2.1 Robotic Process Automation

Four key improvements can be obtained from utilizing RPAs: operational efficiency, quality of service, implementation and integration, and risk management and compliance ⁽⁹⁾. Operational efficiency refers to the time and cost reductions obtained from replacing human resources with RPAs. In many industries, cost reduction is critical because it is directly reflected in company profit. Quality-of-service refers to the use of software bots to prevent user errors in a rule-based workflow. Implementation and integration improvements imply the simplification and clarification of user interfaces in large enterprise systems, including the maintenance and integration efficiencies that result. Finally, risk management and compliance involves the analysis of RPA activity logs to detect mistakes and ensure mandatory performance thresholds are met.

Huang et al. proposed an RPA application framework for an auditing project ⁽¹⁰⁾. Researchers have also automated website login, client selection, account selection, authorization request, confirmation, request monitoring, and account balance download and extraction tasks, shortening them on average from ~30 min to less than 3 min. Aguirre et al. presented a case study that used RPAs to automate business processes ⁽¹¹⁾, resulting in real service improvements, such as automating a customer relationship management receipt-generation process into AS-IS and TO-BE processes, including automatic email generation. According to the results of a one-week trial, these researchers demonstrated that a group with RPAs could handle 21% more cases than a group without RPAs.

Bhatnagar proposed new RPA roles in the pharmaceutical industry ⁽¹²⁾, including a framework for standardized ayurvedic end products and machine learning (ML) integration. The ayurvedic process automates simple and repetitive data acquisition tasks and transmits them to an ML framework for statistical analysis. Romao et al. performed a case study that applied an RPA to AS-IS and TO-BE processes in the banking industry ⁽¹³⁾. They presented and proposed a scheme to apply RPA. As such, RPA is being widely utilized in various industrial fields.

2.2 Open-Source RPA Frameworks

Although companies have worked hard to automate simple and repetitive tasks using commercial RPA frameworks, the licensing costs, security risks, and flexibility issues are driven upward due to each company needing uniquely configured solutions. Hence, open-source RPA frameworks ⁽¹⁴⁾ have been developed to reduce costs and improve flexibility, potentially increasing returns on investment. Moreover, because the implementers directly participate in the development of the source code, security and flexibility are also strengthened.

Earlier, we listed several proprietary RPA frameworks. Several major open-source solutions (e.g., Robot Framework ⁽¹⁵⁾, OpenRPA ⁽¹⁶⁾, UI.Vision ⁽¹⁷⁾, taskt ⁽¹⁸⁾, and TagUI ⁽¹⁹⁾) have been introduced, providing benefits of no- or low-code experience via user-friendly GUIs, human-readable syntax, data “scraping,” feature variety, ubiquitous computer-language integration (e.g., Python, C#, and JavaScript), and open licensing (e.g., Apache, Mozilla Public License, and Affero General Public License) ⁽²⁰⁾.

3.1 Overview

EnterRPA is based on the taskt open-source RPA framework, which requires no-or low-code experience, provides human-readable syntax, and adopts a what-you-see-is-what-you-get (WYSIWYG) user interface. WYSIWYG allows users to directly manipulate layouts without applying hypertext or layout commands ^{(21- 23)}.

We made modifications and extensions on taskt to remedy several drawbacks it has. The taskt takes and executes a script file that contains execution flows among multiple commands, but the execution of taskt slows down as the number of commands increases. Moreover, the execution even occasionally halts when commands run for longer than a week. Inspecting the source code of the taskt manually, we identified such buggy code and fixed the code to work correctly. We also improved the function that recognizes a specific screen image, so that EnterRPA can recognize even small graphical changes on screen. In addition, EnterRPA checks and reports syntactic errors in scripts before executing them, while the taskt just executes scripts and halts on errors.

Figure 1 demonstrates the structural overview of EnterRPA. EnterRPA takes a RPA script and executes it on a dedicated software robot. Users can write a RPA script directly or can also utilize the graphical user interface (GUI) of EnterRPA to create a RPA script for their simple and repetitive tasks. Then, EnterRPA assigns a software robot to the script via a job scheduling algorithm, and the script runs on the robot with automated functions defined in EnterRPA. Once EnterRPA takes a script, the script is registered as a robotic process, so that users can re-execute it anytime via the EnterRPA GUI without re-submitting the script.

We design and implement EnterRPA considering four benefits that can be obtained from utilizing RPAs:

그림. 1. EnterRPA 구조

Fig. 1. EnterRPA architecture

For the operational efficiencies, EnterRPA automates processes that are frequently and repeatedly conducted by human.

For the higher quality of service, EnterRPA prevents user mistakes by applying rule-based workflows and bots.

For the implementation, maintenance, and integration benefits, EnterRPA supports a web-based interface that provides a simple and distinct single-page method for multiple data entries.

For the risk-management, EnterRPA equips status management functions that identify the status of software robots.

In the subsequent subsections, we describe functions that EnterRPA provides (Section 3.2), a web interface for easy-to-use EnterRPA (Section 3.3), and the job scheduling (Section 3.4) in detail.

3.2 EnterRPA Functions

EnterRPA provides new functions in consideration of our company's work environment: Selenium-based browser operation automation. New C# and VBScript source-code execution functions were written, a Windows batch file was added, MySQL, Oracle, and PostgreSQL database management system linkage functions were added, and real-time debugging for task automation was incorporated.

EnterRPA supports a total of 13 functions originally supported by taskt, which are described as follows and illustrated in Figure 1 :

Condition statement to execute functions conditionally

Iteration statement to execute functions repeatedly

Web controller to control Web browsers such as Internet Explorer and Chrome

Excel controller to create, manipulate, integrate, and edit data in excel files

Word controller to create, manipulate, and extract contents in word files

Outlook controller to forward, reply, and delete emails via the Outlook mail client

Database controller to execute database connections and queries

File controller to move, copy, and load files

Folder controller to create, move, copy, and delete folders

Image controller to recognize images and optical characters

Keyboard & Mouse controller to trigger keyboard and mouse events

Window controller to resize, move, close, and perform other window operations

System controller to control execution errors, perform API operations, and etc.

EnterRPA enables users to design automated workflows by combining these functions via GUI. Thus, each user can easily generate RPA scripts for simple and repetitive tasks.

3.3 EnterRPA Web Interface

To implement an automation system for target tasks, we analyzed how the best-performing employee performs the task. The key to the employee’s ability to work well is that they realize the rules clearly about what information to enter depending on the how to operate the system and the task. We subdivided the task process to the level of input device operation, such as the keyboard and mouse. For example we could subdivide the “Log in ERP System” into four operations: 1) input the user ID with the keyboard, 2) left-click on the password input area with the mouse, 3) input the user password with the keyboard, and 4) left-click on the login button with the mouse.

그림. 2. EnterRPA Web UI를 통한 RPA 예시

Fig. 2. Example RPA via EnterRPA Web UI

Subsequently, we classified operations into two categories: information input and system control. The information input category contains operations that require user inputs to perform target tasks. Such operations need to behave differently depending on the details of the target tasks. From the example, inputting a user ID and password belongs to the information input category, since the ID and password are different for each user who wants to log in. The system control category contains the remaining operations that are independent of users’ input. From the example, two mouse click operations in 2) and 4) belong to this category, because they can behave regardless of what an ID and a password are given.

EnterRPA provides a web interface as a user input simplification system to reduce information input operation time. Operations in the information input category cannot be omitted and must be performed by users, while operations in the system control category can be fully automated. Via the web interface, EnterRPA obtains all necessary user inputs on a single web page at once and performs target tasks automatically with the given user inputs.

In addition, the interface effectively reduces users’ burdens in filling in input forms. EnterRPA requires users to put only minimum information necessary for target tasks and generates related information automatically using pre-defined rules for the tasks. EnterRPA then shows the generated information on the web interface, so that the users can review the information and can modify only some parts of the information.

Figure 2 presents an example RPA process via the EnterRPA web interface. For example, if a target task demands six user information in total, the web interface shows only three input forms for a title, contents, and attachments to obtain minimum necessary information from users. When the users fill in the three input forms, EnterRPA generates additional information such as eligible approvers, an appropriate expense account. The information is shown on the web page for the users’ review. After the review, EnterRPA sends the Job Scheduler the task with obtained information from the web interface. The Job Scheduler assigns the task to a dedicated software robot via the job scheduling algorithm, and the robot finally conducts the task automatically with the input information.

3.4 Job Scheduling

EnterRPA parallelizes multiple RPA tasks on multiple PCs and servers simultaneously. Because our base RPA framework, taskt, handles all scheduled RPA tasks sequentially on only one PC where the framework is installed, it is not suitable for enterprises that need to deal with multiple tasks from many users quickly. To overcome the limitation, EnterRPA provides three additional functionalities to distribute and handle multiple RPA tasks on multiple software robots. First, it operates a central server that manages software robots. When EnterRPA is installed on a PC, its host name and network address are automatically transmitted to the server and registered as a client in real time. Second, EnterRPA monitors and transmits script execution status, including the line of code being executed and whether the execution has been completed, to the server. If an error occurs, the error location in the script and the cause of the error are sent to the server and reported to the users. Third, EnterRPA schedules tasks across multiple software robots. It distributes tasks on dedicated software robots to maximize the use of each robot’s resources as well as deactivates or deletes scheduled tasks if needed.

Figure 3 shows the job-scheduling algorithm for allocating jobs to multiple SW robot queues. The jobs are tasks requested by users, which are mapped to RPA scripts and automatically executed on software robots. The algorithm considers two objectives. First, since the ERP system does not allow simultaneous multiple accesses from the same user ID, the algorithm prevents SW robots from logging in with an ID during executing a job with the ID. Second, it distributes jobs on software robots to minimize their idle time. The detailed description of the algorithm is as follows: The algorithm takes three maps (JobsOfId, BotForId, and JobQueue) obtained from the database. The

그림. 3. EnterRPA의 일 분배 알고리즘

Fig. 3. EnterRPA job scheduling algorithm

그림. 4. 펌뱅킹 기반 비용 지급 워크플로우

Fig. 4. Payment via firm banking workflow

JobsOfId stores relations from user IDs to lists of their newly requested jobs, the BotForId stores mappings user IDs to their dedicated robots performing the IDs’ jobs, and the JobQueue maintains mappings from robots to queues containing scheduled jobs to the robots. For each user ID of requested jobs, the algorithm tried to find a SW robot for which the ID’s jobs are currently scheduled. If such a robot exists, all the requested jobs of the ID are also scheduled for the robot to prevent different robots from accessing the ERP system with the same ID simultaneously. Otherwise, the algorithm finds the most idle SW robot with the fewest scheduled jobs, assigns the ID’s jobs to the robot to distribute jobs on all robots as equally as possible, and updates the BotForId to make the robot responsible for further requests from the same ID. After scheduling all the requested jobs, the algorithm updates the database for the modified information and terminates.

4.1 EnterRPA Application Type 1: Corporate Card Usage Settlement

After a certain period, the list of approvals made using a corporate card can be viewed by the employee via the ERP system (Figure 4). During the monthly settlement period, employees enter details about the purpose of each transaction and attach supporting evidence. For the purchase of goods, the employee must create an inspection report for each item, and for business-trip expenses, additional evidence must be attached.

4.2 EnterRPA Application Type 2: Payment via Corporate Banking

This task involves paying employee wages, contract researchers, and interns in support of research and development projects (Figure 5). On a preset date each month, the payroll staff emails an Excel file containing wage payment details to the project managers, and those managers enter the contents of the Excel file into the ERP system for payment accountability. Although one can simply upload the Excel file’s contents directly to the system, nontaxable expenses must be entered separately, and various steps for exceptions are complex. Hence, those unfamiliar with these tasks often make mistakes.