Produced by |Tiger Sniff Think Tank.
Header |Visual China.
With the continuous advancement of large-scale model technology, the combination of AI and RPA is gradually becoming a key driving force for enterprise automation and intelligent transformation, and the application scenarios and capabilities of RPA are undergoing unprecedented changes.
The AI model gives RPA a deeper understanding and learning ability, enabling RPA to handle more complex and abstract tasks, and has stronger problem-solving and innovation capabilities.
However, the key to the implementation of new technologies is to truly solve the actual needs of enterprises and bring business value, while the combination of large models and RPAs also faces challenges such as accuracy, explainability, data quality, team capabilities, computing power costs, and data privacy and security.
However, we still have reason to believe that the combination of AI models and RPA will bring revolutionary changes to the way of working and enterprise operations in the future.
Recently, Tiger Sniff Think Tank 502 online counterparts held an online roundtable seminar with the theme of "AI Large Model + RPA: Intelligent Automation Innovation Practice in the Retail Consumer Industry", which was invited to this seminarHu Bisheng, Head of Operation Assurance Department of BESTORE, Zhou Chunzhao, Partner & Head of Product Innovation Department of Real Intelligence, Wang Yanyuan, Product Director of Yifeng Technology, and Zheng Wenliang, Technical Expert of UIPATHand other senior experts in the field of AI+RPA to share their experience and insights, together with the practice and challenges of AI large model + RPA in the retail consumption industry, and imagine the evolution and change trend of AI technology in the future.
The following is a summary of some of the views of the guests:
Application and exploration of AI large model + RPA.
Zhou Chunzhao, partner of Real Intelligence:RPA has undergone three generations of transformation and upgrading in Real Intelligence. The first generation of expert mode builds this automated process by dragging and dropping components. The second-generation simple white mode, through the mode of point selection, it is based on the internal self-developed intelligent screen semantic understanding technology, through the recommendation of actions, you do not need to enter the editor interface similar to IDE, you can complete the construction of the process. The third generation is based on a large model, which allows the agent to automatically understand, plan, and correspond to the corresponding actions through natural dialogue, and to execute them automatically.
Wang Yanyuan, Product Director of Yifeng Technology Large Model:Historically, automation has been limited by traditional rules that must explicitly tell computers how to perform tasks. However, large models demonstrate the ability to adapt, plan, and choose autonomously, which allows for a full upgrade. Therefore, we have launched a new generation of digital employee robot hyperautomation platform based on large models, based on large model technology, and implemented a number of landing cases. Our focus is on how to provide effective solutions in the field of **chain, and work with industry leaders to build industry application scenarios and digital employees.
Wenliang Zheng, Technical Specialist at UIPATH:With the rise of generative AI, we quickly embed the capabilities of generative AI into all aspects of our automation platform products, such as process building and augmenting proprietary models. At the operation layer, the results of automation are visualized, the return on investment, and the automated test suite is continuously tested to ensure the resilience of automation. Automated operations governance and flexible deployment methods to meet the growing needs of different customers. UiPath's Marketplace also provides best practice accelerators to help customers quickly realize business value. In addition, we have expanded our connections with large language model companies to improve the accuracy of model processing and accelerate model training. At the same time, we also focus on AI controllability and launch a trust layer to ensure the controllability of AI and automation.
Hu Bisheng, head of the operation guarantee department of BESTORE:We have used RPA technology in many scenarios over the past year or so. Our business covers multiple channels, including stores, e-commerce platforms, mini programs, food delivery platforms and MCN agencies, etc., resulting in a large number of orders and customer issues to deal with, although we use some automation tools, there are still many complex and difficult to automate tasks, such as order consolidation, shipping issues, review management, etc. Our role in IT is to provide technical support and training, and assist the team in setting up platforms and processes. Sometimes we act as consultants, designing business processes. Gradually, colleagues in the team learned to use RPA to solve problems. In addition to customer service, RPA is also used in the fields of finance, data integration, and monitoring. We believe that dictating or textually describing the process of generating RPAs and introducing AI capabilities to certain RPA nodes is the way forward to handle more complex tasks. At present, our RPA is mainly used for automation of specific, fixed scenarios.
Which scenarios have already been implemented?
Zhou Chunzhao, partner of Real Intelligence:We have been committed to making RPA truly usable to everyone through AI technology, and in August 23, we released the TARS (TAS) large model, which combined the TARS (TAS) large model with the original screen semantic understanding technology, and developed the real agent agent, which has also been implemented in the new retail, operators and other industries, such as in the financial link, the real agent agent can help users query bank account transaction data according to the user's one-sentence needs. The application scenarios of large models are relatively wide, but each industry has its own characteristics, and the application of any technology can only be used when it is combined with the business. For example, in a scenario of an operator, we use the real agent agent to automatically process customer service tickets, and the large model will automatically analyze and understand the task, and then plan each step of the action, and then execute it through RPA, but the operator's operating software has certain particularity, in order to solve this particularity, we spent two weeks for targeted training, and we can deliver high-performance scenario agents.
Really smart supplementary information
Wang Yanyuan, Product Director of Yifeng Technology Large Model:This year, we launched a new product of CubeAgent, which is an aggregation and training platform for digital employees driven by large model technology for the field of large ** chain (manufacturing, logistics and distribution, wholesale and retail). It can be understood as an entrance to connect the fully automated links of enterprises, which can help enterprises build their own proprietary virtual organizations built by digital employees. There are business scenarios in the organization, finance, administration, sales, each intelligent assistant has its own best practices, business communication, organization construction, new employee training, and employee self-growth can actually be completed through this platform.
Yifeng Technology shares information.
Wenliang Zheng, Technical Specialist at UIPATH:UiPath has done a lot of exploration and innovation in the field of artificial intelligence and automation, and Uipath hopes to provide an open, flexible, and responsible automation platform. We can collect context from all ** and provide it to Gen AI, and combine the capabilities of Uipath's proprietary models and large models, while intervening in the ability to collaborate with humans to help with further validation and correction. For example, in intelligent document processing, UiPath has more than 70 pre-built proprietary models for analyzing and processing documents and communications across industries and domains, embedding large models to handle more complex document types and reducing training and validation time for proprietary models, and human-machine intervention to validate data when necessary before further execution through process automation. UiPath has also recently launched Autopilot, which is automated with the help of large models.
uipath.
The key to the implementation of new technologies is demand.
Hu Bisheng, head of the operation guarantee department of BESTORE:In fact, whether it is AI or RPA, from our perspective as an application company, the core point of considering the implementation of a new technology is the demand, whether there is a scenario that can actually generate value. Large models excel at understanding, perceiving, and generating intelligence, while RPAs perform fixed tasks. Combining the two, fuzzy inputs and precise outputs can be handled, but challenges are presented with accuracy, interpretability, and data quality. Of course, team capacity, computing power cost, and data privacy and security are also key factors. As companies adopt these technologies, they need to assess whether they are really needed and address them.
Wang Yanyuan, Product Director of Yifeng Technology Large Model:I think that large models have subverted most people's perception and understanding of computer capabilities to a certain extent. Some of the challenges or bottlenecks of data security and computing power mentioned above are the areas that need to be filled by the entire ecosystem, and how to form an ecological or cooperative organization may allow us to better cope with these challenges when new technologies are emerging.
Zhou Chunzhao, partner of Real Intelligence:At present, there are three problems that need to be solved urgently, the first is consistency, the large model to handle the same task, he may plan the process and execution actions may be different each time, but the result is the same, thus causing unexplainable problems; The second is performance and speed, through the large model to achieve automation process, it will be relatively more links, more than the traditional information system more nodes, the whole perception, understanding and execution process will be longer, the response speed will be a little slower than the traditional system, now it is also necessary to continue to innovate and optimize the algorithm and technical solutions; The third is the data problem, if the agent wants to use it, it must first become a business expert in the industry, but in subdivided scenarios, specific industries, with special operations and proprietary system software, how to make the large model can quickly learn the corresponding knowledge, is also a challenge, such as establishing the corresponding data security mechanism, trust mechanism, and how to quickly privatize and personalize the training of these models.
Wenliang Zheng, Technical Specialist at UIPATH:I think the talent shortage is a very prominent problem, we especially need talents who understand business and can embrace new technologies to continue to explore and innovate, they know the pain points, but also know which technologies may bring value, such people are scarce. In addition, I think there are very big challenges in the strategic planning of the enterprise, in the organizational structure, in the proportion of resources, whether there is sufficient preparation, and whether the enterprise can create this kind of culture.
The ultimate imagination of AI large model + RPA.
Hu Bisheng, head of the operation guarantee department of BESTORE:I think RPA, combined with the end use case of AI, may eventually kill the traditional IT application system. The communication between us and machines is ultimately data, which connects people and data, and all the processes in between are not important, AI can become a real assistant to help humans solve various problems and needs.
Zhou Chunzhao, partner of Real Intelligence:We speak from the three levels of primary, intermediate, and advanced, and the primary level is that everyone has an intelligent assistant. Do anything on the computer, the assistant can help you with the task, similar to us now; Intermediate level is to change the way the computer is used, like the information system mentioned by Mr. Hu just now, it may not exist in the future, it may be a dialogue window, or a microphone, you tell him, he will help you solve it; At a high level, in fact, all our technologies will prepare for the progress and change of society and business, so in the era of future large models, all our interaction forms may have changed, and the business model has also changed? Productivity has also changed, and there may be new changes in society, but we may not be able to imagine what the specific changes will look like.
Wang Yanyuan, Product Director of Yifeng Technology Large Model:If the intelligence level of the large model can be further emerged and further reach a human-like state, the model of human-computer collaboration may indeed undergo a fundamental change. In the future organizational form, white-collar employees + al agent will become the mainstream office model, making the organization more agile, efficient and intelligent.
During the event, the online participants also actively participated in the interactive exchanges, including people from Wanda, Kid Wang, Honor, Encore Innovation, Hongta Tobacco, Midea, Shanghai Shangmei, Xiaomi and other brand enterprises, as well as practitioners from various investment institutions, industry research institutions and RPAI, and the guests also gave wonderful answers, and also successfully concluded the 502 online peer discussion activity.
Big Whale List Intelligent Automation Selection and Collection.
The latest "Big Whale List Intelligent Automation" technology vendor survey and selection launched by Tiger Sniff Think Tank has been launched, and we sincerely invite technology vendors focusing on task automation, process automation and operation automation related services to participate and accelerate the application of AI employees.
The digital transformation of enterprises is the trend of the times, new technologies are constantly developing, and new digital business scenarios are emerging, Tiger Sniff Think Tank will continue to serve the digital and sustainable development of Chinese enterprises, output timely and high-quality research and insights to the industry, and help decision-makers grasp technology trends and improve their understanding of customers, strategy and management.
To track more industrial digitalization trends, welcome to pay attention to the relevant seminars and reports of Tiger Sniff Think Tank. Go to Tiger Sniff Think Tank to complete the registration to receive our content updates and event notifications. About Tiger Sniff Think Tank.
Tiger Sniff Think Tank is a new research service platform to promote the digitalization and sustainable development of China's industries, focusing on the practical research of digital technology, selecting high-quality technical service providers for digital decision-makers, and promoting the connection and exchange of enterprises in China's digital ecosystem.
This content is the author's independent view and does not represent the position of Tiger Sniff. Do not do without permission**, please contact hezuo@huxiu for authorizationcom
People who are changing and want to change the world are all on Tiger Sniff app