Source | Laser Finance (leishecaijing)
Driven by the dual model of general large model + industrial large model, industrial digitalization is moving towards high-level intelligence, opening up imagination space for a qualitative leap in enterprise operating efficiency and productivity liberation. As a financial industry with deep penetration of digital technology, it is also expected to find new cost reduction solutions in the wave of large models.
From the perspective of the current financial industry, especially retail finance, the cost pressure above the compliance bottom line has increased significantly. High-frequency and concurrent personalized demands continue to raise service response and experience standards. Adjustments to customer group structure and business strategies require more efficient risk control models, while refined operations redefine human efficiency.
These cost challenges faced by the retail finance industry cannot be solved by previous financial technology methods. In the ten years since digital technology has penetrated into the financial field, financial technology has built basic databases, algorithm libraries and computing power frameworks for the financial industry, completing the initial stage of digital finance.
When the demand for digitalization on the financial business end shifts from question and answer to interaction, from online to intelligence, and from copying to reasoning, financial technology will iterate to higher-order technological forms in the evolution of digital technology. Based on the current development direction of AI, large models are likely to be one of the representatives of advanced forms of financial technology.
Just as the war between hundreds of models in the Internet industry is in full swing, first movers in the financial industry are also beginning to gain momentum and develop large industry models in financial segmentation scenarios. On August 28, Immediate Consumption released the country’s first retail financial model, “Tianjing”, at the Financial Model Development Forum, marking the first step in the high-quality development of the financial industry driven by large models.
Although large models are still in the exploratory stage in the financial field and there are still many application problems, judging from the implementation effects of first movers, the trend significance of large models for digital financial transformation is highlighted.
The imagination of financial big models
In the wave of development of the digital economy, the penetration of digital technology into the financial field continues to deepen. Fintech has iterated several times to provide the financial industry with digital solutions such as intelligent customer service, intelligent risk control, and digital middle offices, becoming the key underlying support for the modern financial system.
As digital technology continues to evolve and financial technology gradually enters a high-level intelligent form, how much imagination can emerging large-model technology bring to the financial industry? In other words, what kind of productivity changes can large-scale financial models bring to the financial industry?
Judging from the iterative demands of financial institutions for financial technology at the current stage, the imagination of large-scale financial models mainly lies in reducing costs and increasing efficiency.
According to the "2023 China Fintech Enterprises Chief Insights Report" jointly released by the China Internet Finance Association and KPMG, the confidence index for the future development of the domestic Fintech industry in 2023 will be the highest in the past three years. As competition in the industry intensifies, Fintech companies will become more It has become an industry consensus to attach great importance to enhancing technological competitiveness and reducing costs and increasing efficiency through technology.
In the path of cost reduction through technology, the prospects of large models and AIGC financial applications have become the focus of the industry. More than 90% of the companies surveyed are optimistic about the prospects of AIGC's financial applications, more than 20% of companies already have relevant technologies and product layouts, and more than 70% of companies believe that AIGC can optimize business innovation and content production, and is expected to be deeply integrated into the daily operations of financial institutions.
It can be seen that the application of AI large models has become a new opportunity for financial technology companies. Relying on large model technology to reduce operating costs and increase efficiency, win more market space, and constitute the ultimate imagination of financial large models.
As far as the current operating status of the financial industry is concerned, cost pressure has increased significantly. These cost pressures are the micro-foundation of the imagination of large financial models. First of all, the user side faces an entire user experience chain including reach, interaction, conversion, service, and customer complaints. The needs of different users in the retail financial scenario are diverse and complex, posing challenges for financial institutions to create the ultimate service experience.
Secondly, the risk of fraud and credit risk on the risk control end is high, and comprehensive credit data processing is difficult. Financial institutions urgently need more efficient intelligent risk control model components to effectively reduce risk control costs, improve risk control efficiency, and increase risk assessment and risk prediction. accuracy.
Finally, the operation side involves many departments, large business volume, long processes, and large manpower requirements. Digital technology is needed to reengineer the operation process to achieve the transformation of an agile organization, maximize productivity, and improve human efficiency.
In the actual survey of financial institutions, the demand for large models by financial institutions is mainly reflected in the above three levels. The first is to optimize the user experience and improve customer experience and satisfaction through more natural, smoother and smarter large model applications; the second is to assist in risk control and assist financial institutions in analyzing and processing data; the third is to automatically summarize and retrieve information Generate reports to strengthen the digital capabilities of all employees.
The imagination of large financial models is released, accelerating the implementation of large models in the financial field, and opening up space for a new round of digital financial transformation.
Answers to the Sky Mirror Large Model
Following the evolution trajectory of financial technology and based on the demand for large models, commercial banks, consumer finance companies, financial technology companies and other participants have begun to lay out large financial models in order to build a new growth engine for the financial industry.
Among them, the "Tianjing", the country's first large-scale retail financial model launched by Immediate Consumption, was launched first, taking the lead in kicking off the large-scale model's drive to promote productivity changes in the financial industry. According to Jiang Ning, chief information officer of MaMaConsuming, large models have broad application prospects in the financial field. Promoting the construction of personalized service experiences for users will effectively improve the efficiency of value chains such as marketing and operations in the financial field, and further expand the role of data decision-making in risk control. The innovative application effects in the field have helped the digital transformation of the financial industry to make a substantial leap.
Public information shows that the current Sky Mirror model covers the four core capabilities of gathering wisdom, awakening knowledge, creating value among people, and digital clones.
In terms of gathering wisdom, it is mainly used in artificial customer service scenarios. Through large models, the customer service experience of front-line excellent artificial agents is extracted and aggregated into group wisdom, thus having the ability to serve customers one-to-many. It can also be used as an auxiliary role for artificial agents to help recommend , optimize answers; awakening dormant knowledge mainly solves the pain points of extracting and utilizing data in unstructured documents; crowd-creating data value is mainly to lower the threshold for using data; in terms of digital people, it aims to create "digital appearance + smart brain + emotion" "Inner" three-in-one digital person, a smart secretary who is good at understanding, has warmth, and understands psychology, or a smart "worker" who never sleeps.
Different from the general large model, the Tianjing large model is positioned as a large model of the financial industry. It is based on real financial business scenarios and solves the cost reduction pain points of the industry based on financial business needs. It is understood that Ma Ma Consumption has previously deployed large model technology in intelligent marketing, intelligent customer service and other aspects, and has achieved good results in actual operations.
Data shows that after nearly three months of operation, the Tianjing large model’s intention understanding accuracy reached 91%, higher than 68% of traditional AI; the customer participation rate was 61%, higher than 43% of traditional models, and higher than manual agents. average level of 28%.
In addition, by uploading data and customizing some parameters for the digital clone of the Tianjing large model, employees can have their own digital clone with just 5 minutes of data training to complete a large amount of work on their behalf.
Lu Quan, director of the Immediate Consumer Artificial Intelligence Research Institute, said that from the very beginning of the research and development of Tianjing Large Model, it has focused on helping financial companies implement it and generate actual value.
Different from other industries, the financial industry pays more attention to technological innovation and professional compliance, and the fault tolerance rate is extremely low. Therefore, the base of large financial models must have strong security and stability. In order to ensure the safety and controllability of the Sky Mirror large model, Immediate Consumption has established the underlying capabilities of the "three vertical and three horizontal" large models.
The three verticals refer to the comprehensive capabilities of real-time human-computer collaboration, multi-modal intelligence, and data decision-making intelligence to achieve intelligence in the data field and realize structured data discriminative models. The three horizontals refer to continuous learning, model compliance, and combined AI to form safe, compliant, and trustworthy technical capabilities to ensure that the model becomes smarter as it is used, and at the same time is more stable, safer, and controllable.
Jiang Ning said, "We hope that the answers it gives to customers will be compliant under any circumstances, and that its results will be stable under any unpredictable circumstances."
On the basis of security and compliance, how to make large models better understand finance and more effectively meet the needs of digital transformation in financial scenarios? There are two answers, one is an in-depth understanding of financial business scenarios, and the other is data accumulation and technology accumulation in financial scenarios.
Behind the Tianjing model, Immediate Consumption, as a leading licensed consumer financial institution, has accumulated 179 million users in eight years of development, has more than 2,000 models, 100,000+ variables, and nearly 50PB of multi-modal, high-quality data. By fine-tuning and aligning training on these data, and using inference acceleration technology to achieve model controllability, you can immediately become a large model participant who understands both finance and technology.
At present, financial large models still face key tasks and dynamic adaptability, personalized requirements and privacy protection, group intelligence, security and trustworthiness, infrastructure and other challenges. However, the advent of the Tianjing large model will allow the financial industry to see and Work together to solve these problems and inject new digital momentum into the high-quality development of the industry.