In recent times, the financial big model has been reshaping every key link in the financial service chain. The industry believes that the application of large models can help financial institutions significantly reduce costs and increase efficiency, bringing important value such as new opportunities, new growth points, and model innovation.
Can the implementation of large-scale financial scenarios walk the "last mile" well? How does the big model delve into financial business scenarios, and what are the requirements for financial institutions for its rapid iteration? How does generative technology integrate into consumer finance? Recently, Lu Yong, Chief Technology Officer of Lexin, gave an exclusive interview to Xinhua Finance and Economics reporters.
Walk the Last Mileage of Financial Scenarios in Big Models
For a period of time, many institutions have increased the research and promotion of generative technology. LexinGPT by Lexin, Xuan Yuan by Du Xiaoman, Tianjing by Jima Consumer Finance, Wushi by Lujinsuo, and Qifu GPT by Qifu Technology are all booming.
It seems that all kinds of financial scenarios are exploring adapting to large model interfaces overnight. However, it is worth noting that the implementation of financial big models in the industry is a complex system engineering, and it is not easy to truly achieve practical implementation, continuous operation, and unleash the potential of big models. Therefore, whether the implementation of large model financial scenarios can take the "last mile" well has become a focus of attention in the industry.
Lu Yong introduced that at present, there are three ways to open big models in the industry. One starts with pre training models and has a lot of data, such as Pangu and ERNIE Bot. Such companies need to invest billions or even billions of yuan in funding on a large scale within a few years.
Lu Yong said that currently Lexin is more like a "medium model", with a database size ranging from tens of billions to hundreds of billions. It mainly focuses on fine-tuning financial vertical fields with data sizes ranging from tens of billions to hundreds of billions based on the general pre trained large models of its partners, which can meet the needs of business scenarios and better meet practical needs.
"The third type is small companies that make customized applications by purchasing large model APIs," said Lu Yong.
Lu Yong emphasized that Lexin focuses more on the integration of major models and business scenarios, rather than blindly investing at no cost. LexinGPT is a vertical large model application that serves the business through pre training with Lexin Financial's exclusive data and fine-tuning of business data, based on the general large model.
In Lu Yong's view, only technologies that are implemented in the actual business and penetrate into the capillaries of the business can truly assist the business.
"For the consumer finance industry, the most mature and widely used big models are still in customer service, telemarketing, and collection." Lu Yong gave an example, saying that JD Finance and Huawei Pangu models are excellent representatives in the industry. Applying big models to risk control models is the most profitable for the industry, and Lexin continues to maintain research and actively explore in these areas.
Lu Yong introduced that LexinGPT is a self-developed large model platform by Lexin. Lexin has accelerated the implementation and application of AI big models in the company through financial exclusive data pre training and business data fine-tuning. In terms of business interaction, it has been fully implemented in major business processes such as telemarketing, customer service, and collection; In terms of productivity improvement, it is widely applied in scenarios such as R&D code assistance, design creativity generation, data analysis, RPA for operation and office process automation, to enhance the overall operational efficiency of the company; In terms of risk control core, Lexin continues to pay attention to and actively explore the application of AI big models in the core field of risk control. These measures significantly improve the overall operational efficiency and customer experience of the company.
Lu Yong added that Lexin will also apply the big model to financial market analysis, and the security department of Lexin will use the big model for financial market analysis, including anti fraud, to respond to the marketing water army's "wool picking" behavior.
Generative AI technology helps improve the quality and efficiency of consumer finance
Lu Yong believes that the big model will reshape various industries and bring new productivity, and the consumer finance industry is no exception. Moreover, large model applications are better suited for cognitive work with people, which will help consumer finance institutions significantly improve their efficiency.
Large models can play an important role in building underlying capabilities such as data extraction and model construction, assisting in intelligent decision-making for business and enterprise operations.
According to Lu Yong, as a leading company in the segmented industry, Lexin currently has nearly 200 million high-quality and high growth young users. Its vast user behavior profile data and real transaction link data are no different from a huge data asset. "It has important value in predicting user behavior preferences," Lu Yong said.
But how to sort out data relationships and clarify data bloodlines is like a census of a foreign city, vast and time-consuming. At this point, the advantages of the large model are evident: through deep learning algorithms, the large model can perform data cleaning, classification, and computation more efficiently. It is reported that with the collaboration of technologies such as LexinGPT, the Lexin technology team has worked for more than two years to gradually clarify the massive data relationships and form numerous accurate models for predicting user behavior preferences in different scenarios, including loan willingness models, marketing bias models, offer satisfaction models, repayment willingness models, and customer churn warning models.
According to Lu Yong, based on a massive and accurate prediction model, Lexin has developed the "Turing Decision Simulation System" - which can directly simulate various real business scenarios, equivalent to a "simulation test" of enterprise management. It can quickly output the performance of various key business indicators in just a few seconds, and the accuracy of the simulation results exceeds 95%.
Applying new technologies to achieve refined operations is the key to high-quality development of consumer finance
Lu Yong pointed out that currently, the fintech industry has entered the second half, and refined operation is the key to high-quality development. In the future, Lexin will mainly achieve refined operations by applying AI and big models to the entire financial technology chain.
Lu Yong stated that after the implementation of Lexin Financial's vertical language model, the proportion and efficiency of robots participating in customer service have steadily increased, and the robot resolution rate without human intervention has reached 91.5%; In addition, the Lexin big model has further been implemented in fields such as data analysis, data warehouse design, and risk control data warehouse upgrading and optimization, greatly reducing the participation threshold for data analysis and improving data analysis efficiency.
Lu Yong also added that the application of the big model can further improve marketing efficiency. After applying the big model, the ordering rate of low active users increases by about 15%, GMV increases by 80%, and GMV of high active users increases by 18%.
When it comes to cost control, Lu Yong said that Lexin has always adhered to a technology driven approach since its establishment, and has invested a total of 2.6 billion yuan in the past five years since its listing. In the third quarter of 2023, Lexin's R&D investment reached 127 million yuan, continuing to maintain industry leadership. Lu Yong introduced that Lexin has recently landed more AI large-scale model scenarios. Especially focusing on the essence of financial business, from data lineage sorting, model building, to systematic tool construction, we strive to create a leading full chain quantitative management system to improve operational efficiency and customer experience.
Lu Yong also introduced that Lexin fully leverages the advantages of AI technology and launches a "5S protection system" for consumer protection, comprehensively strengthening consumer rights protection in data security, anti fraud protection, standardized services, intelligent customer service, and cracking down on financial black industries. Utilize technology to comprehensively protect consumer rights and make financial services more warm.
When discussing future plans, Lu Yong stated that the large model has been implemented in Lexin and is accelerating its application in Lexin's business processes. In the future, Lexin will continue to promote in-depth exploration of AI big models in areas such as risk management and anti fraud.