2020, Agostinelli
2020, Agostinelli
previously performed by a human users on the interface of a sualize screen display data from legacy applications (having
computer system, and then emulate their enactment in place no means for automated interfacing) in order to display such
of the user by means of a software robot. Nowadays, in the data using modern UIs. The strength of RPA is that it does
BPM domain, only simple, predictable business processes in- not replace existing applications or manipulate their code,
volving routine work can be automated by RPA tools in situ- but rather works with them similarly to a human user.
ations where there is no room for interpretation, while more In recent years, there has been an increased interest
sophisticated work is still left to human experts. In this paper,
starting from an in-depth experimentation of the RPA tools
around RPA in the Business Process Management (BPM)
available on the market, we provide a classification frame- domain (Kirchmer 2017). BPM is an active area of research
work to categorize them on the basis of some key dimensions. based on the observation that each product and/or service
Then, based on this analysis, we derive four research chal- that an organization offers is the outcome of a number of
lenges and discuss prospective approaches necessary to inject performed activities. Business processes (BPs) are the key
intelligence into current RPA technology, in order to achieve instrument for organizing such activities and improving the
more widespread adoption of RPA in the BPM domain. understanding of their interrelationships. Nowadays, BPs are
enacted in many complex industrial (e.g., manufacturing, lo-
gistics, retail) and non-industrial (e.g., emergency manage-
Introduction ment, healthcare, smart environments) domains through a
The recent developments in Artificial Intelligence (AI) force dedicated generation of information systems, called Process
us to continuously revisit the debate on what should be Management Systems (PMSs) (Reichert and Weber 2012).
automated and what should be done by humans. One of However, while conducting a BPM project is often consid-
these developments is Robotic Process Automation (RPA), a ered too expensive because its “top-down” approach that
fast-emerging automation approach that use software robots forces to develop the PMS from scratch (and system inte-
(or simply SW robots) to mimic and replicate the exe- gration is expensive), the promise of RPA is to rely on an
cution of highly repetitive tasks performed by humans in approach where, instead of redesigning existing information
their application’s user interface (UI). SW robots are mainly systems (that remain unchanged), humans are replaced by
used for automating office tasks in operations like account- SW robots in the execution of those BPs involving routine
ing, billing and customer service. Typical tasks are: ex- work. This allows knowledge workers to have more time for
tract semi-structured data from documents, read and write value added tasks. In the research literature, a number of
from/to databases, copy and paste data across cells of a case studies have shown that RPA technology can concretely
spreadsheet, open e-mails and attachments, fill in forms, lead to improvements in efficiency for BPs involving routine
make calculations, etc. (Willcocks, Lacity, and Craig 2015; work in large companies, such as O2 and Vodafone (Lacity,
Willcocks 2016). Willcocks, and Craig 2015; Aguirre and Rodriguez 2017;
Despite the capabilities of SW robots, the RPA technol- Geyer-Klingeberg et al. 2018).
ogy is still considered to be in its infancy (van der Aalst, Despite this growing attention around RPA, when con-
Bichler, and Heinzl 2018), even if similar solutions have sidering the state-of-the-art technology, it becomes appar-
been around for a long time. For instance, since the mid- ent that the current generation of RPA tools is driven by
nineties, closely related to SW robots, chatbots have been predefined rules and manual configurations made by ex-
used for years to accept voice-based or keyboard inputs pert users rather than by AI (Lohr 2018). According to
and guide customers to find relevant information in web- (van der Aalst, Bichler, and Heinzl 2018), to achieve a more
based applications (Hill, Ford, and Farreras 2015). Similarly, widespread adoption in the BPM domain, RPA needs to be-
in the same years, there has been some interest in the fi- come “smarter”. In a nutshell, with the use of AI techniques,
nancial industry around a preliminary form of RPA, called more complex and less defined BPs could be fully supported
Straight Through Processing (STP) (van der Aalst and van by the RPA technology. For example, by observing human
Hee 2004). The objective of STP was to speed up financial problem resolving unexpected system behavior (e.g., in case
of system errors, changing forms, etc.), RPA tools can adapt • Rapise6
and learn how to handle non-standard cases. After all, the • TagUI 7
Gartner Hype Cycle for AI published in 20181 places RPA
as one of the technologies at the peak of the hype cycle, • UiPath8
meaning that there are nowadays deep expectations on what • VisualCron9
RPA will be able to deliver to the AI community. • WinAutomation10
Starting from the above discussion, in this paper we first
identify and test ten RPA tools available on the market and • WorkFusion11
categorize them by means of a classification framework. The We analyzed each of the above tools leveraging a ded-
results of the classification allow us to derive four research icated case study based on a Purchase-to-Pay process ob-
challenges and identify prospective approaches required to tained from a SAP ERP system (the same one used in
evolve RPA towards AI in the context of BPM. (Geyer-Klingeberg et al. 2018)), which includes many stan-
dardized and highly repetitive transactions with potential
Background for automation. The data of the case study covers around
400.000 purchase orders created over one year, and can be
The state-of-the-art in RPA is plenty of recent works that accessed after registration on the Celonis Academic Cloud
are focused on optimizing specific BPM aspects of a RPA (SE 2019).
project. The work of Bosco et al. (Bosco et al. 2019) fo- After selecting the target process to automate, we em-
cuses on the automated identification of candidate routines ployed the selected tools to design and train various SW
to robotize. The work of Gao et al. (Gao et al. 2019) pro- robots, by recording the manual steps of the process. This
poses a self-learning approach to automatically detect high- has allowed us to identify a list of common tasks that must
level RPA-rules, from captured historical low-level user be- be performed to conduct a RPA project:
havior. The work of Jimenez-Ramirez and Reijers (Jimenez-
Ramirez et al. 2019) proposes a method for the early stages 1. Determine which process steps (also called routines) are
of a RPA project. The work of Leno (Leno et al. 2019) good candidates to be automated.
presents a method to record the performed user actions in 2. Model the selected routines in the form of flowchart di-
Excel and Chrome Web browser in a log, in order to en- agrams, which involve the specification of the actions,
able process mining for RPA. Finally, the works (Lacity, routing constructs (e.g., parallel and alternative branches),
Willcocks, and Craig 2015; Aguirre and Rodriguez 2017; data flow, etc. that define the behavior of a SW robot.
Geyer-Klingeberg et al. 2018) discuss the enactment of three 3. Record the mouse/key events that happen on the UI of
different RPA case studies in large companies. the user’s computer system. This information is associ-
Excluding the works (Bosco et al. 2019) and (Gao et al. ated with the actions of a routine, enabling it to emulate
2019), the majority of the above contributions focus on re- the recorded human activities by means of a SW robot.
fining some existing features of a RPA project, while none of
them is targeted to identify and tackle the concrete research 4. Develop each modeled routine by generating the software
challenges of RPA to enable its adoption in BPM on a large code required to concretely enact the associated SW robot
scale. To fill this gap, the first step of this contribution is to on a target computer system.
identify and test the real RPA tools available on the market 5. Deploy the SW robots in their environment to perform
categorizing them by means of a classification framework. their actions. According to (Jimenez-Ramirez et al. 2019),
Most of the actual deployments of RPA are industry- it should be noted that RPA is often characterized by the
specific, e.g., for financial and business services (Tornbohm lacking of a testing environment; only the production en-
2017). According to (AI-Multiple 2019), nowadays, the vironment is available.
market of RPA solutions includes more than 50 vendors de- 6. Monitor the performance of SW robots to detect bottle-
veloping tools having different prices and features. Among necks and exceptions.
them, we identified 10 vendors that offer to freely try their
RPA tools, i.e., without the need to pay any license. The RPA 7. Maintain the routines, which takes into account each SW
tools in question are: robot’s performance and error cases. The outcomes of this
phase enable a new analysis and design cycle to enhance
• Automation Anywhere2 the SW robots (Jimenez-Ramirez et al. 2019).
• AssistEdge3
Classification Framework
• G1ANT 4
We tested the selected RPA tools with our case study per-
• Kryon5 forming the tasks to conduct a RPA project. This has allowed
1 6
https://www.gartner.com/en/documents/3883863-hype-cycle- https://www.inflectra.com/Rapise/
7
for-artificial-intelligence-2018 https://github.com/kelaberetiv/TagUI
2 8
https://www.automationanywhere.com/ https://www.uipath.com/
3 9
https://www.edgeverve.com/assistedge/ https://www.visualcron.com/
4 10
https://g1ant.com/ https://www.winautomation.com/
5 11
https://www.kryonsystems.com/ https://www.workfusion.com/
Table 1: Results of the application of the classification framework
Tool SW Arch. Coding Recording Self Learning Autom. Routine comp. Log
Cl.-Server St.-alone Strong Low GUI Web Desktop Others Intra-rout. Inter-rout. type Manual Autom. quality
Aut. Anyw. X X X X X X Hybrid X FFF
AssistEdge X X X X Hybrid X F
G1ANT X X X X Hybrid X F
Kryion X X X X X Hybrid X FF
Rapise X X X X X Hybrid X F
TagUI X X X X Hybrid X F
UiPath X X X X X X Hybrid X FF
VisualCron X X Attended X FFF
WinAutom. X X X X X Hybrid X FF
WorkFusion X X X X X X Hybrid X FF
us to realize a classification framework for RPA tools, which • Routine composition: The ability of the RPA tool to or-
consists of the following key dimensions: chestrate, at run-time, through manual support or in an
automated way different (single) routines associated with
• Software (SW) Architecture: The specific SW architec-
different SW robots, when large workflows need to be em-
ture adopted by the tool: either Stand-alone or Client-
ulated.
Server.
• Log quality: The quality of the logs recorded by the
• Coding features: The behavior of SW robots can be clas-
RPA tool (called RPA logs). Since routines consist of col-
sified as:
lections of activities to be enacted according to certain
– Strong coding: based on the realization of explicit routing constraints, logs produced by RPA tools resemble
programming scripts, often with the support of a event logs in process mining. To this end, we measure the
command-line interface (CLI), which instructs the SW quality of such logs using the classification provided in
robots about the routines to emulate; the Process Mining Manifesto (van der Aalst et al. 2012),
– Graphical User Interfaces (GUIs): user friendly envi- where five maturity levels are defined, ranging from logs
ronments providing drag & drop facilities to build the of excellent (FFFFF) to poor quality (F).
flowchart of the routines to emulate; Table 1 shows the results of the application of our classifi-
– Low-code tools: GUIs that – in addition to drag & drop cation framework to the selected RPA tools. The following
facilities – provide low-coding functionalities to semi- aspects become apparent: the majority of the tools provide
automatically create software code. (i) a Client-Server SW architecture, (ii) GUIs with drag &
drop facilities and low-code functionalities, (iii) both web
• Recording facilities: The actions performed by a human
and desktop recording, (iv) a hybrid automation type, (v)
within the software tool can be recorded with:
manual-based features to achieve routine composition, (vi)
– Web recording: detection of user actions performed on logs of poor quality. Interestingly, differently from the other
a web browser; tools, G1ANT and TagUI offer strong-coding functionalities
– Desktop recording: detection of user actions performed with a basic CLI to support the programming of SW robots.
on a desktop UI; Finally, there is no tool that provides self learning or auto-
– Others: some RPA tools do not support neither web mated routine composition features.
nor desktop recording. Nonetheless, they offer record-
ing tools that work on specific applications only, such Research Challenges and Approaches
as Excel, Acrobat, SAP and Citrix. Some RPA tools On the basis of the results discussed in the previous sec-
provide also traditional screen-scraping recording. tion, we have derived four research challenges (and potential
• Self Learning: The ability of the RPA tool to automati- approaches to tackle them) necessary to inject intelligence
cally understand which user actions belong to which rou- into the current RPA technology towards a better support
tines (Intra-routine learning), and which routines are good to BPM. The four identified challenges, which will be ex-
candidates for the automation (Inter-routine learning). plained in the rest of the section, are: (i) Intra-routine Self
Learning, (ii) Inter-routine Self Learning, (iii) Automated
• Automation type: SW robots can either interact with generation of flowcharts, and (iv) Automated routine com-
users and/or act independently. This leads to three differ- position.
ent categories of automation:
1. Intra-routine Self Learning (Segmentation).
– Attended: the SW robots constantly require interaction
Description: Logs recorded by RPA tools are character-
with the users;
ized by long sequences of actions and/or events that re-
– Unattended: the SW robots act like batch processes, flect a number of routine executions. A log can record
i.e., manual intervention is not desired. This is ideal for information about several routines, whose actions and
optimizing back-office work; events are mixed in some order that reflects the particular
– Hybrid: Combination of the two above categories. order of their execution by the user (Bosco et al. 2019). In
addition, the same routine can be spread across multiple To this end, the method combines a technique for com-
logs, making the automated identification of routines far pressing a set of routines into an acyclic automaton, with
from being trivial. techniques for rule mining and for discovering data trans-
Objective: Identify the routines to be (potentially) emu- formations.
lated through software robots by looking at the RPA logs 3. Automated generation of flowcharts.
that keep track of the user actions taking place during a Description: In RPA tools, there is a lacking of testing en-
run of the system. This issue is known as “segmentation”. vironments. As a consequence, SW robots are developed
Approaches: One possible approach to tackle this chal- through a trial-and-error approach consisting of three
lenge is to rely on log analysis solutions in the Human- steps that are repeated until success (Leno et al. 2018):
Computer Interaction (HCI) field (Dumais et al. 2014; (i) First, a human designer produces a flowchart diagram
Dev and Liu 2017; Marrella and Catarci 2018), which fo- that includes the actions to be performed by the SW robot
cus on identifying frequent user tasks inside logs consist- on a target system; (ii) Second, SW robots are typically
ing of actions at different granularity. Alternately, local deployed in production environments, where they inter-
process mining approaches (Tax et al. 2016) or sequential act with information systems, with a high risk of errors
pattern mining (Dong 2009) can be employed to identify due to inaccurate modeling of flowcharts; (iii) Third, if
sequential patterns of non-consecutive actions that tend SW robots are not able to reproduce the behavior of the
to be repeated multiple times across multiple logs (Bosco users for a specific routine, then the designer adjusts the
et al. 2019). An interesting recent approach is the one of flowchart diagrams to fix the identified gap. While this
Gao et al. (Gao et al. 2019), where the authors present a approach is effective to execute simple rule-based logic
learning-based approach that allows for completely auto- in situations where there is no room for interpretation, it
mated RPA-rule deduction, on the basis of captured his- becomes time-consuming and error-prone in the presence
torical low-level user behavior. However, to date, no avail- of routines that are less predictable or require some level
able solution exists that allows for automatically: (i) un- of human judgement. Indeed, the designer should have a
derstanding which user actions have to be considered in- global vision of all possible unfoldings of the routines to
side the log (separating noise from actions that contribute define the appropriate behaviors of the SW robot, which
to routines); (ii) interpreting their semantics on the ba- becomes complicated when the number of unfoldings in-
sis of their granularity and (iii) identifying which routines creases. In cases where the rule set does not contain a suit-
they belong to. Solving the above challenges would al- able response for a specific situation, robots allow for es-
low us to cluster all user actions associated with a routine calation to a human supervisor.
in a well bounded execution trace. Consequently, all such Objective: Once the routines to be automated and the user
execution traces would be organized into a routine-based actions that constitute them (i.e., the routine-based logs)
log. have been identified, the target is to automatically gener-
2. Inter-routine Self Learning (Automated identification ate the flowchart diagrams describing the behaviors of the
of candidate routines to robotize). SW robots required to successfully execute the routines.
Description: While existing RPA tools allow one to auto- Approaches: A possible solution to this challenge can be
mate a wide range of routines, they do not allow one to to resort on discovery algorithms from the process min-
determine in an automated way which routines are good ing field (van der Aalst 2016) and to automatically ex-
candidates for automation in the first place. tract flowcharts in the form of Petri nets/BPMN models
Objective: Given a list of routine-based logs, identify au- from routine-based logs. Thus, it is necessary to investi-
tomatically which routines are good candidates for being gate from the literature on process discovery (Augusto et
automated by RPA tools. al. 2019) which algorithms suit better to extract the base
structure of flowchart diagrams from a routine-based log.
Approaches: To date, current RPA tools provide very lim-
Since such discovered flowcharts will reflect real routine
ited support to this challenge, which is often performed
executions, they may contain/miss (un-)necessary user ac-
by means of interviews, walkthroughs, direct observation
tions with respect to the SW robot’s expected behavior.
of workers, and analysis of documentation that may be of
To mitigate this issue, it is possible to leverage another
poor quality and difficult to understand. This manual ap-
process mining technique, named trace alignment (Adri-
proach allows analysts to identify the most obvious rou-
ansyah, Sidorova, and van Dongen 2011), which would
tines, while it is not suitable to detect those routines that
allow us to properly clean the discovered flowchart dia-
are not executed on a daily basis or that are performed
grams, by filtering out the unwanted behaviors found in
across multiple business units in different ways. The work
the previous discovery stage.
of Jimenez-Ramirez and Reijers (Jimenez-Ramirez et al.
2019) proposes to mitigate this issue through a method to 4. Automated routines composition.
improve the early stages of the RPA lifecycle using pro- Description: In modern contexts, human operators usually
cess mining techniques (van der Aalst 2016). On the other enact not just single tasks but complex workflows, con-
hand, a potential concrete solution to tackle this challenge sisting of many interrelated routines. However, the cur-
is proposed by Bosco et al. (Bosco et al. 2019), where the rent RPA technology allows for developing SW robots for
authors present a method to analyze routine-based logs executing single, independent routines. Only manual sup-
in order to discover routines that are fully deterministic. port is provided to orchestrate multiple routines, i.e., the
Figure 1: Overview of the pipeline of potential approaches required to tackle the research challenges
Flowchart diagrams
Action 1
Routine-based Action 2
logs RPA Workflow
Action 3
RPA
logs Action 1
HCI
Process Automated
Process Mining
Mining Planning
Machine Learning
Action 1 Action 2
management of more complex workflows is completely tions may be missing or not recorded properly), since they
delegated to human supervisors. are mainly used for debugging purposes. Increasing the
Objective: Automated generation of RPA workflows con- quality of RPA logs is a fundamental prerequisite to properly
sisting of many interrelated routines. tackle the proposed research, which leverages a log analysis
Approaches: To synthesize complex workflows through to discover, identify, model and compose routines in an au-
an intelligent orchestration of the robots’ routines, auto- tomated way. To this end, RPA tools should aim at logs at
mated planning techniques in AI can be employed (Ghal- the highest possible quality level.
lab, Nau, and Traverso 2004). The application of plan- To mitigate this issue, we are currently developing an ac-
ning for tackling the composition issue has been already tion logger to be attached to the existing RPA tools, in order
proved to be successful in real world domains (Marrella to enable the creation of routines-based logs of an acceptable
2019). The idea is to consider the robots’ routines as black quality. Apart from the need to increase the quality of RPA
boxes, i.e., as planning actions with specific precondi- logs, as a future work, this research aims at also improv-
tions and effects, and to delegate to a planning system the ing the auditability (RPA logs are auditable), upgradability
generation of a proper strategy to automatically compose (flowchart diagrams describing SW robots’ behavior will be
them in a larger workflow that coordinates their orches- always updated to the current state of the system execution)
tration. and the resiliency (SW robots will be always upgraded to
deal with new behaviors, making them very robust to any
Figure 1 shows a graphical overview of the pipeline of contextual change that may arise during a routine execution)
potential approaches required to tackle the four identified re- of SW robots. Furthermore, scalability must be improved as
search challenges to achieve the aforementioned objectives. well. Human capacity is difficult to scale in situations where
demand fluctuates, instead SW robots operate at whatever
Discussion and Concluding Remarks speed is demanded by the work volume.
RPA recently gained a lot of attention in the BPM domain To conclude, we note that our study has a threat to valid-
(van der Aalst, Bichler, and Heinzl 2018). Since RPA oper- ity, since we analyzed only a sample of the RPA tools avail-
ates at the UI level, rather than at the system level, it allows able on the market. As a consequence, our findings can not
one to apply automation without any changes in the underly- be generalized beyond the scope of the tested RPA tools.
ing information system. Thus, the entry barrier of adopting Nonetheless, we consider this work as an important first
RPA in BPs that are already in place is lower compared to step towards the realization of intelligent solutions for RPA.
conventional BPM (Gao et al. 2019). However, the current Moreover, we also envision that this research will provide
generation of RPA tools is driven by predefined rules and long-term benefits on the companies workforce, e.g., by im-
manual configurations made by expert users rather than by proving the customer service in the front office while at the
AI (Lohr 2018), preventing a widespread adoption of these same time reducing the back office tasks.
tools in the BPM domain.
In this paper, we have tackled this issue starting from an References
in-depth experimentation of the RPA tools available in the
market. Then we have provided a classification framework [Adriansyah, Sidorova, and van Dongen 2011] Adriansyah,
to categorize them on the basis of some key dimensions and A.; Sidorova, N.; and van Dongen, B. F. 2011. Cost-Based
we have derived four research challenges and discussed po- Fitness in Conformance Checking. 2011 Eleventh Interna-
tential approaches necessary to inject intelligence into the tional Conference on Application of Concurrency to System
current RPA technology, from a BPM perspective. Design 57–66.
It is worth to notice that, according to Table 1, the logs [Aguirre and Rodriguez 2017] Aguirre, S., and Rodriguez,
produced by the tested RPA tools have a poor quality (ac- A. 2017. Automation of a Business Process Using Robotic
Process Automation (RPA): A Case Study. In Applied Com- [Kirchmer 2017] Kirchmer, M. 2017. Robotic Pro-
puter Sciences in Engineering, 65–71. Cham: Springer In- cess Automation-Pragmatic Solution or Dangerous Illusion.
ternational Publishing. BTOES Insights, June’17.
[AI-Multiple 2019] AI-Multiple. 2019. All 52 RPA Soft- [Lacity, Willcocks, and Craig 2015] Lacity, M.; Willcocks,
ware Tools and Vendors: Sortable List [2019]. https://blog. L. P.; and Craig, A. 2015. RPA at Telefonica O2. The
aimultiple.com/rpa-tools/. London School of Economics and Political Science.
[Augusto et al. 2019] Augusto, A.; Conforti, R.; Dumas, M.; [Leno et al. 2018] Leno, V.; Dumas, M.; Maggi, F. M.; and
Rosa, M. L.; Maggi, F. M.; Marrella, A.; Mecella, M.; and La Rosa, M. 2018. Multi-perspective process model discov-
Soo, A. 2019. Automated Discovery of Process Models ery for robotic process automation. CAiSE Doct. Cons.
from Event Logs: Review and Benchmark. TKDE 31(4). [Leno et al. 2019] Leno, V.; Polyvyanyy, A.; Rosa, M. L.;
[Bisbal et al. 1999] Bisbal, J.; Lawless, D.; Wu, B.; and Dumas, M.; and Maggi, F. M. 2019. Action logger: Enabling
Grimson, J. 1999. Legacy information systems: Issues and process mining for robotic process automation. In Proceed-
directions. IEEE Software 16(5):103–111. ings of the Dissertation Award, Doctoral Consortium, and
Demonstration Track at 17th International Conference on
[Bosco et al. 2019] Bosco, A.; Augusto, A.; Dumas, M.; Business Process Management, (BPM’19), Vienna, Austria,
La Rosa, M.; and Fortino, G. 2019. Discovering Au- 124–128.
tomatable Routines From User Interaction Logs. In 17th
International Conference on Business Process Management [Lohr 2018] Lohr, S. 2018. The Beginning of a Wave:
(BPM’19), Forum track, Vienna, Austria, 144–162. Cham: A.I. Tiptoes Into the Workplace. https://www.nytimes.com/
Springer International Publishing. 2018/08/05/technology/workplace-ai.html/.
[Dev and Liu 2017] Dev, H., and Liu, Z. 2017. Identifying [Marrella and Catarci 2018] Marrella, A., and Catarci, T.
Frequent User Tasks from Application Logs. In Proceedings 2018. Measuring the Learnability of Interactive Systems
of the 22Nd International Conference on Intelligent User In- Using a Petri Net Based Approach. In Proceedings of the
terfaces, IUI ’17, 263–273. New York, NY, USA: ACM. 2018 Designing Interactive Systems Conference, DIS ’18,
1309–1319. New York, NY, USA: ACM.
[Dong 2009] Dong, G. 2009. Sequence Data Mining. Berlin,
[Marrella 2019] Marrella, A. 2019. Automated Planning for
Heidelberg: Springer-Verlag.
Business Process Management. J. Data Semantics 8(2):79–
[Dumais et al. 2014] Dumais, S.; Jeffries, R.; Russell, D. M.; 98.
Tang, D.; and Teevan, J. 2014. Understanding User Behavior [Reichert and Weber 2012] Reichert, M., and Weber, B.
Through Log Data and Analysis. In Ways of Knowing in 2012. Enabling Flexibility in Process-Aware Information
HCI. New York, NY: Springer. 349–372. Systems - Challenges, Methods, Technologies. Springer
[Gao et al. 2019] Gao, J.; van Zelst, S. J.; Lu, X.; and van der Berlin Heidelberg.
Aalst, W. M. P. 2019. Automated robotic process automa- [SE 2019] SE, C. 2019. Academic Alliance. https://www.
tion: A self-learning approach. In On the Move to Mean- celonis.com/academic-alliance.
ingful Internet Systems: OTM 2019 Conferences, 95–112.
Cham: Springer International Publishing. [Tax et al. 2016] Tax, N.; Sidorova, N.; Haakma, R.; and
van der Aalst, W. M. 2016. Mining local process models.
[Geyer-Klingeberg et al. 2018] Geyer-Klingeberg, J.; Journal of Innovation in Digital Ecosystems 3(2):183–196.
Nakladal, J.; Baldauf, F.; Veit, F.; van der Aalst, W.; Casati,
[Tornbohm 2017] Tornbohm, C. 2017. Gartner market
F.; Conforti, R.; de Leoni, M.; and Dumas, M. 2018.
guide for Robotic Process Automation software. Report
Process Mining and Robotic Process Automation: A Perfect
G00319864. Gartner.
Match. In 16th International Conference on Business Pro-
cess Management (BPM’18), Dissertation/Demos/Industry [van der Aalst and van Hee 2004] van der Aalst, W., and van
track, Sidney, Australia, 124–131. Hee, K. 2004. Workflow management: models, methods, and
systems. MIT press.
[Ghallab, Nau, and Traverso 2004] Ghallab, M.; Nau, D.;
and Traverso, P. 2004. Automated Planning: theory and [van der Aalst et al. 2012] van der Aalst, W.; Adriansyah, A.;
practice. Elsevier. de Medeiros, A. K. A.; Arcieri, F.; Baier, T.; Blickle, T.;
Bose, J. C.; van den Brand, P.; Brandtjen, R.; Buijs, J.; Bu-
[Hill, Ford, and Farreras 2015] Hill, J.; Ford, W. R.; and Far- rattin, A.; Carmona, J.; Castellanos, M.; Claes, J.; Cook,
reras, I. G. 2015. Real conversations with artificial intel- J.; Costantini, N.; Curbera, F.; Damiani, E.; de Leoni, M.;
ligence: A comparison between human–human online con- Delias, P.; van Dongen, B. F.; Dumas, M.; Dustdar, S.;
versations and human–chatbot conversations. Computers in Fahland, D.; Ferreira, D. R.; Gaaloul, W.; van Geffen, F.;
Human Behavior 49:245–250. Goel, S.; Günther, C.; Guzzo, A.; Harmon, P.; ter Hofstede,
[Jimenez-Ramirez et al. 2019] Jimenez-Ramirez, A.; Rei- A.; Hoogland, J.; Ingvaldsen, J. E.; Kato, K.; Kuhn, R.; Ku-
jers, H. A.; Barba, I.; and Del Valle, C. 2019. A Method to mar, A.; La Rosa, M.; Maggi, F.; Malerba, D.; Mans, R. S.;
Improve the Early Stages of the Robotic Process Automation Manuel, A.; McCreesh, M.; Mello, P.; Mendling, J.; Mon-
Lifecycle. In 31st International Conference on Advanced tali, M.; Motahari-Nezhad, H. R.; zur Muehlen, M.; Munoz-
Information Systems Engineering (CAiSE’19), Rome, Italy, Gama, J.; Pontieri, L.; Ribeiro, J.; Rozinat, A.; Seguel Pérez,
446–461. Cham: Springer International Publishing. H.; Seguel Pérez, R.; Sepúlveda, M.; Sinur, J.; Soffer, P.;
Song, M.; Sperduti, A.; Stilo, G.; Stoel, C.; Swenson, K.; Ta-
lamo, M.; Tan, W.; Turner, C.; Vanthienen, J.; Varvaressos,
G.; Verbeek, E.; Verdonk, M.; Vigo, R.; Wang, J.; Weber, B.;
Weidlich, M.; Weijters, T.; Wen, L.; Westergaard, M.; and
Wynn, M. 2012. Process Mining Manifesto. In Business
Process Management Workshops, 169–194. Berlin, Heidel-
berg: Springer Berlin Heidelberg.
[van der Aalst, Bichler, and Heinzl 2018] van der Aalst, W.
M. P.; Bichler, M.; and Heinzl, A. 2018. Robotic Process
Automation. Business & Information Systems Engineering
60(4):269–272.
[van der Aalst 2016] van der Aalst, W. M. P. 2016. Process
Mining: Data Science in Action. Heidelberg: Springer, 2
edition.
[Willcocks, Lacity, and Craig 2015] Willcocks, L. P.; Lacity,
M.; and Craig, A. 2015. The IT Function and Robotic Pro-
cess Automation. The London School of Economics and
Political Science.
[Willcocks 2016] Willcocks, L. 2016. Service Automation :
robots and the future of work. Warwickshire, United King-
dom: Steve Brookes Publishing.