The Second Half of AI in the Insurance Industry: Focus on “Certainty,” Not Just “Large Models”
Hello everyone, I’m your financial observer.
Recently, both the tech and finance communities have been discussing a crucial question: Can AI truly help insurance companies make money, save costs, and better serve their customers?
At the Bund Conference on September 11th, industry leaders such as Ant Group, Taibao Property Insurance, and Alibaba Cloud came together and reached a very clear and pragmatic conclusion: The competition in AI for the insurance industry has moved beyond the initial stage of “who has the biggest model” and has entered a deeper phase of “who can implement solutions most effectively.”
In simple terms, before, the focus was on “whose AI is smarter or can mimic human conversations better.” Now, the focus has shifted to “whose AI is more reliable and can actually get the job done without making mistakes.”
Below, I’ll break down the key points from this high-level forum into four easy-to-understand aspects, showing you what the “second half” of AI in the insurance industry is all about.
---
1. Stop with the “Universal Chatbot” – First, Figure Out What Pain Points You Want to Solve
Many tech companies create AI products with a large chatbot that claims to be able to answer any question. However, this “all-in-one chatbot” approach doesn’t work in the insurance industry.
Chen Guanhua, General Manager of Ant Insurance, made a vivid analogy: Installing a motor on a manual razor doesn’t make it an electric razor. You first need to determine what kind of electric razor the user wants (rotating or reciprocating? Waterproof or portable?), and then decide on the size of the motor and the battery to use.
What does this mean for the insurance industry?
- Product definition first: When developing AI for insurance, you shouldn’t ask “How big is my model?” Instead, you should ask, “Which specific problem of the user do I want to solve?”
- Do you want to help users quickly calculate premiums?
- Or track the status of claims?
- Or recommend the right insurance policies based on their family situation?
- Customization for different scenarios: Different scenarios require different combinations of capabilities. For example, recommending insurance to the elderly requires knowing their health, age, and existing policies, while recommending it to young people might focus on different factors.
- Diversity rather than dominance: Future AI products for insurance should be developed by each company based on their understanding of their customers, resulting in unique services, rather than all companies using the same generic AI chatbot.
In plain language: Don’t expect one AI to do it all. A good insurance AI is a “specialist,” not a “generalist.” It needs to know the specific problem you want to solve before finding the right technical solution.
---
2. The Gap Between AI’s “Probability” and Finance’s “Certainty”
This is the core and most challenging logical conflict in the news.
What is the essence of AI? It’s large models that predict the probability of the next word based on massive data. AI is good at guessing, estimating, and possibilities.
What is the essence of financial services? Certainty. Premiums must be exact, claims must be processed according to regulations, and risk control must be flawless. Financial services cannot tolerate approximations or uncertainties.
This creates a huge contradiction: How can you use AI, which is good at guessing, to handle financial tasks that require precision?
Zhang Chi, Vice President of Alibaba Cloud, pointed out that although many financial institutions have released AI agents, their functions are often limited. When these agents are used in real business processes, problems arise:
- Inaccurate calculations: The premiums calculated by AI don’t match the company’s policies.
- Black-box processes: It’s unclear how AI makes its decisions, and there’s no traceable reasoning.
- No way to revert: If the model is upgraded, how can previous errors be corrected or audited?
In plain language: AI is like a talented intern who’s creative but often makes basic mistakes without explaining why. Financial services, on the other hand, need a “rigorous accountant” where every step is traceable and accountable. Overcoming this gap requires engineering solutions, not just relying on AI to figure it out on its own.
---
3. Regulatory Measures: The “Document No. 8” is a Barrier, Not a Restriction
On June 18, 2026, the China Banking and Insurance Regulatory Commission issued the “Guidelines on the Safe Development and Application of Artificial Intelligence in the Banking and Insurance Industries” (Document No. 8). Some might think regulatory documents are just there to restrict companies.
However, Zhang Chi sees it as an open challenge, requiring financial institutions to answer four critical questions when deploying AI:
- Is the conclusion based on evidence? (No random assumptions.)
- Can the process be audited? (No black boxes.)
- Can mistakes be corrected? (No irreversible errors.)
- Is responsibility clearly assigned? (No shifting of blame to AI.)
How to implement this?
Zhang Chi proposed a “credible control framework” that converts regulatory requirements into specific measures:
- Data contracts: The data used by AI must have clear sources and reliable quality.
- Tool contracts: The tools AI uses (e.g., for checking policies or calculating premiums) must have standard interfaces and verifiable results.
- Result validation contracts: AI’s outputs must be verified by a rule engine to meet financial standards.
- Traceability and rollback mechanisms: Every AI action must be recorded, allowing for auditing and recovery in case of issues.
In plain language: Regulation isn’t about preventing the use of AI; it’s about ensuring it operates safely and responsibly. Each step must be documented, justified, and traceable. This framework has already been implemented in some leading banks and insurance companies.
---
4. The “Five Levels of Financial AI Maturity”: Where Are You?
Zhang Chi also proposed a practical “five-level maturity model” for financial AI:
- Level 1: Can chat: Answers simple questions, like a customer service robot. Current status: Most companies are at this level.
- Level 2: Can write documents: Generates reports, summaries, emails, etc. Current status: Most companies are at this level.
- Level 3: Can verify sources: Provides answers and explains the source and relevant policies. Current status: A few companies have reached this level.
- Level 4: Meets regulatory requirements: The AI’s process is compliant, with clear data and tool contracts, and its results can be audited and rolled back. This should be the minimum standard for financial institutions.
- Level 5: Can perform complex tasks independently: Works reliably like a human employee. This is the future direction.
In plain language: If your insurance AI is still just for chatting or writing reports, it’s a toy, not a productive tool. True financial AI must reach Level 4 or above, meaning it’s auditable, traceable, and responsible. Only then can it be effectively integrated into business processes and create value.
---
Conclusion: The Second Half of AI Depends on Engineering and Business Understanding
The discussions at the Bund Conference sent a clear message:
The competition in AI for the insurance industry has shifted from showing off technology to implementing practical solutions.
- For insurance companies: Stop chasing large model sizes. Focus on your business processes, identify real pain points, and develop targeted AI products. Invest in building a reliable control framework to ensure compliance and safety.
- For tech companies: Stop selling models alone. Provide industry-specific intelligent solutions that transform probabilistic models into reliable business processes and implement regulatory requirements through engineering.
- For consumers: In the future, you’ll enjoy more accurate, convenient, and transparent insurance services. AI will no longer be just a cold chatbot; it will be a knowledgeable, reliable, and accountable “intelligent assistant.”
In one sentence: AI is not magic; it’s a tool. In the insurance industry, only by putting AI within the constraints of compliance and certainty can it truly add value.
Thank you for listening.