Copilot: The Key to Hyperautomation

NOV 25, 2024 Eunju Hong

- This article is based on the keynote and session details presented at REAL SUMMIT 2023, hosted by Samsung SDS in September 2023. –

Since 2019, Samsung SDS has been offering Robotic Process Automation (RPA) solutions, supporting business process automation for companies. We currently hold the largest market share in South Korea, serving numerous clients. Traditional RPA solutions are primarily focused on automating simple, repetitive tasks and UI or document processing, which posed limitations in advancing towards hyperautomation. However, these limitations have been overcome with the advent of generative AI, triggered by the release of ChatGPT. Generative AI excels in understanding, interpreting, and expressing the complexities of human language, which has long been a challenge in office automation. This breakthrough has enabled the automation of intellectual tasks. In other words, with the emergence of Large Language Models (LLMs), a technological foundation has been established that can accelerate automation. This not only expands the scope of tasks that can be automated but also significantly improves the quality of work and dramatically enhances productivity, paving the way for a new level of automation—hyperautomation—that is fundamentally different from traditional automation.

Companies typically have eight mega processes: development, procurement, manufacturing, logistics, marketing, sales, service, and management support. Members of an organization carry out research, creation, and development tasks across all areas of these processes, applying the analysis results and interpretation to solve problems. To support this, Samsung SDS is preparing to seamlessly connect all areas of company processes using generative AI, thereby enhancing productivity through hyperautomation. This includes automating works using generative AI across various domains, such as the "common work systems" for internal collaboration, "core work systems" like ERP, CRM, SCM, HCM, and PLM for company operations, and "system development" and "system operation" areas tailored to the company's specialized needs.

Hyperautomation innovation in companies using Generative AI (Source: Samsung SDS)

Hyperautomation in Common Work Areas

What should be considered when using generative AI in corporate tasks? Regardless of the task at hand, the processes of preparation, communication, and creation are repeated countless times. Projects and issues often arise simultaneously, requiring constant communication and collaboration with relevant departments, partners, and clients, necessitating email writing, meetings, and document creation/management. The system that supports such collaborative work is known as the common work system. In common work, writing (creating) documents involves repeated information analysis, review, and insight generation processes. Rather than relying on general external information or arbitrary creation, generating outputs using recent emails, documents, accumulated business knowledge within the company, and data from various work systems is essential. The key consideration to effectively use generative AI lies in "how to leverage the complex relationships, systems, and diverse information within a company."

In this context, it is necessary to check four perspectives: data (intellectual assets and corporate data), tools (work/collaboration tools), output (quality of deliverables), and security (prevention of information leaks). First, the system should be capable of using recent business information and external data effectively, and tools such as meetings, messaging platforms, and document creation applications must be easy and convenient to use, fitting the specific work situation. As the quality of the output is also crucial, it is necessary to generate results through optimal queries based on the company's data and work tools. Lastly, since the leakage of company secrets can pose significant management risks, the system must be capable of preventing sensitive information from being leaked and managing access rights appropriately.

In other words, for a company to effectively use generative AI, it is crucial to have concrete strategies for three key points: △ integration of in-house data, △ contextual and seamless usability, and △ generative AI security management tailored to the company.

Key points for using generative AI in enterprises (Source: Samsung SDS)

Brity Copilot: Generative AI-Based Collaboration Solution

Brity Works is a collaboration solution (common work solution) provided not only to Samsung employees but also to many corporate clients. It is a suite of tools optimized for corporate communication, offering services like email, messaging, meetings, and a drive service for managing personal and shared documents and an approval platform integrated with company work systems. As a front-end solution, it serves as the closest interface for using generative AI. Samsung SDS offers Brity Copilot, a service that combines the key points mentioned earlier for using generative AI with Brity Works, supporting hyperautomation in common company works. What sets Brity Copilot apart regarding key points for using generative AI?

1) Integration Structure for Using Internal Data

To perform tasks, you need both external information and internal data, such as recent emails, documents, and the latest updates from work systems, and you must address the question, "Which LLM should we choose and how do we access internal data?" to ensure safe and appropriate use. Brity Copilot addresses this by offering a Multi-LLM option, which allows you to integrate multiple LLMs. With the recent release of high-performance, open-source LLMs like Meta's LLaMA that can be optimized for specific purposes, it will become common to customize and use them as Private LLMs. Brity Copilot allows companies to use multiple LLMs simultaneously, using a private LLM for internal HR and management information, and connecting to external public LLMs for market trends and academic materials, all tailored to the company's needs and policies. Brity Copilot supports fine-tuned private LLMs based on the company's optimized data, while also offering the flexibility to use only public LLMs when private LLMs are not required.

Brity Copilot - Multi LLM Option Support (Source: Samsung SDS)

Brity Copilot also provides a connector (plug-in)-integrated structure that enables the use of emails, documents, and work system information from Brity Works. Since company information is constantly being updated, it is challenging to train LLM with all data continuously, so there needs to be a method to search for and use the information as needed. Brity Copilot addresses this by offering a "knowledge search," which indexes and preprocesses (vectorizes) unstructured, lengthy knowledge assets, such as the latest emails, messenger conversations, stored documents, and meeting minutes saved in Brity Works, and provides a basic structure for searching and using this in conjunction with LLM. Also, structured data from internal work systems like ERP and CRM can also be integrated through OpenAPI.

Brity Copilot – Internal data plug-in integration (Source: Samsung SDS)

2) Contextual & Seamless Usability

Brity Copilot functions as both a personal assistant and a work facilitator, providing an easy, intuitive user experience. It provides an interface that offers an easy, intuitive user experience. It allows users to access features directly from their work environment (system) and use functions appropriate to the communication context. Brity Copilot can summarize emails or draft approval documents and search for and share information instantly during a messenger conversation. During meetings, it offers real-time Korean subtitles for better communication and provides summaries of previous meetings, helping users to maintain their workflow. Brity Copilot can also automatically generate meeting minutes based on a recorded transcript and quickly share action items with relevant stakeholders. Moreover, Brity Copilot plans to offer an "MS Office Add-in" feature optimized for document creation. Adding Add-in to MS Word, Excel, and PowerPoint allows users to search, summarize, and draft emails, schedules, tasks, and documents directly within these programs. When writing documents, it is crucial to apply the company's unique format and style, aligning with the background and purpose based on recent documents, discussions, and current situations, and the editing process should be easily manageable.

Brity Copilot offers prompt UI features, such as templates, conversational queries, chip menus, and follow-up question suggestions, to facilitate quick access to necessary resources and data and easy formulation of optimal queries. For Excel tasks, Copilot can quickly summarize and analyze data into summary tables, helping users derive insights that typically require significant time and effort. When creating planning documents using Word, you can quickly find and use the necessary information with the following step-by-step prompts provided through a chip menu*—for example, you can freely query by selecting private LLM to search and summarize internal documents, but selecting the "Data Search" chip menu offers options to choose search targets (mail drive, file drive, etc.)—which makes it quick and easy to generate content efficiently. * Chip Menu: A UI element consisting of brief text or images representing user input options is used for tasks such as selection or filtering. In other words, it is used to display the value chosen by the user or to provide options the user can select.

Brity Copilot – Office Add-in functionality (example of chip menu) (Source: Samsung SDS)

Employees often switch between numerous tools, such as email, meeting tools, and document creation tools, to complete their tasks. Brity Copilot provides a seamless user experience that maintains workflow continuity and enhances productivity. For instance, you can start a meeting directly from a messenger conversation and share automatically generated meeting minutes via email through Copilot. After receiving an email, you can search for necessary files in the drive and use the "Office Add-in" feature to work on documents. Once the work is complete, the document can be immediately shared with colleagues via email or messenger, making a smooth connection from the beginning to the end of the work.

3) Security Management of Generative AI for Enterprise

Compliance concerns, such as security breaches and privacy leaks, are major company challenges. While generative AI is anticipated to enhance business competitiveness, it becomes useless if security risks are not adequately addressed. In fact, this is one of the biggest reasons why many domestic and international companies hesitate to adopt generative AI. So, how can these security issues be resolved?

Using a private cloud rather than a public cloud to configure their work systems or Private LLMs for security-sensitive companies is crucial. Of course, using a private cloud is optional, and multiple clouds can be used. However, does using a private cloud guarantee perfect security? Not entirely. External access is often necessary for performing tasks within the company, which means the risk of information leakage always exists. Therefore, a robust filtering mechanism is essential to manage this risk effectively. While using a public LLM to search for market trends or publicly available information poses no issues, querying with confidential company information carries significant risks. Brity Copilot offers a keyword filtering feature. You can set message patterns for keywords, source code, personal identification numbers, etc., that should not be leaked according to the company's policy. The system analyzes the message before sending prompt content to generative AI and automatically blocks its transmission. In reality, it is not easy for all companies to set up private LLMs, so most businesses will likely use services based on external public LLMs. In this context, keyword filtering becomes an essential feature that enables companies to use generative AI within their policies and guidelines.

Also, detailed data access management is necessary since the data that different departments and individuals need to access varies. As more companies integrate various internal work systems via plug-ins with generative AI, the need for precise control over data access will grow. While there are common intellectual property and policy guidelines that everyone can access, certain information, such as HR or financial data, is restricted to specific departments. Even within the same role, like sales, there might be confidential information accessible only to certain individuals. Brity Works is a company-optimized collaboration solution that offers granular permission options based on company, department, individual, job level, usage location, and device, and the Brity Copilot service provides the same level of access control. For example, if a user requests system data through Brity Copilot, and they do not have the necessary access rights, the system will notify them that the data is inaccessible, and the user can then request access through the appropriate approval process, and the system administrator can grant access to the allowed data.

Brity Copilot – Security options for using generative AI within enterprises (Source: Samsung SDS)

Hyperautomation in Core Business Areas

Large-scale enterprise systems, such as ERP, SCM, HCM, PLM, and CRM, are necessary for business operations, and global solution providers are developing Copilots with generative AI for these systems. However, to use these features effectively, it's crucial to connect them seamlessly with a company's legacy systems and have a deep understanding of processes specific to the Korean market. Based on its extensive experience and success in B2B IT innovation, Samsung SDS supports hyperautomation in core business areas by offering services that combine the best solutions.

Among these core systems, SCM is one of the most challenging areas to automate. This is because businesses need to make decisions not based on transaction data but by predicting future data to plan and determine where and how much resources should be allocated. However, through generative AI, even these aspects can be automated. Until now, SCM planners have conducted scenario planning based on their knowledge and experience, which was an asset that couldn't be reused from a company's perspective. However, if the actions taken by planners during the scenario planning process, such as comparing data between versions and adjusting plans, are all recorded and accumulated in the system and used as training data, similar issues can be automatically resolved when they arise.

For example, by using the SCM Copilot that we are preparing in partnership with o9 Solutions, a strategic partner, the process of forecasting demand and increasing production capacity accordingly can be automated. To explain a detailed use case, user A is planning an aggressive promotion for Black Friday in the North American market in November and calls the SCM Copilot to modify the existing scenario plan. They then re-run the plan with a scenario that increases demand by 20% and request a comparison with the previous base version. The Copilot summarizes the comparison results between the re-executed plan and the previous version based on user A's request and provides a link to view the detailed information. User A checks the detailed information page in the SCM system through the link and identifies a discrepancy between the demand forecast and supply availability. They then request the Copilot to send a message to B, the production support manager, informing them of the predicted supply shortage situation and asking them to adjust the production plan. Upon receiving the request, the production support manager B reviews the message and attached files on the collaboration screen and adjusts the production capacity plan accordingly. User A requests the Copilot to compare the scenario plan with the modified version to review the changes. They can see that the November demand and supply forecast values have increased according to the scenario with a 20% increase in demand. By using generative AI in this way, it will be possible to automate planning, analysis, and problem-solving across not only SCM but all areas of a company's eight major processes. In this context, let's take a closer look at the automation of the product development process using generative AI that Samsung SDS is considering.

Generative AI-based Product Development, PLM Copilot

Product development is evolving into an R&D management system based on data generated throughout the entire product life cycle, known as Product Data Management (PDM). In the 90s, PDM focused on expanding product development information across the organization, focusing on collaboration within the design department to improve time-to-market. Later, through Collaborative Product Commerce (CPC), the focus shifted to open collaboration and the creation and sharing of information between companies. In the 2000s, product development evolved towards integrating product information and company-wide operational data to enable collaboration across the entire company value chain from a digital enterprise perspective. Since 2010, it has been advancing towards integrating and using product information on a cloud basis for digital transformation. So, how will the product development system change when generative AI is applied?

Product development trends (Source: Samsung SDS)

Product development data should encompass all accumulated data, including know-how and problem-solving processes, allowing easy access to similar required information for issue resolution, and the stored information should be valuable with a logical precedence. Product development has a very long life cycle from development to discontinuation. Information for cars is managed for at least 10 years, while data for warships or military supplies needs to be managed on a 50- to 100-year cycle. Also, since personnel or companies may change, having information with a logical precedence is essential.

Examining the pain points in using such data, lead researchers who plan concepts struggle to identify necessary information or intermediate outputs, while design engineers often stick to existing methods due to the overwhelming number of factors to consider during design. In 50-80% of overall tasks, professional engineering tools such as CAD are used. However, information is often accumulated within these tools without being reflected in related systems when changes are made. Validation engineers require more test cases as products become more complex. Still, due to a lack of time and manpower, they are often unable to generate sufficient test cases for thorough validation.

However, in the product planning stage, applying generative AI can support searches linked with market, technology, and internal system information, enabling fact-based content review and generating ideas for concept planning. In the product development (R&D) phase, it generates design drafts in consideration of various aspects. It can also generate software code, experimental plans, and test cases and support design and software code reviews.

Product development system using PLM Copilot (Source: Samsung SDS)

Design Draft Creation Process

Let's go through a scenario for creating a design draft by assuming a persona. An automotive parts design engineer, Kim, has received an urgent request to design a part for a strategic model targeting the Indian market. However, he lacks the time to create a design draft and is unsure how to proceed. To ideate the design, he needs to review base models and examples, analyze gaps, and establish a plan, all of which require considerable effort. The problem is that it's not only difficult to verify this information, but also challenging to know where to find the necessary materials, and there are many factors to consider in the design process, but the work often proceeds in the usual way due to time constraints.

If Copilot is applied, it will detect the design request and provide notification with a brief message. Kim then asks Copilot, "Compare the key specs and load differences between the base model and the modified version for the India-targeted strategic product in detail. Also, check the internal knowledge hub for design cases where only the load was modified." The generative AI creates a summarized analysis report that compares the specifications of the models and provides the location of similar cases in the internal knowledge hub using Retrieval-Augmented Generation (RAG) to verify the accuracy of the information. Based on the analysis report, Kim requests the generative AI by including constraints and parameters to adjust: "Generate 10 lightweight design options by referring to the truss support reinforcement case, which is an improvement design method for model S with load-bearing constraints, and adjust the support thickness, position, and quantity." The generative AI generates design drafts and explanatory content in conjunction with Samsung SDS's ongoing research on generative design in product development. Kim can then review and finalize the design options.

Scenario for design draft creation using generative AI + Gen. Design (Source: Samsung SDS)

It is essential not to forget the preliminary tasks to drive innovation in the product development system through introducing generative AI. First, it is important to △ continuously accumulate and collect product development data that has logical connections and learning value. Although it might seem like there's a lot of data in the product development field, only CAD files and analysis results are available in many cases. The most important data includes meeting content and troubleshooting issues, which should be consistently collected to ensure they have learning value and then used to identify potential use cases. Without securing use cases that can enhance productivity, generative AI is meaningless. Also, since sending internal company information outside poses risks, △ adopting a scalable AI platform for enterprises that can facilitate learning from internal data and orchestrate various external LLM models is necessary.

Samsung SDS is preparing a "product development offering" based on a scalable generative AI service platform. The generative AI service platform consists of three main components: the data module, which handles data collection, preprocessing, and the connection of structured and vectorized data with LLMs; the training module, which supports LLM training, training data management, and fine-tuning; and the service module, which supports model orchestration, filtering, and knowledge retrieval. In other words, the product development offering is structured to connect key apps for product development, such as design/development, product management, software management, and digital twins, with the generative AI platform, focused around the common work environment of Brity Works. This supports productivity improvement through the automation of the product development process.

Product development service offering with generative AI (Source: Samsung SDS)

Epilogue

Factories have been automated through the application of rules, but office work automation has remained incomplete due to the difficulty in applying such rules. With the advent of generative AI, it is now possible to automate not only simple repetitive tasks but also intellectual tasks. This advancement enables hyperautomation across all process areas within an organization, dramatically increasing productivity. The key to hyperautomation based on generative AI is not "autopilot" but "Copilot." Humans use their insight to oversee the entire workflow, give instructions to the Copilot, and then make final decisions based on the results provided. In other words, it is a concept of streamlining work through an assistant called Copilot.

Samsung SDS's Brity Copilot integrates generative AI into common work systems (email, messenger, meetings, and data storage) to enable seamless collaboration and integration with existing legacy systems, while supporting a private cloud environment for secure usage without security concerns. In the core work systems (ERP, SCM, HCM, PLM, CRM, etc.), Samsung SDS is preparing Copilot in collaboration with global solution companies. For example, applying SCM Copilot can automate the adjustment of production plans according to changes in demand forecasts. In the system development and operation areas, a pilot application showed that development speed improved by 30%, validation became twice as fast, and 60% of customer requests in ERP operations could be handled automatically.

To apply generative AI, companies should first define the areas to automate from an overall business perspective and then design the detailed tasks to be carried out. Ultimately, it is a task that requires collaboration between personnel with domain knowledge and technical expertise, and finding solutions in partnership with IT service companies like Samsung SDS that have extensive B2B business experience is crucial. Samsung SDS will guide and implement generative AI in complex corporate tasks, making it easy and convenient to use, and will accompany companies on their journey toward a hyperautomated enterprise.

▶ This content is protected by the Copyright Act and is owned by the author or creator.
▶ Secondary processing and commercial use of the content without the author/creator's permission is prohibited.

Eunju Hong

Strategic Marketing Team at Samsung SDS

Had been in charge of digital transformation in samsungsds.com, solution page planning/operation, based on her work experiences in IT trend analysis, process innovation, and consulting business strategy, and is now in charge of content planning and marketing through trend/solution analysis for each main business sector of SDS.

App	설계/개발	제품관리	SW 관리	디지털 트윈
협업환경	단일 업무 환경(Brity Wroks) - 대시보드/위젯	단일 업무 환경(Brity Wroks) - 메신저/화상미팅	단일 업무 환경(Brity Wroks) - 작업공간(문서관리)	단일 업무 환경(Brity Wroks) - 3D 협업
AI 플랫폼	생성형 AI 서비스 플랫폼 Data 모듈 수집/전처리 정형 데이터 Vectorization	생성형 AI 서비스 플랫폼 학습모듈 LLM 모델 서빙 LLMOps 지원 모델 저장소	생성형 AI 서비스 플랫폼 학습 모듈 학습 Data 관리 Fine Tuning 사전 학습	생성형 AI 서비스 플랫폼 서비스 모듈 Orchestrator Filtering 지식 검색

Copilot: The Key to Hyperautomation

Hyperautomation in Common Work Areas

Hyperautomation in Core Business Areas

PDM

CPC

PLM

PLM

Next

Engineer

AI Copilot

생성형 AI 서비스 플랫폼

Data 모듈

생성형 AI 서비스 플랫폼

학습모듈

생성형 AI 서비스 플랫폼

학습 모듈

생성형 AI 서비스 플랫폼

서비스 모듈

Epilogue