At the Eighth China Fintech Forum, held in Beijing on September 9 as part of the China International Fair for Trade in Services, Wang Kaibo, business partner and head of the data science team at DeepTech Intelligent, delivered a speech on how AI is reshaping growth in the retail banking sector. His presentation focused on the theme of AI-empowered decision-making as a catalyst for new growth, emphasizing that the discussion is about much more than operational efficiency.

DeepTech Intelligent, founded in 2009, is a leading Chinese decision-AI technology company specializing in marketing and sales scenarios. In May of this year, the company listed on the Main Board of the Hong Kong Stock Exchange under the ticker 2723.HK, marking itself as the first listed entity in the enterprise decision-AI agent space. The complexity of marketing arises from the multitude of channels, direct interactions with C-end customers, and the vast array of public and private domain data that must be managed.

Since its inception, the company has focused on refining a digital operations platform with three core capabilities. The first layer is data management, addressing the need to handle extensive customer requirements and public data. The second layer involves deriving value from that data through algorithms, including both mainstream large models and a portfolio of industry-specific small models and machine learning models, such as recommendation algorithms, which have been developed over many years. The third layer consists of a matrix of product offerings, including tools for public domain advertising and private domain capabilities like CDP and MA for managing customer data.

With the advancement of large models, DeepTech Intelligent launched its enterprise-level AI agent platform, DeepAgent, in February 2025. This platform is designed to construct a series of AI agents that address single-point issues, but more importantly, to provide a complete link of capabilities that ensures the entire enterprise operation improves, not just isolated nodes. In addition to large models, the platform incorporates the small models and business knowledge accumulated within the industry. The numerous models developed from public and private domain cases are not for showcasing technology but are directly applied to business scenarios, with each model solving a specific decision-making node. While industry recognition is valuable, the most significant validation comes from the many Fortune 500 and leading enterprises that are clients of the company.

The banking industry currently faces three major pressures. First, there is significant revenue pressure due to narrowing net interest margins and related profitability challenges. Second, competition from internet platforms and other cross-industry players is diverting customer demand, making the competitive landscape cross-sectoral. Third, customer requirements have evolved, demanding not just basic deposits and savings but also fast, comprehensive services. These shifts represent the most prominent changes of the era. Additionally, several pain points are prevalent in banking operations. There is often a disconnect between systems and operations, resulting in data silos and fragmented processes. Many companies have implemented numerous systems like CDP and MA, but the data across these systems is often inconsistent in standards and definitions, preventing much of the available data from being fully converted into customer service value. There is also a fragmentation between online and offline operations, where customers may not receive a cohesive experience when switching between channels, leading to inconvenience and repeated procedures. The lack of a closed operational loop further hampers customer service, as operations remain in silos and rely heavily on manual work for data sorting, strategy formulation, and system integration, all of which impede service efficiency. AI is poised to take over these tasks while adhering to compliance and security requirements.

The path to transformation requires a systems engineering approach that begins with top-level design. The core principle is a user-centric strategy, which means being fully committed to customers. All operations should revolve around customer experience and needs. To achieve this, three key threads must be maintained: complete data integration across the entire chain, a customer lifecycle perspective, and uninterrupted business processes. With these supports, three primary scenarios are targeted: customer acquisition, marketing, and customer service, which are central to growth and user engagement. Organizationally, the head office should act as the brain, setting goals, strategies, and building platforms, while branches execute best practices. Their feedback loops back to the head office, creating a closed-loop system that allows for the accumulation of opportunities, strategies, tools, and accurate KPI tracking through continuous iteration.

The first of three breakthrough strategies is driving an efficiency revolution through refined operations, which is impossible without detailed customer insight. This requires two key elements. AI-driven decision-making at critical nodes, such as selecting target audiences and matching products for specific activities, helps free up human resources by making decisions based on data and validated models. AI also enhances insight analysis by helping business personnel understand customer needs, going beyond basic demographics to align with specific demands while ensuring data security and rule compliance. The second element is reducing repetitive work for staff, allowing them to focus on higher-value services. A case study with a city commercial bank illustrates this: by integrating multiple models for target audience, product, and scenario analysis, the bank could match appropriate content and services at the right time, addressing core operational bottlenecks. This approach yielded significant results in asset and payment transaction scenarios, with standardized data and strategies becoming valuable assets for future applications.

The second strategy is a phased approach to transformation. Taking too large a step at once can strain both organizational capacity and understanding, often yielding subpar results despite significant effort. Starting with single-point breakthroughs not only builds specific tools and capabilities but also gives the organization tangible feedback, fostering recognition and consensus. The next step is to abstract the capabilities from the first phase into a platform-wide offering, followed by the final stage of achieving highly automated execution. In one bank's three-step journey, the first phase focused on automation and integration, including the implementation of CDP and MA tools. The second phase involved digitization and intelligence, embedding AI capabilities at key nodes to bridge disconnected business segments. The third phase emphasized personalization, using a combination of large models for insights and smaller, stable models for specific decisions, resulting in significantly reduced processing times and improved decision and execution efficiency.

The third strategy revolves around integrated search and recommendation systems. As banking interactions increasingly move online, these systems are essential for better understanding customer intent. By combining contextual awareness and large model capabilities with traditional recommendation models, the system can match relevant content and services. An ontology is also embedded within this framework as it converts core business processes into accurate knowledge blocks, providing a reliable foundation for the integrated search and recommendation functionality. While only one AI agent was showcased at the forum due to time constraints, the DeepAgent platform is capable of supporting multiple agents across different business scenarios. DeepTech Intelligent is committed to advancing the integration of enterprise-level AI agents into real-world business processes and welcomes further discussions.