Application of Microservices for Digital Transformation of

Data-Intensive Business Processes

Jānis Grabis and Jānis Kampars

Information Technology Institute, Riga Technical University, Kalku 1, Riga, Latvia

Keywords: Business Process Redesign, Digital Transformation, Data Analysis Model, Microservices.

Abstract: Business processes are redesigned as a part of business process management lifecycle and data intensive

activities such as image processing, prediction and classification are increasingly incorporated into business

processes. Data intensive activities often involve usage of data analysis models. It is argued that successful

development and execution of data intensive business processes requires synchronization of business

process redesign and data analysis models development activities. The business process architecture

integrating core business process with data analysis model setup and updating sub-processes is developed.

Business process transformation stages for incorporating data-intensive activities are outlined. The process

redesign and execution is supported by the technical architecture based on microservices. An example of

business process redesign is discussed.

1 INTRODUCTION

Business processes are subject to continuous

improvement not least because of new technologies

becoming available. There have been multiple cycles

of technology driven business process redesign and

the most recent cycle is frequently referred as to

digital transformation (Zimmermann et al., 2016).

This cycle is characterized by increasing usage of

solutions enabling business process execution

autonomy what is often achieved by relying on

advanced data processing capabilities (Roedder et

al., 2016). Bringing these data processing

capabilities into business processes requires a new

set of development technologies. These technologies

should support simultaneous business process

redesign, development of complex data analysis

models and process execution software as well as

establishing appropriate infrastructure. These

activities concern various domains and require

domain specific knowledge not only during the

process redesign phase but also during business

process execution.

Business process management lifecycle models

and methodologies cover various stages of business

process development (De Morais et al., 2014). They

provide process redesign patterns and guidelines.

However, they provide limited guidance concerning

data processing issues and implementation aspects

are platform dependent. Traditional implementation

platforms such as ERP systems and Business

Process Management (BPM) suites are better

equipped for transaction processing rather than

analytical processing. Additionally, data analysis

models incorporated in data-intensive business

processes have their own lifecycle and often require

extensive computational resources at their disposal.

That requires integration of core business process

activities with activities associated with data

analysis and an ability to provision the required

computational resources.

Business process architecture (Weske, 2007) is

an approach allowing integration of various

dimensions of business process redesign and

execution. Cloud-computing and service-orientation

are two key technologies providing infrastructural

services for running computationally intensive

operations (Moreno-Vozmediano et al., 2013).

The overall goal of the proposed research is to

elaborate a framework for development of data-

intensive business processes. The framework

integrates core business process activities and

activities associated with data analysis for enabling

data-intensive activities (i.e., horizontal business

process architecture integration) as well as considers

development of appropriate computational

infrastructure (i.e., vertical business process

736

Grabis, J. and Kampars, J.

Application of Microservices for Digital Transformation of Data-Intensive Business Processes.

DOI: 10.5220/0006805207360742

In Proceedings of the 20th International Conference on Enterprise Information Systems (ICEIS 2018), pages 736-742

ISBN: 978-989-758-298-1

architecture integration). It focuses on process

redesign and implementation stages of the business

process management lifecycle.

The objective of this paper is to outline the

proposed transformation approach and to introduce

key components of business process architecture for

transformation of data-intensive business processes.

The data-intensive business process is a process

requiring data of various types and from different

sources as well as analytical data transformations to

guide and automate business process execution. In

this paper, conversion of paper-based documents

(e.g., travel receipts) into meaningful business data

is considered as an example of the data intensive

business process. Optical characters recognition is

used to convert document images into characters and

neural networks are used to extract business data

from the text.

The rest of the paper is structured as follows.

Section two introduces technologies used in

elaboration of the approach. The transformation

approach is presented in Section 3. An application

example is provided in Section 4. Section 5

concludes

2 BACKGROUND AND

REQUIREMENTS

2.1 Business Process Redesign

Business process is a sequence of activities

performed to achieve specific business goals.

Business processes are continuously improved

through their lifecycle (Van der Aalst et al., 2003).

The improvement cycle includes activities of

process design, modeling, implementation,

monitoring and optimization (De Morais et al.,

2014). Becker et al. (2011) provide guidelines for

the design of business process. The guidelines

include As-Is modeling, analysis of As-Is models,

To-Be modeling and optimization. Process

documentation, usage of reference models and

benchmarking as well as measurement and

simulation methods are used for these purposes.

The process improvement methods rely on a set

of general principles (De Morais et al., 2014) and are

often tailored to specific needs and organizations.

Therefore, there is a large variety of methods.

Reijers and Mansar (2005) define a set of heuristic

rules for business process improvement. The

qualitative evaluation of the heuristic rules is also

provided which helps business analysts to justify

their BP improvement decisions. Barros (2007)

attempts to formalize business improvement

guidelines as reusable patterns. These patterns can

be used to construct business process from existing

best practices. Damij et al. (2008) propose a Tabular

Activities Development methodology, which

addresses both business process improvement and

implementation of information system supporting

the improved business processes. Process simulation

is used to determine the process cycle time and

related measurements are one of the key parts of the

methodology. Dumas et al. (2013) categorize

various process improvement alternative and their

potential impact on process improvement as well as

advocates usage of quantitative models in business

process design. The review of BP improvement

approaches is provided by Zellner (2011). Sidorova

and Isik (2010) present an extensive review of

business process research. They have identified

business process design and business process on-

going management and control as two of the four

cornerstones of core business process research.

Decision Model and Notation allows representing

decision-making logics in business processes (Biard

et al. 2015; Mertens et al., 2017). Although the

authors acknowledge importance of quantitative

analysis and decision-making in business process

design and execution, integration of quantitative

models into business processes is rarely explore

what is especially important since quantitative and

analysis models have their own life-cycle, which

should be synchronized with business process life-

cycle. It is also desired that models should not be

developed for every client using a particular data

intensive business processes. The models are pre-

packaged or provided as a service and only their

configuration is required for particular clients.

The business process architecture is an approach

allowing to combine various dimensions of business

process redesign (Lapouchnian et al., 2015).

Lapouchnian et al. (2017) show that business

process architecture is suitable for design of business

processes requiring cognitive capabilities. The set of

related business processes include processes for

analytical model creation, validation and business

improvement.

Business processes are implemented and

executed using various technologies. ERP systems

provide monolithic business process development

environment, which is efficient for stable processes

while being difficult to modify and to implement

custom requirements. Business process management

(BPM) suites allow implementing custom processes

and jointly with service-oriented architecture create

Application of Microservices for Digital Transformation of Data-Intensive Business Processes

737

an environment support flexible modification of the

processes by selecting appropriate services.

However, BPM and to some extent SOA are not

well-suited for data analysis purposes, especially,

requiring integration of various data processing

technologies. Scalability is also achieved at the

expense of efficiency. Business intelligence

technologies are often used to develop data analysis

models though their integration in business

processes is often performed in off-line or ad-hoc

manner (Chou et al., 2005).

Microservices (Dragoni et al., 2017) is a

technology recently gaining prominence as self-

contained light-weight containers of business

services. They allow for high degree of modularity,

containers are created on-demand quickly and with a

little overhead as well as various data processing and

storage technologies can be utilized.

2.2 Requirements

The proposed research focuses specifically on

design and execution of data intensive business

processes. The existing research and industrial

experience suggest that the following requirements

should be considered:

• Support for development and tuning of data

analysis models - business process and data

analysis models’ life-cycles should be

synchronized and model development should be

a part of the overall business process redesign

and execution;

• Presentation and understanding of data analysis

models - user of data analysis model should

have utmost understanding of models used in

business process execution. That can be

achieved by explaining levers available for

controlling behavior of the models as well as by

providing facilities for experimenting with

models. The experimentation allows

identification of appropriate model usage modes

and understanding of model usage

consequences;

• Modification - data analysis models are subject

of frequent revision as more data become

available and models are refined. The new

models should be evaluated and integrated in

the business processes without affecting the rest

of the process;

• Scalability - data analysis models often require

significant computational resources and

decisions should be made during business

process execution for potentially large number

of users. That requires both vertically and

horizontally scalable execution environment;

• Flexibility - data analysis models, data used in

the models and model solving algorithms come

in different forms and the development and

execution environment should be flexible to

support usage of the most appropriate data

processing methods and technologies.

3 TRANSFORMATION

APPROACH

The transformation approach consists of three main

components: 1) business process architecture; 2)

transformation process; and 3) technical

architecture. The business process architecture

defines an interrelated set of sub-processes required

to design and run data-intensive business processes.

The transformation process defines steps to be

performed to redesign a traditional business process

into a data-intensive business process. The technical

architecture specifies technologies used to execute

data-intensive business processes.

3.1 Business Process Architecture

The business process architecture consists of three

sub-processes: 1) sub-process of setting up models

of data analysis; 2) actual execution of data-

intensive business process or operations; and 3) sub-

process of updating the data analysis models once

additional data are accumulated during the

operations (Figure 1).

Figure 1: Data-intensive business process architecture.

The operations sub-process is the core business

process providing the required business functions.

Many activities of this process rely on data analysis

models, which are invoked during the process

execution. These models are setup prior to their

usage. The setup sub-process might involve

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

738

configuration of several models for all data-intensive

activities in the business process. Assuming that a

number of data analysis models for solving the

decision-making problem are available, typical

activities of the setup process are model selection

(e.g., exponential smoothing or moving average for

forecasting purposes), selection of structural

parameters (e.g., input variables for a regression

model or layers for neural networks), estimation of

models’ parameters (e.g., estimation of regression

coefficients) and evaluation of modeling

performance. The evaluation activities are of

particular importance to ensure acceptance of data

analysis results and their adoption for business

process execution.

The setup sub-process requires training and test

data to estimate and to evaluate models,

respectively. The business process architecture

represents these as coming from data stores, which

physically could belong to different types of storage

facilities. It also uses configuration data. The

configuration data can be divided in three groups: 1)

models’ parameters; 2) performance indicators; and

3) design of experiments. The configuration data can

be manipulated by business process users. All

models’ parameters are not included in the

configuration data but specifically promoted

parameters important for experimenting, tuning and

understanding of the models. The performance

indicators set targets for both modeling and process

performance. That is important because in many

situations data analysis models can be used only if

reasonably accurate. The design of experiments is

necessary for evaluation purposes. That allows

business process users to gain understanding of

models’ behavior and implications of modeling

results.

The operations sub-model is business problem

specific. The data analysis models are invoked

during execution of its activities. The execution

results are stored as transactional records, which

could be used for further refinement of the models.

The refinement is handled by the update sub-

process. The main issues represented in the update

process are updating conditions (i.e., when to

replace existing models with the updated ones) and

handling of model life-cycle issues (e.g., model

versioning for long-running processes).

3.2 Transformation Stages

The transformation process defines steps of process

redesign into a data intensive business process and

processes execution. The transformation process

itself is a part of the overall business process

management lifecycle though other stages of this

lifecycle are not considered in this paper.

The following steps are performed to develop a

data intensive business processes:

• Define business process - representation of the

core business process model is developed

assuming that process identification has been

performed

• Identify data intensive activities - the process is

vetted to discover opportunities for automation

or improved decision-making due to data

analysis. A data intensive activity requires data

entities, which are not directly associated with

the data record currently processed by the

activity, and involves analytical

transformations;

• Integrate data - data sources for training data are

identified and data delivery channels are

established;

• Develop data analysis models - data analysis

models appropriate for a specific business area

are developed. The models are developed as

packaged products configured and tuned for a

particular case;

• Deploy models as microservices - the models

are packaged and deployed as microservices to

achieve high scalability and modifyability;

• Redesign process around data intensive

activities - introduction of data intensive

activities might result in significant changes in

the business process to achieve full benefits of

digital transformation;

• Integrate models in business process - bind data

intensive business process activities with

corresponding microservices and development

of data flows;

• Execute business process - the business process

is executed including invocation of the

microservices;

• Update data analysis models - as additional data

are accumulated as the result of business

process execution, the data analysis models are

updated to account for the latest tendencies.

Some of these steps are performed in parallel, for

example, data integration and development of data

analysis models.

3.3 Technology

The technical architecture (Figure 2) consists of four

main components: 1) operations execution

component; 2) model setup component; 3) model

execution component; 4) data storage. The

Application of Microservices for Digital Transformation of Data-Intensive Business Processes

739

operations execution module is a container for

running data intensive activities and invoking

microservices. It can be implemented using various

technologies suitable for developing business

processes. The model execution component is built

using microservices as lightweight containers of

executable data analysis models. The containers are

created and disposed on-demand depending on

computational requirements. The microservices are

decoupled from operation execution component by a

queue thus providing scalability and load balancing

capabilities. The model setup component also can be

implemented in various technologies. However, it is

proposed that notebooks are suitable technology

because they combine model development,

execution, experimentation and documentation thus

clearly explicating the data analysis model. The

model setup often results in a model specification

(e.g., PMML standard), which can be used to

configure the microservices. Data storage

technologies are selected depending on requirements

and include traditional relational databases, object

storages and document-oriented databases. All

components access the storage component though

isolated storage facilities usually are created for the

setup and operations purposes.

Figure 2: Main components used to create technical

architecture.

4 EXAMPLE

The proposed approach is illustrated using the travel

management processes as an example. Business

travelers are required to report their expenses after

returning from the trip what includes filling out

expense forms and submitting travel receipts (Figure

3.a). The current business process captures images

of travel receipts without much of valuable use of

these data and requires manual filling of the expense

report. Apparently, there are at least two data-

intensive activities in the current process, namely,

Gather receipts and Fill expense report. The images

captured during the Gather receipts activity can be

transformed into text making them available for

further processing. The text extracted from the

images can be analyzed to fill the expense report

automatically. The redesigned process is shown in

Figure 3.b. Character recognition (OCR) is used to

extract text. Neural networks are used to recognize

type of travel expenses, date and amount. The

training data needed for these purposes are a data set

containing images of travel receipts and actual

expenses associated with these images. The

extracted expenses still need to be verified by a

human to ensure correctness.

The setup sub-process starts with preparation of

input data including training and test data. These

data are either previously accumulated images of

travel receipts with known (manually filled) expense

data items or specifically generated data sets.

Assuming that neural networks are used to identify

expense data items, a structure of the neural network

is defined and training of the model is performed.

The final activity is evaluation of the accuracy of

expense data identification. The automated

identification is enabled only if this accuracy is

satisfactory as defined by performance indicators.

Every new enterprise using the travel expenses

reporting business process requires tailoring of the

model and the setup sub-process provides clear

Figure 3: Travel expenses reporting: a) original process; and b) operations sub-process of the redesigned process.

Capture travel

receipts images

Fill out travel

expenses form

Attach images

to the travel

report

Submit the

report

Capture travel

receipts images

and do OCR

Extract expenses from

the receipts and prefill

the report

Review the

report

Submit the

report

Data intensiveData intensive

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

740

Figure 4: Technical architecture (the numbers indicate the sequence of interactions).

guidance for that. The configuration parameters and

the design of experiments are of particular

importance because they allow for understanding of

the model and its behavior. In this case, the

estimated expenses extraction accuracy threshold is

one of the configuration parameters. If the estimated

accuracy is below the threshold then the extracted

data are not used and manual entry is requested.

Another configuration parameter for travel receipt

capture is the number of images taken of the single

receipt to improve OCR precision.

At the infrastructure level, the redesign process

is supported by a scalable data analysis solution

(Figure 4). The image capture activity creates an

image processing job and places it in the queue

(implement using RabbitMQ

). The images captured

are also stored in object storage implemented using

Swift

. The image to text conversion is realized as a

microservice using the Tesseract

library. Incoming

image conversion messages are pulled from the

queue by microservice instances and the instances

read image data from the object storage. The

recognized text is stored in the document-oriented

database (implemented using MongoDB

) and the

queue is notified that the text is available for further

processing. That triggers the expense data extraction

using neural networks (realized using TensorFlow

https://www.rabbitmq.com/

https://docs.openstack.org/swift/latest/

https://github.com/tesseract-ocr

https://www.mongodb.com/

https://www.tensorflow.org/

and dedicated neural networks (NN) are created for

each expense item). The neural network was trained

in the model setup component and production neural

networks are again implemented as microservices.

The extraction results are stored in the document-

oriented database and the data intensive activity

captures events indicating that extraction results are

available for verification.

Following the business process architecture

allows to redesign the business process as well as to

establish supporting processes for model setup and

update. The technical solution is highly scalable and

the components are decoupled one from another.

That allows to modify the components, to use

specialized technological components as necessary

and to integrate the solution in the enterprise’s

existing technological landscape. In this case, the

expense data extraction algorithm is regularly

modified to achieve better accuracy. Once the new

version of the algorithm is approved for application

it is deployed as a microservice without a delay.

5 CONCLUSION AND OUTLOOK

The paper proposes a framework for design

execution of data-intensive business processes.

Integration of setup, operations and update sub-

processes allows to implement data intensive

business processes as one package rather than

implementing the key process and its supporting

processes separately as it is often a case. At the

current stage of the proposed research, a proof-of-

Process engine

Capture travel

receipts and

do OCR

Extract

expenses and

prefill the

report

Review the

report

Queue (RabbitMQ)

Object storage

(Swift)

Document

storage

(MongoDB)

Microservices

OCR

…

Application of Microservices for Digital Transformation of Data-Intensive Business Processes

741

concept has been developed and the further

elaboration of framework is required.

The example described is currently implemented

as a prototype and experiments are conducted with

this prototype to tune data analysis models and to

evaluate efficiency of the redesigned processes.

Performance of the technical solution is also being

evaluated with focus on scalability.

The paper has proposed the business process

architecture intended for data intensive business

processes. The further formalization of this

architecture is required. That includes elaboration of

transition from the setup sub-process to the

operations sub-process both at the business process

and infrastructure levels. From the transformation

process perspective, guidance on identification of

data intensive activities should be provided.

Additionally, redesign around data-intensive

activities often results in substantial changes in other

business process activities. These changes are

similar across business processes depending of data

analysis technique used and patterns for process

redesign could be formulated.

The prototype developed relies on the

microservice architecture and described

technological choices. These design decisions are

also subject of further investigation.

REFERENCES

Barros, O., 2007.Business process patterns and

frameworks: Reusing knowledge in process

innovation. Business Process Management Journal 13,

47-69

Becker, J., Kugeler, M., Rosemann, M. 2011. Process

Management: A Guide for the Design of Business

Process, Springer, New York

Biard, T., Mauff, A.L., Bigand, M., Bourey, J.-. 2015.

Separation of decision modeling from business

process modeling using new decision model and

notation (DMN) for automating operational decision-

making. IFIP Advances in Information and

Communication Technology 463, pp. 489-496.

Chou, D.C., Bindu Tripuramallu, H., Chou, A.Y. 2005. BI

and ERP integration, Information Management,

Computer Security, Vol. 13, No. 5, 340–349.

Damij, N., Damij, T., Grad, J., Jelenc, F. 2008. A

methodology for BP improvement and IS

development. Information and Software Technology

50, 1127-1141

De Morais, R. M., Kazan, S., de Pádua, S. I. D., Costa,

A. L. 2014, An analysis of BPM lifecycles: From

a literature review to a framework proposal, Business

Process Management Journal, vol. 20, no. 3, pp. 412-

432.

Dragoni N. et al. 2017. Microservices: Yesterday, Today,

and Tomorrow. In: Mazzara M., Meyer B. eds. Present

and Ulterior Software Engineering. Springer

Dumas M., La Rosa M., Mendling J., Reijers H.A., 2013.

Fundamentals of Business Process Management.

Springer, Berlin, Heidelberg

Lapouchnian, A., Yu, E., Sturm, A. 2015. Design

dimensions for business process architecture. Lecture

Notes in Computer Science 9381, pp. 276-284.

Lapouchnian, A., Babar, Z., Yu, E., Chan, A., Carbajales,

S., 2017. Designing process architectures for user

engagement with enterprise cognitive systems. Lecture

Notes in Business Information Processing, 305, 141-

155.

Mertens, S., Gailly, F., Poels, G. 2017. Towards a

decision-aware declarative process modeling language

for knowledge-intensive processes, Expert Systems

with Applications, vol. 87, pp. 316-334.

Moreno-Vozmediano, R., Montero, R.S., Llorente, I.M.

2013. Key challenges in cloud computing: Enabling

the future internet of services, IEEE Internet

Computing, vol. 17, no. 4, pp. 18-25.

Reijers, HA, Liman Mansar, S. 2005. Best practices in BP

redesign: An overview and qualitative evaluation of

successful redesign heuristics. Omega 33, 283-306

Roedder, N., Dauer, D., Laubis, K., Karaenke, P.,

Weinhardt, C. 2016. The digital transformation and

smart data analytics: An overview of enabling

developments and application areas, Proceedings -

2016 IEEE International Conference on Big Data, Big

Data 2016, pp. 2795.

Sidorova, A., Isik, O. 2010. BP research: A cross-

disciplinary review, Business Process Management

Journal, vol. 16, no. 4, pp. 566-597.

Van der Aalst, W.M.P., Ter Hofstede, A.H.M., Weske, M.

2003. Business process management: A survey. In:

van der Aalst W.M.P., Weske M. (eds) Business

Process Management. BPM 2003. Lecture Notes in

Computer Science, vol 2678. Springer, Berlin,

Heidelberg.

Weske, M. 2007. Business process management:

Concepts, languages, architectures in Business Process

Management: Concepts, Languages, Architectures,

Springer.

Zellner, G., 2011. A structured evaluation of Business

Process improvement approaches, Business Process

Management Journal, vol. 17, no. 2, pp. 203-237.

Zimmermann, A., Schmidt, R., Jugel, D., Möhring, M.

2016. Adaptive enterprise architecture for digital

transformation. Communications in Computer and

Information Science 567, pp. 308-319.

Zimmermann, A., Schmidt, R., Jugel, D., Möhring, M.,

2015. Evolving enterprise architectures for digital

transformations, Lecture Notes in Informatics LNI.,

Proceedings - Series of the Gesellschaft fur Informatik

GI., pp. 183.

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

742