"By inputting a few personalized details, the test system can automatically generate a suspicious transaction analysis report. Once 'AutoReport' is launched, it will liberate us from the tedious task of report writing, allowing us to focus more on business development," said an excited anti-money laundering specialist involved in testing "AutoReport."
"AutoReport" is based on CIB’s independently developed intelligent suspicious transaction report generation model (AML-GPT). Utilizing large-scale models and natural language processing technology, it intelligently analyzes the characteristics of suspicious customer behavior, suspicious entities, and suspicious transaction information to generate auxiliary analysis reports for suspicious transactions. By employing intelligent Q&A technology, personnel can quickly adjust and optimize the initial drafts generated by the model, marking a breakthrough for CIB’s GPT-like large model from 0 to 1. It also won the first prize in the comprehensive track of the Innovative Project Award at the 2023 Annual In-House "Blazing Innovation" Marathon.
In December 2022, following the release of ChatGPT based on the GPT-3.5 model, a wave of generative AI technology swept across the globe. How to leverage generative AI technology to empower business development has become a new challenge for financial institutions.
Currently, financial large model applications are emerging in scenarios such as customer marketing, operations assistants, and intelligent customer service. However, using large models and natural language processing technology to empower compliance and risk control in financial institutions is still rare. CIB has carved out a unique path, focusing on the niche area of anti-money laundering, and has introduced the "CIB Solution" by independently developing the AML-GPT model.
According to reports, this model is the first in the financial industry to use a large model in the field of intelligent generation of suspicious transaction reports. AML-GPT focuses on the specific field of anti-money laundering, offering high levels of specialization and targeted services. It not only reduces the daily workload of grassroots anti-money laundering personnel but also improves the quality and efficiency of case identification. Under the guidance of the People’s Bank of China’s Fujian Provincial Branch, the technical solution of "AutoReport" has been included in "China Anti-Money Laundering Practices."
"Without intelligent auxiliary tools, we need to process approximately 30 suspicious transaction reports daily. Each report requires conducting investigations, analyzing transaction flows and behavior characteristics, and writing the report, taking between 20 to 60 minutes per report," said a grassroots anti-money laundering employee at a branch.
However, "AutoReport" leverages advanced artificial intelligence technology to train an intelligent model for generating anti-money laundering suspicious case reports. Utilizing the system's multi-dimensional features, it conducts comprehensive analysis and evaluation to form suspicious analysis reports that align with anti-money laundering expert logic and provides preliminary handling recommendations. Additionally, embedding the large language model into the anti-money laundering system, based on knowledge retrieval tools to expand the knowledge base, allows the large model to fully exercise its generative capabilities, continuously revising and improving the content of suspicious analysis reports. This effectively equips each anti-money laundering worker with a personalized digital assistant.
In the past two years, CIB’s anti-money laundering staff have written over one million suspicious transaction reports. "By applying the large model for intelligent report generation, grassroots employees can be freed from the tedious manual reporting process, allowing more focus on business development, researching new money laundering methods, and optimizing expert rules," said the project leader of "AutoReport."
It is reported that CIB plans to upgrade AML-GPT to a model with hundreds of billions of parameters and advanced adjustable algorithms, combined with machine learning, knowledge graphs, natural language processing, and text mining technologies to further enhance the overall performance of the model.
In the digital age, CIB is comprehensively accelerating its digital transformation, creating a new blueprint for digital CIB, with more and more innovative projects like "AutoReport" blooming on the CIB stage. Currently, CIB has around 8,000 tech talents and an annual tech investment exceeding 8 billion RMB. It has initially formed a "1 (Mobile Banking) + 5 (CIB Inclusive Finance, CIB Butler, CIB Life, Money Manager, CIB Platform for Integrated Banking) + N (scenarios)" digital CIB system, with invention patent applications increasing 50-fold over the past three years.