FSIs Are Banking on AI for Productivity Gain

2025 is challenging for the banking sector. The tariff wars, stock market roller coaster, and macroeconomic fluctuation put many FSIs under tremendous pressure on profitability. In pursuing healthier financial performance, FSIs are addressing these external risks by turning to AI to transform internal operations and drive productivity. 

This finding comes from IBM’s 2025 Global Outlook for Banking and Financial Markets and is also supported by market researchers. Bloomberg Intelligence and Deloitte research indicated that AI adoption could lift pretax earnings by up to 17%, or USD 1.5 billion for local banks. With the expectation of AI boosting productivity and profitability, Hong Kong technology and banking leaders recently convened at the  FSI & AI Hong Kong Summit to discuss ways to transform AI potential into profitability.

Boosting productivity and profitability

“AI is the answer to automation for higher productivity and increased operation efficiency,” said Kayton Wan, software leader at IBM Hong Kong. He added that banks are eager to harness automation advantages to boost productivity, and 60% of the banking chief executive officers stated in the IBM survey that they are willing to accept the significant risk associated with these initiatives.

“Banks have always wanted to maintain a competitive edge, and they’re always looking for ways to maintain operational efficiency,” added Altaf Rehmani, executive of a local bank and author of the book “Generative AI for Everyone”. “AI is simply helping us do things we have defined.”

Rehmani said that AI is a tool, but it can be transformative when used right. This explains why most FSIs are experimenting with AI. According to the Hong Kong Monetary Authority’s 2024 survey, 75% of local banks reported adopting or planning to adopt AI for daily operations. This number also aligns with IBM’s finding, where 78% of the banks had a tactical approach to AI.

Meanwhile, IBM’s study also stated that only 8% of banks were systematically developing generative AI in 2024, indicating that enterprise-wide AI adoption and an AI-first approach remain limited.

The tipping point to become AI-first

Wan said successful AI adoption is easier said than done, quoting Gartner’s report, which stated that as many as 54% of AI pilot projects are expected to fail to move into production. But he added banks are reaching a tipping point to move from being “+AI,” meaning businesses are simply adding AI to their existing process into “AI+.”

“The tipping point comes when businesses can combine AI with IT Automation inside the enterprise,” Wan said. “When AI really gets woven into the fabric of the company, this is AI+.”

To pivot from “+AI” into “AI+,” organizations need a good technology foundation and adoption framework, according to Peter Lee, chief technology officer and pre-sales engineering leader at IBM Hong Kong. The rapid development in AI and continuous breakthroughs mean new buzzwords are in the market every quarter. “Earlier this year, DeepSeek was one of the hottest topics; today, everyone talks about agentic AI,” he said.

AI adoption framework provides a holistic roadmap for organizations to stay focused on their adoption journeys and be open to emerging trends. “You need to have a framework in place; otherwise, you cannot plan for future growth, and the latest trends will pull you everywhere,” he said.

Building blocks for future-proof AI

As a banking executive and AI expert, Rehmani agreed with the value of a framework. He added that the framework should focus on major elements like data, APIs, and governance to enable efficient and responsible AI initiatives.

“Without these foundations, all you can do are PoCs, and you can’t be ready for the real world for AI to operate securely and at scale,” he said. “Most financial service organizations are still laying the foundation but are far from the future state they envision.”

To help businesses become AI-first, IBM offers watsonx, a portfolio of AI products designed around five major AI building blocks: data, models, governance, assistants, and agents. Wan added that these building blocks work hand-in-hand to build use cases from the pilot to the “AI+” implementation stages.

While data and models are two foundational building blocks, Lee said the focus is shifting as technology matures. It is given that organizations have a strategy to enable data access for analysis, and open-sourced models are no longer the conversation. More importantly, Lee said organizations should consider the ownership of data and models.

“Your data is your data; your model is your model,” he said. “You need to train your model for your benefit, not other company’s benefit.”

Although open-source LLMs are easily accessible, he urged organizations to train their models using their data for accuracy, governance, and compliance. He explained that LLMs are developed using public Internet data, which includes inappropriate wording, biased data, and unethical information.

The cost and risk associated with unethical AI systems are steep. He added the European Union AI Act could fine organizations as much as 7% of their global income.

As regulations worldwide around AI evolve with increased fines, organizations should select and train models that best fit their operations. Thus, Lee suggested organizations mitigate their risks by training their models using enterprise data within a controlled and governed environment to future-proof their AI systems.

“AI governance is becoming a mandatory requirement,” said Lee. “IT helps organizations to guardrail the AI outputs that comply with the regulations.”

In addition to different technical considerations, Rehmani said non-technical factors also power organizations to become AI-first.

“Leadership commitment from the top is absolutely essential,” he said. Since most organizations are still in the phase of experimentation with AI, failures are unavoidable. The ability for organizations to pivot from these failures to become AI-first successes lies in the leadership’s commitment and AI governance.

Image credit: iStockphoto/uzenzen