings imply a redesign of the advisory process and provide first indications on the impact AI-based
RA application types may have when augmenting a human financial advisor. In a further study on tablet-
based financial advisory, (Nueesch et al., 2016) proposed tablet-supported PB investment advisory process
variants.

Conceptualization
Methodology
The research gap is addressed through the application of design science research (DSR) according to Hevner
et al. (Hevner et al., 2004). DSR was applied since the present study develops practically utilizable results,
(Winter, 2008). Furthermore, the research approach follows the design
cycles proposed by Peffers et al. (2007) to ensure rigor and relevance. First, the problem is identified
through a literature review and the study objectives are defined (Webster and Watson, 2002). The design
and development phase contains the process characterization, derivation of process-relevant RA
characteristics and an impact analysis of the AI-based RA application types CA, CS and PA on the PB
investment advisory process. The demonstration is presented by the re-designed investment advisory
process, including an integration of the three AI-based application types. The AI potentials for process
redesign were discussed in ten semi-structured PB expert interviews and the evaluation was conducted in
four separate semi-structured interviews with both PB and technology experts. Finally, the design
knowledge is presented by three implications for future AI-based PB investment advisory (Figure 1).

Figure 1. Research design cycles and iterative results based on Peffers et al. (2007).

4
 Implications of AI-based Robo-Advisory

Characterization of the Private Banking Investment Advisory Process

The
which encompasses the client-bank interface. To create an adequate understanding of the tasks connected
with investment advisory, this paper applies a PB advisory process, which was proposed by Nueesch et al.
(2016) and employed in an impact analysis of tablets onto bank advisory. The process represents the
traditional end-to-end investment advisory process, which is carried out by a human advisor being
separated into six process steps: (1) initiation, (2) profiling, (3) concept, (4) offer, (5) implementation, and
(6) maintenance. The process steps are divided into 26 sub-processes, which are the object of investigation
for the impact analysis of the RA-related AI application types (Figure 2).

Figure 2. The private banking investment advisory process based on Nueesch et al. (2016).

Process-relevant Robo-Advisory Characteristics

A structured literature review following Webster and Watson (2002) was conducted to identify process-
relevant RA characteristics for the subsequent iterative process redesign.

IEEE Xplore, ACM DL and Springer Link database including a backward-forward search. The literature
search resulted in a total set of 617 documents, which were analyzed for relevance to the RQs. More
precisely, the existing body of knowledge was screened for articles containing RA characteristics, which are
relevant in the context of the investment advisory process. Inappropriate research papers and duplicates
were excluded during this process. Another full-text screening resulted in 19 articles that were reviewed in
detail by two of the authors. The authors extracted seven RA characteristics, which are considered to
manifest the impact of RA within a process (Table 1): Accessibility, automation, availability, efficiency,
standardization, transparency, and usability are considered to manifest the impact of RA within a process.
During the impact analysis of the subsequent iterative process redesign of the present study, the seven
process-relevant RA characteristics are applied on each of the investment advisory sub-processes through
the three AI application types. The RA characteristics and corresponding scientific literature are shown in
Table 1.

5
 Implications of AI-based Robo-Advisory

Robo-advisory
Explanation References
characteristic
A characteristic feature of robo-advisors is the ease of access to the
Becchi, et al. (2018); Jung et al. (2018a); Kaya (2017);
Accessibility service. In principle, this also enables customers with low assets to use
O'Keefe et al. (2016); Ruf et al. (2015)
the service of an asset management company.
Becchi et al. (2018); D'Acunto et al. (2019); Fein (2015);
Automated activities by the robo-advisor allow the overall advisory
Fisch et al. (2017); Gold & Kursh (2017); Jung et al.
Automation process but also particular sub-processes to be executed without human
(2018a), Kaya (2017); Moulliet et al. (2016a); Moulliet,
intervention.
(2016b); Rühr (2020); Sironi (2016)
The advisory process can be accessed at any time and from any location, Becchi et al. (2018); Bellani et al. (2016); Fisch et al.
Availability provided the user has an internet-enabled device and the robo-advisor is (2017); Jung et al. (2018a); Moulliet et al. (2016a);
not affected by server problems. Singh & Kaur (2017)
Bellani et al. (2016); D'Acunto et al. (2019); Fisch et al.
The robo-advisor executes the individual sub-processes or activities
(2017); Gold & Kursh (2017); Jung et al. (2018b), Kaya
Efficiency within the advisory process in less time and at lower costs compared to
(2017); Moulliet et al. (2016a); Singh & Kaur (2017);
the bank advisor.
Sironi (2016); Mesbah et al. (2019)
Some process steps are unchangeable and standardized in their sequence
and the activities they contain. For this reason, every customer can Fisch et al. (2017); Jung et al. (2018a); Kaya (2017);
Standardization
always be guaranteed consistent quality and exclusion of human Singh & Kaur (2017)
advisory errors.
Bellani et al. (2016); D'Acunto et al. (2019); Gold &
Kursh (2017); Jung et al. (2018a); Jung et al., (2018b),
Transparent robo-advisors allow each step within the end-to-end process
Transparency Kaya (2017); Moulliet et al. (2016a); O'Keefe et al.
to be rationally and logically justified creating trust towards the user.
(2016); Ruf et al. (2015); Rühr (2020); Sander (2021);
Singh & Kaur (2017); Mesbah et al. (2019)
Content is presented in a modern way and customers are actively Becchi et al. (2018); Bellani et al. (2016); D'Acunto et
involved in the process, in which, e.g., they must make investment al. (2019); Fisch et al. (2017); Jung et al. (2018b); Kaya
Usability
decisions in a playful way based on sample situations and are thus (2017); O'Keefe et al. (2016); Ruf et al. (2015); Singh &
involved in determining their risk attitude. Kaur (2017); Sironi (2016)

Table 1. Process-relevant RA characteristics.

Quantifying the fit of the three AI application types in processes, the indicative correlations from Figure 3
are applied during the impact analysis. The correlation scale indicates the extent to which the respective AI
application type has the potential to introduce the respective RA characteristic into a process and
consequently improve it. The assignment of AI applications and RA characteristics was derived during the
expert interviews. The interfaces of both axes show either a very high, high, moderate, low, very low or no
correlation. The higher the sum of correlations per application type, the greater the effects on the respective
sub-processes to be analyzed in chapter 4.1. Therefore, it is expected that the intelligent CA has the highest
impact on the process, also because of its high relevance for client interaction, while the impact of CS and
PA is expected to have less direct and more indirect impact on the process through the intelligent CA
application.

Figure 3. Indicative correlation of AI application process optimization potentials.

6
 Implications of AI-based Robo-Advisory

Iterative Impact Analysis

Impact on Sub-Processes
The impact analysis is conducted by applying the three AI-based RA application types CA, CS and PA with
their individual causative characteristics onto the PB investment advisory process. The analysis of each
process step results in a transformation, which is expressed by either a
merger (due to AI application functionalities addressing process steps at the same time)
elimination (process steps that, due to AI application functionalities, become so inherent to other

shifting (execution of the process step elsewhere in the process)

or acceleration (faster execution of the process step due to AI application functionalities)

of sub-processes providing indications for process redesign which are based on the condition statements of
Netjes et al. (2008). Subsequently, each AI-based application type and corresponding transformations are
applied for process re-design. The concrete influences on the end-to-end process are presented in a matrix
(Table 2). The matrix representation offers the benefit of linking the induced transformations of the sub-
processes with the causative process characteristics and the underlying AI application types. The numbers
in the matrix represent the sub-processes in chronological order. For better readability, sub-processes that
merge are indicated with the same letter. The process transformations induced by the three AI application
types cause ten mergers, six eliminations, three shifts and nine accelerations of and within the investment
advisory sub-processes. Most process mergers occur in the initiation, profiling, concept and offer phases,
which are characterized by intense client-advisor interaction.

Induced transformation Addressed process-relevant robo-advisory characteristics AI application type

Process step
1 a x x x x x x x
2 b x x
Initiation 3 a x x x x
4 a x x x
CM1 x x x x x x
5 x x x x
6 x x x
7 x x x x x x
Profiling
8 b x x x x
9 c x x x x
10 x
11 c x x x x
12 x x x
13 d x x x x x
Conecpt
14 d x x x
15 x x x
CM2 x x x x
16 d x x x x
17 x x x x x
CMn x x x x
Offer 18 x x x x
19 x x x x x
20 x x x
21 x x x x x x
22 x x x x x
Implemen-
23
tation
24 x x x x x x x
25 x x x x x x x x
Maintenance
26 x x x x x x
Total 10 6 3 9 3 14 5 19 14 1 10 18 13 8

Table 2. Summary matrix of induced transformations on the sub-process level.

7
 Implications of AI-based Robo-Advisory

Transformative Process Change Analysis

Sub-processes like (1) initial customer contact, (3) clarification with compliance and (4) analysis of last
customer contact may be integrated since client conversations and segmentations can be analyzed in-time
and updated based on the customer profile saved. Also, the sub-processes (2) identification of customer
needs and (8) goal formulation merge since both can be derived with an intelligent CA. Sub-processes (9)
determination of restrictions and (11) creation of risk profile are integrated applying a CA combined with
PA for risk analysis. Finally, sub-processes (13), (14), (16) merge because the investment simulation and
proposal generation are conducted by applying PA. While sub-process eliminations and shifts are observed
through the whole process, accelerations take mostly place in the second part of the process where highly
standardized and administrative tasks such as (20) settlement of contract and (24) customer reporting are
located, which profit from digital interactions, digital data storage and the transmission of information and
insights between the three AI-based application types.
The changes are mostly initiated by CA (18) and CS (14), while the PA only causes eight process
transformations. An analysis of frequency distributions of process-relevant RA characteristics and induced
transformation per AI application type reveal further insights into the impact of RA onto the investment
advisory process (Figure 4). First, most sub-process mergers are based on the intelligent CA and CS. This
is because such systems are frontend systems affecting the activities and its sequences between the bank
advisor and the clients. Second, eliminations are relatively evenly distributed with CS in the lead. This
makes sense because customer needs are matched with the risk profile by the CS application type which
used to be extensive work for the advisor. On that basis, the application seamlessly identifies cross-
/upselling potential, creates an investment profile, and prepares a solution. Process accelerations are
mostly driven by the CA since it enables fast communication between clients and the bank. The majority of
investment advisory process changes are caused by the efficiency (15), automation (9) and (8)
standardization the CA offers. Besides of that, the standardization (10) and efficiency (8) of the intelligent
CS account mostly for sub-process mergers and eliminations, while PA brings efficiency (6), standardization
(4) and usability (4) into the process. In addition, usability is generated by all three AI application types for
both clients and advisors. In contrast to the beforementioned RA characteristics, accessibility (2),
availability (4) and transparency (1) only cause a minority of the overall process impact. The reasons lie in
the fact that accessibility and availability can only emerge at mandatory contact points. Transparency in the
investment advisory process is still provided by the advisor as studies imply (Rühr, 2020) and can only be
supplemented by AI-based application systems in (13) simulation of the best solution.

Figure 4. Frequency distribution of transformations and adressed characteristics.

8
 Implications of AI-based Robo-Advisory

Investment Advisory Process Redesign

The impact analysis results of the three AI-based applications CA, CS and PA imply a process redesign with
changes in both the sub-process sequence and the system-support. In contrast to the traditional PB
investment advisory process, the first customer meeting can be conducted either by a CA or bank advisor.
Sub-processes (1), (3), and (4) merge due to automation and efficiency gains from the use of intelligent CS
and the communication via CA. An analysis of existing customer data along with a set of non-compliant
examples allows to automatically estimate risky and flag fraudulent cases. The self-learning systems are
continuously improved through iterative feedback to evaluate unknown cases. AI-based chatbots and voice
assistants learn and memorize (2) individual client needs to support (8) goal formulation. Within client
profiling, CA-collected data is used to assign a client segment and for (5) positioning in the right phase of
life. With this, the combination of intelligent CS and CA allows (7) to identify and actively offer suitable
cross-/ upselling options. Further, the CA asks clients questions about risk tolerance and portfolio
restrictions in the combined sub-processes (9) and (11). The recorded information is then interpreted by CS
and possible market scenarios are contextualized for an accurate risk analysis. The final consolidated client
profile (10) is subsequently used to offer the client an investment solution. Compared to the traditional
advisory process, the sub-processes (13), (14), and (16) occur simultaneously due to efficiency gains. Here,
PA is used for financial asset selection and allocation. Considering the predicted returns, risks and the
-learning algorithms select the best fitting investments. Further suggestions for
improvement made by the customer can be discussed interactively in the subprocess (19) customer approval
for the overall solution using intelligent CAs. Once the investor is satisfied with the investment solution, a
standardized contract can be drafted. With the conclusion of the contract (20), the consultation ends.
Analogous to the traditional PB investment advisory, the following implementation of the agreed contract,
i.e. (22), (23) and (24), runs without further customer interaction. In the last process step, maintenance,
the forecasting capability of PA can be used for the successive assessment of the compatibility of the solution
and market development (25). As proposed by Yang and Almahdi (2017), portfolio profitability is
continuously checked, and alternative investment instruments are recommended to either the bank advisor
or the client (Figure 5).

Figure 5. The re-designed AI-based private banking investment-advisory process.

Evaluation
Four semi-structured, one-hour expert interviews were conducted and transcribed to critically assess the
results by applying an interview guideline, which participants received in advance. The questions focus on
the applicability of the three application types in the identified process steps and their impact on the process
and beyond. All interviewees work in the financial industry and have an association with AI and PB: Expert
1 works as head of innovation management at a German bank, expert 2 is head of digitization at a Swiss
private bank, expert 3 is head of business technology with a focus on AI and expert 4 is managing director
at a Swiss AI-based asset management Fintech company. The interview results largely confirm the process
redesign, but also reveal surprising findings. Regarding CA integration, experts 1, 3, and 4 saw potentials

9
 Implications of AI-based Robo-Advisory

the more complex

the product, the more likely a bank advisor is to add value. Expert 2 views CA-based client onboarding

bile device-implemented chatbots and voice assistants contribute to the

investor's independence of location and time-consuming phone calls or email queues (experts 1, 2 and 4).
Yet, efficiency gains of CA depend on domain knowledge and speech flow, since immature assistants
decrease client trust (expert 3 and 4). In addition, CAs in PB must follow HNWI regulations, which offers
additional potential for advisor support in the consultation documentation (expert 2). Besides of that, all
experts confirm the applicability of CS in customer acquisition but find external client data availability and
processing due to data protection regulations a barrier (expert 3). Moreover, product recommendations
based on CS promote service individualization and risk-adequate investments while improving the client
journey (expert 2 and 4). While expert 1 recognizes potential for CS-based, dynamic client segmentation,

different 3 and 4
mention that analyzing alternative and unstructured data helps to identify lucrative investment solutions.
Expert 4 thereby emphasizes the potential of standardizing processes internally which would enable banks
to process client information across channels and systems and hence individualize services and products
towards the clients. Nevertheless, the interviewees fear that efficiency gains tend to favor banks rather than
customers. AI-
internal bank tensions may arise as human advisors find themselves in competition with intelligent
applications and overwhelmed by the AI-based automation (expert 2 and 3). Expert 2 even warns against
automating organizations too quickly: In the case of a Swiss private bank, automation led to the loss of
employees because they felt overstrained by the rapid digitalization.

Discussion
Implications for AI-assisted Private Banking Investment Advisory
Based on the impact analysis and expert interviews, three implication dimensions for PB investment
advisory could be derived. First, AI-based automation is driven by the AI application types CA and CS which
standardize the process and hence enable advisory flexibility. On the one hand, CAs help the advisor to stay
informed about the clients by identifying needs and reporting them. This increases consulting service
efficiency, as the bank advisor can discuss uncommunicated needs with clients and proactively present
solutions. On the other hand, CS and partly PA drive process automation and relieve advisors of analytical
tasks, enabling an intense relationship management across the different channels as well as regular market
and product insights. All relevant AI-based market and product insights should be communicated regularly
or even ad hoc to the clients online or offline. Hence, advisory can focus on their banking professional role,
while the AI application types increasingly take over the analytical role. Second, AI-based process
automation requires strategic integration of the three AI-based application types in the form of an
omnichannel approach, which mainly evolves around CA. The degree of accessibility and availability that
CA brings into the process for both clients and advisors allows to connect different online and offline
channels. Nevertheless, advisors only remain the client focus if they are reachable through different direct
and indirect channels since the application type also provides clients with intelligent self-services such as
analyses and interpretations which are created by CS and PA systems. Hence, client journeys need to be
formalized to seamlessly integrate AI-driven insights for both clients and advisors. Direct channels such as
branch, telephone, email, and video conferencing are time-intensive from the advisors viewpoint but
remain important to establish personal client relationships. Above all, it is important to ensure that there
are no media discontinuities between the direct and indirect channels (e.g. online banking and bank-
specific apps), to maximize customer flexibility and experience. Third, an omnichannel approach that
integrates the three AI application types into a consulting-intensive process requires structured human-AI
interaction. The combined use of all three applications can contribute to improving efficiency, transparency
and usability, provided that the collaboration between humans and machines is purposeful and thus
sustainable. While advisors remain the face to the clients especially when it comes to complex products,
they increasingly generate value through collaboration with AI-based applications. To encourage clients to
use intelligent services, system design should place emphasis on usability to gain client trust regarding

10
 Implications of AI-based Robo-Advisory

innovative applications. Sequential decision-making between advisors and AI systems is proposed with the
advisor having a veto right, which is required by regulation to ensure transparency. If banks offer their
customers intelligent self-services, the definition of human-machine interaction at the customer-bank
interface is also necessary to provide seamless client journeys, avoid communication breaks between
solutions, advisors, and clients, as well as to build trust regarding AI use. The multiple channels require the
advisor to act as an internal and external network-coordinator and deliver hybrid, AI-assisted services. All
in all, CA has the overall highest impact on PB investment advisory since the process is communication-
intensive by nature and the NLP capabilities of CA are well-suited to support PB advisors. The effects of the
CS and PA application types, in turn, unfold in an indirect way, since these, especially CS, are for internal
bank use and, unlike CA, are not representing a direct customer interface. CS and PA, nonetheless, are the
underlying analytical components of the CA application type that ultimately lead to added customer value.
Hence, AI-assisted PB investment advisors are asked to (1) perform professional banking tasks such as risk
profile creation, (2) interpret AI-based analyses and interpretations to create valuable insights, (3)
coordinate between AI-based systems and interpersonal client actions and finally (4) integrate knowledge
into these systems. Moreover, client needs can be related to non-banking services and need to be
orchestrated together with other service providers. Due to the AI integration into the PB investment
advisory process, the advisor task range increases and raises the question whether the advisor role should
be split into one role covering banking professional tasks towards the client and a second role being
responsible for data analysis and interpretation.

Figure 6. Implications and client benefits on AI-assisted private banking investment

advisory.

Limitations and Future Research

As a first limitation, the process impact analysis, redesign and implications presented are tailored for PB
and are not generally applicable to retail banking as other AI-based transformations occur at the sub-
process level due to the customer segment. Nevertheless, the results can be used for the design of AI-based
service processes. Second, the process impact analysis and re-design is based on a reference process which
can vary in other financial institutions. Hence, the present study is not replacing a detailed analysis of
individual processes. Third, the three AI application types are abstract, generic descriptions of varying
degrees of complexity, which only allow for implications at the conceptual level. For the redesign of
company-specific processes, the concrete applications and their respective process-related effects must be
analyzed. While the process re-design indicates usage of the AI application types CA, CS and PA in private
banking investment advisory, the paper is not providing concrete human-AI interactions along the sub-
processes. The work of Dellermann et al. (2019) on human-AI collaboration can be used as a foundation.
Further research could also examine the proposed sequential decision-making design between AI systems
and advisors and investigate effects on the system adoption from both the client and advisor perspective,
e.g. based on the research of Shrestha et al. (2019). Additionally, client trust towards AI usage within the
process should be analyzed. Since AI-based systems affect the available channels, customer experience
should be investigated, and appropriate multichannel solutions designed. Besides of that, further research
could focus on required PB advisory skills in the newly designed process. Moreover, AI system design
research must consider that the human advisor may suffer from both a high degree of automation and
information overload within the AI-based process. The latter has already been indicted by the research

11
 Implications of AI-based Robo-Advisory

results of ([blinded for review]). Future research could therefore address how to increase employee well-
being in AI-based workflows so that employees achieve long-term success in collaboration with the
algorithms. Approaches for such research include the work of Kellogg et al. (2020), which describes the "6
Rs" (recommending, restricting, recording, rating, replacing, and rewarding) of algorithmic control of
employers, which in turn elicit 6 corresponding positive and negative experiences among employees.

Conclusion
This article presents an AI process impact analysis of RA-based AI application types on the private banking
investment advisory process developed by Nueesch et al. (2016). The impact analysis was performed by
examining three AI-based RA application types and the associated seven process-relevant RA
characteristics for their transformation potential in the sub-processes, which can manifest itself in a merger,
elimination, shifting, or acceleration of the respective sub-processes. The impact analysis results imply that
efficiency, standardization, usability, and automation gains induced by the AI application types trigger
process changes. Additionally, the expert interviews largely confirmed the proposed process redesign and
three organizational implications of AI for PB are derived. First, AI-based and automated systems support
client advisors in analytical tasks and enable them to act more flexible towards the client through different
direct and indirect channels. Hence, the bank advisor role may be divided into one banking-professional
and one analytical role with the latter role eventually being covered by an AI system. Second, AI-based
process automation and the provision of intelligent self-services require an omnichannel approach. This
aspect is crucial to enable individual client journeys and create value by providing both bank advisors and
clients with AI-based analyses and interpretations. Third, AI-based automation and the integration of an
omnichannel approach demand active design of task-specific human-AI interaction towards bank advisors
and clients. Overall, the advisor should remain the face to the client and orchestrate identified client needs
also beyond banking products and services. Clients are expected to use AI systems depending on product
complexity, system usability and trust in the solution. To avoid a decrease of client trust through AI
implementation, a sequential decision-making structure from AI to human should be integrated and
customer advisors as well as clients should thus be assigned a veto right for the AI applications. If private
banks consider the identified implications, the use of AI in PB investment advisory can generate added
value in the form of personalized advice and individualized product and service offerings. It should be noted
that the PB advisor will have an even more comprehensive role, as a better understanding of customer needs
beyond financial aspects will emerge because of AI systems. Hence, the AI-driven disruptions will further
open the banking industry and require advisors to increasingly operate as coordinators in the internal and
external network to generate client value.
From a scientific perspective, the present study contributes to the business process redesign research
stream by providing (1) AI application and application-related process characteristics on the example of
RA, (2) an impact analysis approach to assess AI-based business process redesign and (3) guidance for AI-
driven organizational design on the example of industry-specific implications. Furthermore, the analysis of
the four sub-process transformations merger, elimination, shifting and acceleration implies that AI-driven
process redesign is mostly induced by interaction-intense sub-processes affected by CA. By applying AI
application characteristics and the four mentioned process transformations, the impact analysis approach
extends knowledge on the technology-induced business process redesign (Mansar and Reijers, 2007). The
generated knowledge may serve research and practice as a fundament for investigations on AI deployment
in advisory processes. Moreover, practitioners receive a blueprint for a structured AI impact analyses for
process redesign that is based on three AI application types and exemplary application-related process
characteristics that allow companies to derive organizational implications along the causal chain of the
impact analysis. Additionally, the present study results provide private banking practitioners with concrete
suggestions for the future of private banking investment advisory, where advisors interact and decide
together with AI-based systems.

References
a/pdf-
79/accenture-pov-digital-wealth-management-private-banking-in-your-pocket.pdf).

12
 Implications of AI-based Robo-Advisory

nvestment Decisions with Robo-advisors: The

Role of Anthropomorphism and Personalized Anchors in Recommendations
Proceedings, May 15. (https://aisel.aisnet.org/ecis2019_rp/33).
-Return Portfolio
Optimization Using Recurrent Reinforcement Learning with Expected Maximum Dra
Systems with Applications (87), pp. 267 279. (https://doi.org/10.1016/j.eswa.2017.06.023).

Electronic Markets (28:3), pp. 235 243. (https://doi.org/10.1007/s12525-018-0310-9).

-Oriented Banking - Electronic Markets Are Paving
215.
(https://doi.org/10.1007/s12525-012-0106-2).
Althelaya, K. A., El-Alfy, E.- -
and Long-
Communication Systems (ICICS), , April, pp. 151 156. (https://doi.org/10.1109/IACS.2018.8355458).
- -Advisory, Cham: Palgrave
Macmillan, pp. 227 243.
elligence in FinTech: Understanding Robo-

Publishing Limited, pp. 1411 1430. (https://doi.org/10.1108/IMDS-08-2018-0368).

pp. 93 113.

(https://www.bcg.com/publications/2019/what-does-personalization-banking-really-mean).
Bruckes,
Adopting Robo-

(https://stories.wf.com/helpful-banking-assistanton-facebook/).

Transformation (44), pp. 45 57.

-Advice On Financial Advisers: A Qualitative Case

Davenport, T. H., and Ronanki

Review (96:1), pp. 108 116.
Day, M. Y., Lin, J. T., and Chen, Y. C. 2018. Artificial Intelligence for Conversational Robo-Advisor,
presented at the Proc. of the 2018 IEEE/ACM Int. Conf. on Advances in Social Networks Analysis and
Mining (ASONAM), Barcelona, Spain, pp. 1057 1064.
Day, M.-Y., and Lin, J.-
ternational Conference on Advances in Social
Networks Analysis and Mining, Vancouver British Columbia Canada: ACM, August 27, pp. 1026 1033.
(https://doi.org/10.1145/3341161.3344822).
Dellermann, D., Calma, A., Lipusch, N., Weber, T., Weigel, S., and Ebel, P. 2019. -AI

Proceedings, pp. 274 283.

Eickhoff, M., Muntermann, J., and Weinrich, T. 2017.
FinTech B
-
Fein Law Offices.
- e Disruptive
Impact of FinTech on Retirement Systems, Oxford University Press, pp. 13 37.
(https://doi.org/10.1093/oso/9780198845553.003.0002).
-
Fulk, M., Grable, J. E., C. F. P., Watkins, K., and -Advisory Services, and
188.
Gu, C.-S., Hsieh, H.-P., Wu, C.-S., Chang, R.-I., and Ho, J.- -Advisor with
Deep-
and Cybernetics (SMC), , October, pp. 2845 2850. (https://doi.org/10.1109/SMC.2019.8914183).

13
 Implications of AI-based Robo-Advisory

Influencing Factors in Robo-Advisor

MIS Quarterly (28:1), Management Information Systems Research Center, University of Minnesota,
pp. 75 105. (https://doi.org/10.2307/25148625).

Experience, Firm Percep

Marketing Science (49:4), pp. 659 676. (https://doi.org/10.1007/s11747-020-00753-z).
Hoelscher, R., and Nelde, M. 2018. -Advisor
Zeitschrift Für Das Gesamte Kreditwesen (2), pp. 68 73.

Robo-Advisory: Digitalization and Automation of

86.
(https://doi.org/10.1007/s12599-018-0521-9).
Jung, D., Dorner, V., Weinhardt, C., and Pusmaz, H. 2018. -Advisor for Risk-Averse,
Low- 380. (https://doi.org/10.1007/s12525-
017-0279-9).
-Advisory: Opportunities and Risks for the Future of
dvances in Consulting Research: Recent Findings and Practical Cases,
Contributions to Management Science, V. Nissen (ed.), Cham: Springer International Publishing, pp.
405 427. (https://doi.org/10.1007/978-3-319-95999-3_20).
Jung, D., and Weinhardt, C. 201 -Advisors and Financial Decision Inertia: How Choice Architecture

(https://aisel.aisnet.org/icis2018/hcri/Presentations/4).
Kellogg, K. C., Valentine, M. A., and Christin,
410.
(https://doi.org/10.5465/annals.2018.0174).
Khadivizand, S., Beheshti, A., Sobhanmanesh, F., Sheng, Q. Z., Istanbouli, E., Wood, S., and Pezaro, D.
-
in Proceedings of the 18th International Conference on Advances in Mobile Computing & Multimedia,
Chiang Mai Thailand: ACM, November 30, pp. 74 83. (https://doi.org/10.1145/3428690.3429172).
Kilic, M., Dolata, M., and Schwabe, G. 2017. Why Do You Ask All Those Questions? Supporting Client
Profiling in Financial Service Encounters, presented at the Hawaii International Conference on System
Sciences. (https://doi.org/10.24251/HICSS.2017.572).
Kilic, M., Heinrich, P., and Schwabe, G. 2015.

Social Computing, Vancouver BC Canada: ACM, February 28, pp. 1324 1335.
(https://doi.org/10.1145/2675133.2675289).

Kothari, K., and

(https://worldwealthreport.com/wp-content/uploads/sites/7/2020/12/Top-Trends-2021-in-Wealth-
Management.pdf).
Law, K. S., and Chung, F.- -Driven Decision Analytics for Commerc
Journal of Management Analytics (7:2), pp. 209 230.
(https://doi.org/10.1080/23270012.2020.1734879).
444.
(https://doi.org/10.1038/nature14539).
Li, R., K

Liao, L., Ma, Y., He, X., Hong, R., and Chua, T.- -
in Proceedings of the 26th ACM International Conference on Multimedia, Seoul Republic of Korea:
ACM, October 15, pp. 801 809. (https://doi.org/10.1145/3240508.3240605).
-
Identifying and Streamlining Low Risk Customers or Applicants: Techniques &

14
 Implications of AI-based Robo-Advisory

Mesbah, N., Tauchert, C., Olt, C., and Buxmann, P. 2019. -

AMCIS 2019 Proceedings.
(https://aisel.aisnet.org/amcis2019/ai_semantic_for_intelligent_info_systems/ai_semantic_for_int
elligent_info_systems/6).

Decision-Making with Robo- CIS 2020 Proceedings, , June 15.

-Advisory in Wealth

-Income Ratios and Robo-Advisory: Why

Wealth Managers Need to Engage with Robo-
Nueesch, R., Puschmann, T., and Alt, R. 2014. -Supported Customer Advisory:

Nueesch, R., Zernd

Nüesch, R., Alt, R., and Puschmann, T. 2015.

Systems Engineering (57:1), pp. 73 78. (https://doi.org/10.1007/s12599-014-0366-9).

Using Twitter to Predict Returns, Volatility, Trading Volume an

Systems with Applications (73), pp. 125 144. (https://doi.org/10.1016/j.eswa.2016.12.036).
-Enabled Robotic Advisory for Personalized
2020 Proceedings.
(https://aisel.aisnet.org/amcis2020/ai_semantic_for_intelligent_info_systems/ai_semantic_for_in
telligent_info_systems/1).

Methodology for
pp. 45 77. (https://doi.org/10.2753/MIS0742-1222240302).
-
Machine Learning for Asset Management - 328.
-Traded Funds: A New Investment Option for Taxable
427.
Renaux, M., Rudnicki, M., Beniere, A., and Lorain, -Advisory Reshape Asset

-Advisor Configuration: An Investigation of User Preferences and the Performance-

Rühr, A., Berger, B., and Hess, T. 2019. -Advisor? Trade-Offs in Automation and

23.
Russell, S. J., and Norvig, P. 2016. Artificial Intelligence: A Modern Approach, Malaysia: Pearson Education
Limited.
Sachse, S., Puschmann, T., and Alt, R. 2012. -Oriented Electronic Markets A Survey

Schabicki, T., Quint, Y., and -) Launching a Robo- -

Advisory, Cham: Palgrave Macmillan, pp. 163 185.
Schindler, J., de Souza Moraes, A., Li, S.-P., Bruno, G., Marcucci, J., Moscatelli, M., Yamaoka, H., Elizondo,
A., ter Braak, M., Pang

Stability Board.
Shrestha, Y. R., Ben-Menahem, S. M., and von Krogh, G. 2019. -Making Structures
83.
Sironi, P. 2016. FinTech Innovation: From Robo-Advisors to Goal Based Investing and Gamification,
Chichester: Wiley.
Statista -Advisor - -
advisors/worldwide).
-Advisors - -
finance/robo-advisors/deutschland).
Tauchert, C.,
Robo-Advisors, ICIS 2019 Proceedings.

15
View publication stats

Implications of AI-based Robo-Advisory

pp. 70 84.
Tsiptsis, K., and Chorianopoulos, A. 2009. Data Mining Techniques in CRM: Inside Customer
Segmentation, Chichester: Wiley.
Wang, P.-Y., Liu, C.-S., Yang, Y.-C., and Huang, S.- -Advisor Design Using Multiobjective
RankNets with Gated Neural Network S
(ICA), pp. 77 78. (https://doi.org/10.1109/AGENTS.2019.8929188).

2020 TREOs. (https://aisel.aisnet.org/treos_amcis2020/8).

Management Information Systems Research Center, University of

Minnesota, pp. xiii xxiii.

pp. 470 475. (https://doi.org/10.1057/ejis.2008.44).

Woodyard, A -
Financial Service Professionals (72:5), Society of Financial Service Professionals, pp. 56 66.
ed on Financial Social Network for
Robo- 54535. (https://doi.org/10.1109/ACCESS.2018.2871131).
-Chitty-
Twenty-Seventh International Joint Conference on Artificial Intelligence, Stockholm, Sweden:
International Joint Conferences on Artificial Intelligence Organization, July, pp. 5520 5526.
(https://doi.org/10.24963/ijcai.2018/778).

16