Banking’s AI transformation starts with the customer

Jason Cao, CEO
Global Digital Finance, Huawei

People often assume that the changes reshaping the banking industry are being led by AI. But while technology plays an important role, the deeper force behind this transformation is simpler and more human: changing customer behavior.

We saw this pattern during the rise of digital banking in the mid-1990s. Customers embraced online services and, over time, migrated to mobile payments and banking apps. As their behavior changed, financial institutions had to change with it.

The same dynamic is now playing out with AI. Today, more businesses expect payments to settle in real time and want financing to keep pace with their operations. More consumers are becoming accustomed to services that happen instantly and often disappear into the background. Before long, both groups will expect banks to understand their needs and respond more proactively, with appropriate consent and controls.

These changes will fundamentally alter the relationship banks have with customers. Increasingly, customers will use their own AI assistants to interact with a bank’s AI agents through secure, authorized channels. For consumers, agents may help monitor bank balances, compare financial products, support portfolio decisions, and identify avoidable fees. For businesses, agents will assist with daily tasks such as processing invoices and sending payment reminders, while also helping with more complex work such as cash-flow management, foreign-exchange decisions and compliance checks. Bank agents will work alongside human employees to anticipate customers’ evolving needs, while human judgment and accountability remain central to important decisions.

This is the emerging model of Agentic Banking, in which customer AI, bank AI, and human expertise work together within clear governance boundaries. Customer behavior is changing. The question is whether banks can respond quickly enough, and at sufficient scale, to remain relevant and trusted.

More than just an add-on

Piloting a conversational AI assistant can be relatively straightforward. Re-engineering a bank’s technology, processes, and culture so that AI can work reliably at scale is much harder.

AI alone does not create banking value. It can answer questions, summarize documents, or make recommendations, but those capabilities become useful only when AI can connect securely with governed data, institutional knowledge, business processes, and production systems. This is often where the limitations of legacy technology become visible.

Many financial institutions still rely on applications and platforms built decades ago. Designed for a different era, they can be hard to change, difficult to connect to newer applications, and poorly suited to services that need to respond in real time. These systems were often designed for stability and longevity—strengths that remain essential—but tightly coupled architectures can make rapid change and integration difficult. Modernization must therefore be carefully engineered to reduce risk while preserving business continuity.

As a first step toward that change, financial institutions need to rethink the architecture beneath their AI. For many, that means modernizing tightly coupled applications where necessary and building more flexible environments in which applications, governed data and AI services can work together securely.

The objective is not to move everything to the public cloud. Banks need a hybrid architecture that allows models, data, and workloads to run where regulatory, security, resilience, performance, and cost requirements can best be met—across on-premises and cloud environments—while connecting AI securely to the parts of the business that matter.

As banks move from experimentation to production, four interconnected challenges become decisive: sovereignty, security, scalability, and sustainability. Sovereignty means retaining control over data, domain knowledge, models, and deployment choices. Security means protecting sensitive data and financial transactions while ensuring that AI systems remain governed and auditable, with clear human accountability. Scalability is not simply about supporting more users or replicating isolated use cases. It means embedding AI at scale across a broader range of critical business operations and using it to re-engineer end-to-end processes across the institution. Sustainability means ensuring that compute, subscription, and operating costs remain manageable, and that the investment case remains viable, as adoption grows. These four challenges reinforce one another; addressing any one of them in isolation is not enough.

Where engineering makes the difference

While modern infrastructure lays the foundation for change, engineering determines what a bank can actually do with it. Powerful AI models are becoming increasingly accessible, making access alone less of a differentiator. What matters is what a bank can reliably build, deploy, and operate with the models it has.

A general-purpose AI model may understand language. A bank understands banking. It knows its customers, its risk profile, its regulatory obligations, and the way its business works. Engineering allows the bank to turn that knowledge into operational and customer value, transforming a general model into something built to accomplish a specific financial task.

One approach is to start with an open-source foundation model and adapt or ground it using the bank’s governed data, knowledge, and expertise. The result can be a domain-specific AI system designed for a particular job, such as detecting fraud or assessing credit risk.

But building the model is only part of the work. It must be tested, connected securely to existing applications, and monitored throughout its lifecycle. Banks also need to trace the information on which AI outputs are based, understand and audit how AI contributes to important decisions, and preserve appropriate human accountability. When an application moves from a limited deployment to one serving millions of customers, the requirements for security, resilience, auditability, reliability, and performance become much more stringent, while controlling costs becomes paramount.

Meeting the 4S challenges requires systems engineering and governance across models, data, infrastructure and business processes. A better model, by itself, will not solve them.

Learning from China’s banks

Many Chinese banks have spent years rebuilding the digital foundations on which AI now rests. Consequently, they offer a useful reference point for financial institutions in other countries.

Some of China’s largest banks have modernized core systems serving hundreds of millions of customers. They have also combined open-source foundation models with governed banking data, knowledge, and expertise to build domain-specific AI systems for tasks such as risk analysis, fraud detection, and credit assessment.

No two banking markets are identical, and approaches that work in China cannot simply be transplanted elsewhere. But China’s experience shows what is possible when modernization is pursued systematically and at scale. For banks elsewhere, the value lies less in copying a particular architecture than in learning from approaches that have already been tested in large and complex financial environments.

Huawei has worked with Chinese financial institutions through many of these modernization programs. That experience has given us a close-up view of what happens after a promising AI use case moves beyond the laboratory to become a reliable banking service.

The challenge is not to export a single architecture, but to share proven engineering practices and adapt them to the needs of each market. That requires working with local banks, software providers, and systems integrators that understand the market and can co-create solutions that meet local business, regulatory, and operational requirements.

As ever, customer behavior will set the pace of change. Banks that succeed in this new environment will not necessarily be those with the largest models or the most AI projects. They will be the institutions that retain control over their data and models, connect AI with institutional knowledge, modernize the underlying systems, and embed AI securely and sustainably across critical business operations and end-to-end processes to deliver services customers trust.