The integration of artificial intelligence (AI) techniques such as machine learning (ML) and natural language processing (NLP) with robotic process automation (RPA) can enable businesses to efficiently automate higher-order tasks. Businesses can also improve their productivity and reduce operational costs using AI in RPA. This article discusses the importance of AI in RPA, an RPA + AI implementation approach, applications in internal audit and invoice processing, and recent developments in this field.
Image credit: Fit Ztudio/Shutterstock
Importance of AI in RPA
RPA refers to software that can automate computer programs to perform tasks complying with a set of rules consistent with the business process. Public security, government, logistics, insurance, new retail, manufacturing, banking, taxation, finance, and e-commerce are the major application areas of RPA.
AI techniques, such as deep learning (DL), ML, NLP, and optical character recognition (OCR), can enable RPA to have robust autonomous learning and cognitive capabilities and correct its behavior continuously through big data. Thus, RPA with intelligent operations research and decision-making capabilities can be realized by integrating AI.
AI-powered RPA bots/RPA bots combined with AI technologies, such as OCR, DL, and ML, can gather data automatically from different sources, eliminating the need for manual entry and substantially reducing errors in the data acquisition process compared to conventional data acquisition methods.
AI-powered RPA bots can streamline information processing by automating complex workflows and executing tasks faster than humans to reduce processing times. AI can also optimize workflow steps and perform data analysis for better decision-making.
An RPA + AI Implementation Approach
Initially, a user interface captures the user requests, and then the purpose of the user is understood by classifying natural language understanding (NLU) pipelines into predefined categories. These NLU pipelines consist of several NLP libraries such as SpaCy or NLTK to offer multi-language support for word vectors/embedding tokenization.
This layer extracts user requests for the entity type, a crucial function as entities are part of a text of interest to the company/data scientist, such as people’s locations, account numbers, addresses, and names. Advanced algorithms and techniques, such as conditional random fields (CRF) and stemming, are employed to train the Named Entity Recognition (NER) models.
NER support is available in several libraries such as SpaCy and allows the extraction of entities by identifying their statistical properties such as patterns. Additionally, primary information, such as date and time, can be obtained using libraries such as duckling. The extracted intent and the objects of the data pass through the machine-to-machine interface.
A software robot can be plugged in at this stage to drive automation and orchestration. In the design process, details such as leveraging technology, managing exceptions, tracking and monitoring, performance and quality, access specifications, data structure, and protocols must be discussed and evaluated. The control logic can be decoupled and externalized to realize versatile design.
Applications of RPA
AI and RPA in Internal Audit: Internal audit initially requires the identification of problems through sampling tests. However, several problems are missed in a large data volume environment, necessitating real-time monitoring of full data.
Additionally, the process and risks are understood through interviews, which are limited by the subjective wishes and experience of the interviewee, increasing the difficulty of obtaining a comprehensive and objective understanding.
Thus, audit activities require significant costs and are extremely time-consuming as they depend highly on experts. Intelligent automation can automatically execute regular audit procedures and release resources, unlike conventional internal auditing, which is resource-dependent and primarily focuses on post-event inspection.
Intelligent internal audit can be implemented by quickly developing an intelligent audit monitoring platform using RPA and AI technologies. The robot develops an analysis model, identifies process abnormalities and system and behavioral abnormalities to perform cause analysis, and then risk warning and abnormal handling.
AI and RPA-based intelligent monitoring platforms can ensure 24/7 automatic monitoring of full data, rapid adaptation to business rules and process changes, automatic abnormal cause analysis, and timely detection of new fraud techniques, violations, and system abnormalities caused by rapid iteration.
AI and RPA in Invoice Processing: The processing of documents with RPA was performed using OCR, which has several limitations while documenting formats or detecting scanned text, such as issues in detecting blurred text and interpreting natural language.
RPA and AI techniques can be combined to perform intelligent document processing (IDP) to overcome these issues. IDP utilizes ML for the clear and better processing of documents. The data can be structured in a viewable manner and digitized using IDP in three steps.
Initially, manual actions can be replaced by software robots, followed by proper structuring of all data, and lastly, digitizing the data and enabling an automated flow of structured data between third parties, employees, and suppliers, who can utilize this data for analytics and decision making.
Recent Developments
Several routine business tasks and processes are performed by qualified experts, which results in a loss of productivity. RPA can be used to perform such repetitive tasks in an automated manner with higher accuracy and consistency to substantially reduce the workload of experts and enable them to focus on other tasks with more relevance.
In a paper published in the 2020 15th Iberian Conference on Information Systems and Technologies (CISTI), the authors proposed an RPA application that can dynamically detect objects in software applications interface in real-time to ensure accuracy, performance, and flexibility irrespective of the OS and interface tool location.
A convolution neural network (CNN) was trained with several menus and interfaces and utilized to classify software interfaces in real time. Moreover, software was developed that takes automated actions, such as clicking, editing text, and moving the mouse pointer.
Results demonstrated that the proposed technique based on DL can effectively detect objects in real time, classify the detected objects with exceptional accuracy, and take dynamic actions to efficiently automate all computer tasks.
RPA has quickly evolved from automating simple rule-based tasks to mimicking more sophisticated human tasks, which require the integration of AI. However, the integration of AI and RPA can lead to several challenges as it involves professionals from both AI and RPA fields who have different backgrounds and skills and AI models often degrade over time, which adversely impacts the performance of the overall solution.
In a paper published at the International Conference on Business Process Management, the authors described the AIRPA project that addresses these challenges by proposing a software architecture that enables the abstraction of the robot development from the AI development and monitoring, controlling, and maintaining intelligent RPA developments to ensure its performance and quality over time.
AIRPA was developed with a microservice architecture and service contract standardization. This architecture is composed of four modules, including the document repository, the deployment manager, the AIRPA Control Room, and the tracking and exploitation panel. The major goals of the AIRPA project were based on Servinform industrial experience and the background within the IWT2 group RPA research line.
These goals include creating an AI component collection to empower RPA solutions, creating a nexus of union between both domains by presenting the results in an understandable manner in a platform for business experts and technical staff, ensuring automation of end-to-end processes that enable the integration between AI components and existing RPA solutions to reduce the need for human participation and decision-making, and reducing and simplifying cost to access AI-powered RPA solutions with licensing restrictions.
The goals also include enabling RPA professionals lacking AI and ML skills to use AI components, defining a development methodology, production, lifecycle, and integration roadmap of RPA solutions with AI components, verifying the developed AIRPA framework in several realistic scenarios, and integrating an AI components library in RPA solutions.
The AIRPA project was performed in the Servinform context, and the prototype was validated with reality settings. Specifically, the average handle time (AHT) was reduced in processes AIRPA deployed cognitive robots, with the initial AHT of 9 min being reduced by 75% to a final AHT of 2 min and 15 s after AIRPA implementation, which indicated the potential of the intelligent RPA development. However, the final results can vary as the AIRPA project is still in the validation phase.
References and Further Reading
Martínez-Rojas, A., Sánchez-Oliva, J., López-Carnicer, J. M., Jiménez-Ramírez, A. (2021). Airpa: An architecture to support the execution and maintenance of AI-powered RPA robots. International Conference on Business Process Management, 38-48. https://www.researchgate.net/publication/354056493_AIRPA_An_Architecture_to_Support_the_Execution_and_Maintenance_of_AI-Powered_RPA_Robots
Martins, P., Sá, F., Morgado, F., Cunha, C. (2020). Using machine learning for cognitive Robotic Process Automation (RPA). 2020 15th Iberian Conference on Information Systems and Technologies (CISTI), 1-6. https://doi.org/10.23919/CISTI49556.2020.9140440
Jha, N., Prashar, D., Nagpal, A. (2021). Combining artificial intelligence with robotic process automation—an intelligent automation approach. Deep Learning and Big Data for Intelligent Transportation: Enabling Technologies and Future Trends, 245-264. https://doi.org/10.1007/978-3-030-65661-4_12
Zhang, X., Wen, Z. (2021). Thoughts on the development of artificial intelligence combined with RPA. Journal of Physics: Conference Series. https://doi.org/10.1088/1742-6596/1883/1/012151
Desai, D., Jain, A., Naik, D., Panchal, N., Sawant, D. (2021). Invoice processing using RPA & AI. Proceedings of the International Conference on Smart Data Intelligence (ICSMDI 2021). https://dx.doi.org/10.2139/ssrn.3852575