At the Eighth China Fintech Forum held in Beijing on September 9th, themed around "Technology Empowerment - Digital Innovation and Application in the Financial Industry," Bank Of Shanghai Co.,Ltd.'s Vice President and Chief Information Officer Hu Debin delivered a keynote speech. The topic of discussion was the transition from digitalization to intelligent transformation. Although there's only a subtle difference in terminology, this shift marks two distinct phases in the banking sector's technological evolution.
Why This Evolution Matters
The banking industry's digital journey spans three decades, primarily focused on migrating business operations online - moving branch services to mobile apps, tellers to remote cloud platforms, and process controls into integrated systems. The achievements have been substantial, allowing customers to handle most transactions without visiting physical branches. However, a critical examination reveals three significant costs of digitalization. First, menus replaced conversations, forcing customers to navigate through hundreds of options, which often feels overwhelming and confusing. Second, channels developed in silos, with mobile banking, internet banking, WeChat, and branch services each operating on separate logic, requiring customers to relearn processes and lacking integrated information sharing. Third, service gave way to sales pressure, as more advanced channels led to more aggressive sales pitches, making customers feel bothered rather than genuinely served.
Notably, monitoring data indicates that average daily effective usage time for mobile banking apps has dropped from 4.93 minutes in 2023 to approximately 2.70 minutes by mid-2025, a decline of about 45%. Similarly, daily usage frequency has fallen from 4.54 times to 2.86 times. These figures suggest that the initial benefits of moving services online have been exhausted. Customers don't simply need an app that processes transactions; they need a bank that truly understands them. Digitalization addressed where customers conduct business, but intelligent transformation focuses on how to understand them. Large language models have finally equipped machines with the ability to comprehend natural language and customer intent, filling the final gap in three decades of digital evolution. Intelligent transformation isn't about adding more features - it's about making intelligence a productive force and returning banking services to their fundamental purpose.
Practical Framework: The Five-Component Intelligent Teller System
What constitutes the foundation of banking services? It's the traditional counter experience - customers walk in, express their needs, and tellers understand, process transactions, and occasionally offer professional advice. Banking services have evolved through multiple channels - from counters to ATMs, telephone banking, internet banking, and mobile platforms. This fragmentation essentially represents a history of compensating for technological limitations. Because machines couldn't understand human language, customers had to learn machine language: memorizing menus, clicking buttons, and filling out forms. Now, with artificial intelligence enabling machines to understand human speech, banking service logic can be reconceptualized around the most fundamental counter model: obtaining services through conversational interaction.
Based on this insight, Bank Of Shanghai Co.,Ltd. has prioritized developing an intelligent teller system as part of its "AI Plus" strategy implementation. Importantly, this isn't a single capability but a comprehensive framework comprising five integrated components. The AI platform serves as the brain, combining large models with intelligent agents for understanding, reasoning, and generation. The data platform provides memory, housing customer profiles, behavioral insights, and industry knowledge so the intelligent teller can recognize each customer. The business platform offers capabilities, encompassing all transaction functions, product offerings, and risk control rules. Digital humans represent the interface - the interactive face between digital employees and customers, providing approachable and trustworthy presentations. Human employees handle complex judgments, emotional connections, and ultimate accountability.
Regarding the human element, when discussing AI, the common question arises about machines replacing people. In the customer service ecosystem, digital and human employees aren't competing but complementary partners. Machines provide scale; humans provide warmth. Machines deliver standard responses; humans make complex decisions. Machines understand every word; humans honor every commitment. Human employees are freed from repetitive tasks to focus on what machines cannot do - understanding customer difficulties, building genuine trust, and taking final responsibility. A key design principle is maintaining a single source of capability, where mobile banking, branch counters, cloud branches, and self-service terminals all share the same intelligent teller infrastructure. This ensures channels no longer operate independently but deliver standardized, consistent service across all touchpoints.
Implementation Case Studies
Strategy requires practical validation, and I'd like to share three case study groups. The first involves AI-powered mobile banking, transforming menu-based navigation into conversation-as-service. In 2025, we released a test version of our AI mobile banking application. In retrospect, that was essentially a prototype - customers could find functions and ask questions through conversation, but it functioned more like an intelligent customer service embedded within the app. It demonstrated that machines could understand speech, though transaction capabilities remained limited. In 2026, we're developing a new agent-driven version. The design goal isn't simply adding a chat window; it's reorganizing knowledge, functions, and services distributed across menus, pages, and processes according to customer needs. We've mapped over 40,000 professional financial knowledge points, connected more than 100 functions, and established 18 natural language processing flows. The objective is that customers don't need to understand menu structures - they express needs in natural language, and digital employees comprehend intent, orchestrate processes, and directly execute transactions. This moves from conversation-for-navigation to conversation-for-execution. Critical steps require customer confirmation, and functions beyond digital employee capabilities can connect to human staff through cloud branches.
The second case involves cloud branches and digital humans, representing the virtualization of teller services and evolution of customer-facing interfaces. Bank Of Shanghai Co.,Ltd. introduced cloud branch services in 2019, bringing human employees online and migrating services previously requiring physical branches. Features like one-touch calling and voice-command processing enable large-sum transfers, password resets, and other services, covering 90% of non-cash routine business scenarios, with cumulative service to over 1.35 million customers. Within this model, two digital employees - Hai Xiao Zhi and Hai Xiao Hui - have been deployed sequentially, handling product explanations, service guidance, and customer inquiries before switching to human employees for specific transaction processing. These digital employees are powered by financial-sector vertical large models with dedicated knowledge bases and security review mechanisms to ensure compliant and controlled responses. Remarkably, they're also capable of communicating in the Shanghai dialect.
The third case addresses intelligent risk control, moving from experience-driven to intelligence-driven approaches. Unlike the first two cases visible to customers, this operates behind the scenes. For anti-fraud, we've developed a large model system driven by both behavioral analysis and fund network analysis, breaking through traditional single-point detection limitations. The system comprehensively assesses whether customers are being induced, whether transaction motivations are abnormal, and whether fund flows are suspicious, enabling real-time early warning and interception. In the first half of 2026, this system identified and intercepted 12,000 suspicious transactions involving 380 million yuan, with false alarm rates reduced by 40% compared to traditional models. While customers may not directly perceive these capabilities, such invisible intelligence safeguards every transaction. The depth of banking intelligence isn't in front-end demonstrations but in robust back-end infrastructure.
Building Trust Beyond Technology
The "AI Plus" initiative faces its greatest challenge not in technology but in institutional mechanisms. Banking is fundamentally a trust-based industry. When customers entrust their money, they're entrusting their confidence. When we delegate services to AI, we must first answer whether AI deserves that trust. We've focused on several priorities. First, business-technology integration - bridging the gap between business and technical departments whose different languages and objectives often derail AI projects. The root cause lies in org structure, not individuals. We've implemented three measures: deploying frontline engineers embedded in business units, sharing offices, attending morning meetings together, and visiting customers to identify high-value scenarios; establishing "iron triangle" teams combining business experts, product managers, and technical managers as virtual units jointly accountable for AI application outcomes, with two-way secondment of analysts and domain experts; and creating scenario evaluation mechanisms that prioritize quick-win and strategic initiatives based on both business value and implementation complexity.
Second, human-machine collaboration - managing digital employees with the same rigor as human tellers. Digital employees represent a new workforce requiring full lifecycle management, including unified asset management to prevent redundant development, standardized positions defining responsibilities and compliance boundaries, operational loops with monitoring, feedback, and iteration mechanisms, and performance assessments with recalibration or exit provisions for underperformers. Our collaboration approach starts with human-primary, digital-support, gradually evolving toward digital employees operating autonomously within defined boundaries. Machines handle standardized work while humans focus on complex decisions and emotional communication. This boundary isn't a technical question - it's a governance question. Without clear boundaries, digital employees may overstep; with them, humans and machines can function effectively in their respective roles.
Third, security governance - maintaining inviolable bottom lines through a four-layer protection framework encompassing institutions, processes, data, and ethics. Institutionally, we benchmark regulatory requirements to establish model management policies covering the complete AI application lifecycle from initiation to retirement, with clear accountability and audit trails at every stage. Process-wise, we implement three checkpoints: admission assessment requiring risk management committee approval for scenarios involving fund transactions or credit decisions, with mandatory human review of core decisions; pre-release testing requiring multi-dimensional validation before deployment; and ongoing operational evaluation tracking output quality and drift, with automatic alerts, review, and rollback when performance degrades. For data, we enforce strict classification: personal and private information is prohibited from generative AI training, training data undergoes desensitization, AI-generated content requires prominent labeling, and testing/production environments maintain strict separation. Ethically, major AI applications undergo pre-deployment ethical assessment and regular algorithm audits.
I often tell my team that financial technology innovation requires bold experimentation but cautious implementation. Large models deserve aggressive exploration, but every customer-facing step must be deliberate. Intelligence determines how fast we can progress; trustworthiness determines how far we can go.
Future Outlook: Returning to Conversational Engagement
Reviewing three decades of banking service channels - from counters to ATMs, telephone banking, internet banking, mobile banking, and cloud branches - reveals a history of continuous channel fragmentation. Artificial intelligence now enables all channels to converge on a single service model: customers articulate needs, and intelligent tellers understand and serve. History has followed a spiral path - starting from conversational engagement, traversing extensively, and returning to conversation with all accumulated capabilities intact. This isn't a return to the past. Where tellers once served dozens of customers daily, digital employees can simultaneously understand thousands of needs. Where tellers memorized hundreds of business rules, digital employees carry comprehensive institutional knowledge in their systems. The conversational form has returned, but insight, scale, and consistency have all been elevated.
These changes transcend technology, fundamentally altering banking's underlying production relationships. Previously, systems development followed prescribed processes with sequential coding - systems were process rigidification. Requirements flowed from business departments through approval, development, testing, and deployment cycles measured in months or quarters. In this model, technology played executor to business's requester, akin to contractor-client relationships. The intelligent era fundamentally changes this logic: capabilities precede scenarios, and systems gain learning and evolutionary properties. Once large models are deployed, capability boundaries aren't defined during development but expand through customer interactions. This necessitates technology and business units transforming from contractor-client relationships into partners - jointly defining problems, validating outcomes, and accepting results in shared trenches. Without organizational change, AI projects inevitably fail - a conclusion validated across multiple scenarios in our experience.
Talent evaluation standards must also evolve. In the digital era, technology talent was assessed on translating requirements into code and meeting delivery deadlines. In the intelligent era, core competencies shift to problem definition, model calibration, and judging AI output boundaries. Developers who lack business understanding diminish in value, while business professionals who can't engage with models similarly decline. The truly scarce talent combines business fluency with model literacy - what we internally call composite professionals. Bank Of Shanghai Co.,Ltd. is restructuring technology talent evaluation systems and promotion pathways, incorporating model optimization effectiveness and business scenario implementation outcomes as core assessment metrics rather than solely code volume and project delivery counts.
The transformation in risk control logic is more fundamental. In the digital era, risks were known and rule-based - violations and anomalies were predefined, and systems enforced rules mechanically. In the intelligent era, risks are unknown and emergent: models may hallucinate, data distributions may shift, and training-time compliance doesn't guarantee inference-time compliance. This shifts risk control from primarily preemptive design toward concurrent monitoring with rapid rollback, and from rule-driven approaches to model-driven strategies with human safeguards. We need operational systems capable of real-time model behavior sensing, automatic alert triggering, and rapid issue isolation. Consequently, intelligent transformation isn't merely technological upgrading - it's a systematic restructuring of banking organizational forms, talent structures, and governance logic. Technology can be acquired, but production relationships cannot be purchased; they must be transformed from within.
Over three decades of digitalization, we taught customers to use machines. The path of intelligent transformation is about machines learning to understand customers. Bank Of Shanghai Co.,Ltd. stands ready to walk this path steadily alongside industry peers and partners. Thank you.
Comments