Banks handle thousands of loan approval tasks daily. How can they conduct sampling reviews more efficiently with limited headcount at headquarters? Artificial intelligence (AI) large language models are transforming banking workflows: by leveraging these models, banks can comprehensively inspect all loans, shifting from random checks to full coverage. This approach has enabled many banks to identify numerous compliance violations and reduce losses.

At an insurance company, a new agent repeatedly practices with an AI "training partner" on their phone, preparing for a meeting with a business owner client the next day. Questions like, "What products should I recommend?" or "What tricky questions might the client ask?" are rehearsed. Quickly, this agent evolved from a novice struggling to close deals into a top performer successfully securing policies.

On March 17, 2026, at the release of the report "AI Driving Renewed Upgrades in Mainland China and Hong Kong's Financial Services Industry," Wang Jianping, Management Consulting Partner at PwC China, shared impressive AI application cases. From October 2025 to January 2026, PwC surveyed 201 financial services professionals from the banking, insurance, and asset management sectors in mainland China and Hong Kong, supplemented by 20 in-depth interviews, painting a picture of the industry's embrace of AI: high enthusiasm, but distinct challenges.

**Strategic Importance, Insufficient Investment**

As a capital-rich sector, what is the current state of AI application in finance? According to PwC's research, financial institutions' use of AI can be categorized into three tiers. The first tier involves internal use, such as searching internal knowledge bases, which is relatively mature in leading banks but less visible to external clients. The second tier includes client-facing applications like intelligent customer service, investment advice, and post-trade services. In banking, however, applications like anti-money laundering detection and internal control audits have shown significant benefits. The third tier involves directly facilitating client transactions or providing investment and wealth management advice. This level requires balancing AI transparency, fairness, and traceability of logic, making large-scale implementation challenging in the short term.

Ni Qing, Mainland China Asset and Wealth Management Industry Leader at PwC, observed that different sectors prioritize AI deployment differently. Banking focuses on risk control, anti-money laundering, and compliance; insurance emphasizes agent capability enhancement, customer service, and claims processing; while asset and wealth management prioritizes investment and portfolio management, alongside data and market analysis.

"Surveyed institutions have already achieved initial returns of 10-15% on AI investments, but they place greater value on AI's long-term potential to enhance market position and strategic development space," Wang Jianping stated. However, a significant gap exists between ideal and reality: the survey revealed that 61% of financial institutions allocate less than 10% of their technology budget to AI.

Why not increase investment? Wang Jianping analyzed that in the current market environment, boosting overall technology spending is difficult for financial institutions. They must adjust their structure—reducing traditional technology investments and lowering costs associated with conventional business development and testing—to reallocate savings into AI. With a fixed total budget, significantly increasing the proportion allocated to technology and AI would heavily impact traditional tech. Furthermore, financial institutions prioritize stability, leading to a slower transformation pace compared to other industries. Meanwhile, AI technology evolves rapidly, making it challenging for finance to keep up, resulting in a situation where AI is "strategically important, investment should increase, but significant short-term boosts are difficult."

**Talent and Data Dilemmas**

"Internal training takes too long, while external recruitment faces intense competition from tech companies offering high salaries," commented a human resources professional from a major state-owned bank, highlighting the recruitment challenges for AI-related talent.

The survey found that only 29% of financial institutions have successfully built an "AI-first" culture. Talent shortages and rigid organizational structures are more significant obstacles than insufficient investment or technical issues.

Li Weibin, Management Consulting Partner at PwC China, pointed out: "Respondents widely reported a major challenge is recruiting versatile talent who understand both business operations and algorithms. Training and upskilling existing employees, along with establishing incentive mechanisms that encourage using AI as a transformation tool, are crucial for fostering an AI-first culture. Some financial institutions are exploring special mechanisms to break traditional hiring models. For instance, certain insurers have established AI research institutes with unique compensation structures to attract senior talent; some banks are driving the transformation of traditional tech staff through training and assessments to identify talent. Equally important, senior management must lead by example and actively advocate for AI application."

Beyond talent and organizational culture, data is a key constraint. The survey identified the top three barriers to increasing AI investment as data availability (30%), regulatory pressure (20%), and the need to prioritize maintaining existing core systems (14%). Data security and privacy protection were listed as the primary challenges in data management, leading 90% of financial institutions to rely on internal proprietary data to support their AI applications.

From Wang Jianping's perspective, the data required for AI is shifting from structured to unstructured formats. However, unstructured data in financial institutions—such as banks' credit policies, risk control manuals, and review knowledge, or insurers' survey manuals—has not traditionally been included in data governance systems.

The "hallucination" problem inherent in large models makes it difficult to meet the data accuracy requirements of traditional business. "Take auto insurance claims as an example. When a client reports a claim, the backend might identify relationships between the claiming vehicle, the collided vehicle, the surveyor, and the repair shop, potentially indicating fraud. If the model only focuses on the claim data itself and ignores these relationships, it may fail to detect fraud risks. Therefore, ontological modeling is necessary to find connections between various entities and inject these relationships into the large model to improve identification accuracy," Wang explained.

With the advancement of AI technology, financial institutions also face new challenges in risk control and security. The report found that while basic protections are in place at the input and output stages, effective automated monitoring tools are still lacking for addressing dynamic risks during model operation.