Insurance AI Starts to Enter the "Income Statement": From Technical Verification to Business Value Realization
The insurance industry is one of the most suitable industries to test the commercial value of AI (Artificial Intelligence).
The reason is not complicated. Essentially, the operating results of insurance companies depend on their risk identification, risk pricing, and risk management capabilities. From sales and underwriting to claims settlement and anti - fraud, the insurance business is naturally filled with a large amount of complex unstructured information and high - frequency decision - making scenarios, which are exactly the areas where large models are most likely to create value.
Therefore, compared with many industries that are still in the proof - of - concept stage, the insurance industry has a better chance to answer a key question first: Can AI truly enter the business operation system and create quantifiable business value?
Based on the currently disclosed industry practices and public data, the application of large models in the insurance industry is gradually moving from the scenario exploration stage to the value verification stage, and some core business scenarios have begun to show relatively clear return on investment (ROI) effects.
Insurance AI is entering the "business value realization" stage
From the currently disclosed industry practices, different types of insurance institutions and insurance technology companies are exploring AI applications based on their respective business characteristics, and have begun to form relatively clear value feedback in some core scenarios.
For example, Ping An Property Insurance disclosed that the scale of loss reduction through intelligent anti - fraud will exceed 10.5 billion yuan in 2025; China United Property Insurance launched the "AI Xiaohang" intelligent sales platform, covering links such as quotation, renewal, marketing, and customer management. After the platform was launched, the quotation efficiency increased by 50%, the data query efficiency increased by 90%, and the output efficiency of marketing copy increased by 3 times; Yuanbao Insurance built an intelligent customer service system based on large models, achieving 7×24 - hour service coverage, saving 25% - 30% of the working hours of manual customer service, realizing full - scale coverage of customer service quality inspection, and the detection rate of problem calls reaching 95%.
These cases cover different links such as risk management, sales operations, and customer service, but they all reflect a common trend: AI is gradually entering the core business operation processes of insurance companies from an auxiliary tool and beginning to affect key indicators such as cost control, risk management, and operational efficiency.
This characteristic also determines the difference between the insurance industry and many other industries in the path of AI application.
Essentially, the insurance industry is highly dependent on risk identification, process operation, and complex information processing. From sales, underwriting, claims settlement to customer service, risk control, and anti - fraud, a large number of business scenarios are characterized by high frequency, complexity, and dense rules.
Especially in the fields of health insurance and life insurance, a large number of businesses have long relied on manual processing of unstructured information such as medical records, inspection reports, policy terms, customer service recordings, and claims materials.
And the understanding, summarization, and reasoning of complex unstructured information are precisely one of the most advantageous capabilities of current large models.
From this perspective, the insurance industry has strong potential for AI value transformation and is one of the industries that were among the first to have the conditions to form a commercial closed - loop.
In fact, the insurance technology field has always faced a common challenge in the past few years: Many AI projects can demonstrate technical capabilities, but it is difficult for them to stably enter the main business process and continuously affect business indicators. Some so - called intelligent applications still essentially remain at the levels of customer service automation, process automation, or rule engine optimization, and there is still a large gap between them and the core business operation system.
What's new at the current stage is that with the gradual maturity of large models, multi - Agent collaboration, knowledge base systems, and local deployment capabilities, AI has begun to have the ability to call across systems, disassemble complex tasks, and coordinate processes. More and more insurance institutions are starting to try to embed AI into core business links such as underwriting, claims settlement, sales, compliance, and risk control.
From the perspective of the industry development stage, the focus of competition in insurance AI is also changing: from focusing on the display of technical capabilities to focusing on the realization of business value.
Intelligent underwriting and claims settlement: The AI scenarios closest to the profit statement
From the current industry practices, the first scenario where AI forms a value closed - loop is not marketing and customer acquisition, but the most core "two - core" businesses in the insurance industry - underwriting and claims settlement.
This phenomenon is not accidental.
Underwriting and claims settlement are essentially the core links of risk management for insurance companies. At the same time, they are also the most typical scenarios in the insurance business chain with high complexity, high professional thresholds, and high density of unstructured information.
Behind a single insurance policy, there are often a large amount of complex information such as medical records, inspection indicators, past medical histories, medical terms, imaging reports, accident descriptions, and liability clauses. For a long time, these links have highly relied on the experience judgment of professional personnel. Not only is the processing efficiency restricted by human resources, but there is also the problem that experience is difficult to standardize and replicate.
From the perspective of the industry development history, the training cycle for underwriters and claims adjusters is generally long, and their core capabilities often come from long - term business accumulation. How to transform individual experience into replicable and scalable organizational capabilities has always been an important issue for the insurance industry.
The emergence of large models provides a new technical path for this problem.
Its core value lies not only in automating the processing process but also in being able to understand, summarize, and reason about complex unstructured information and precipitate some professional experience into reusable digital capabilities.
Currently, this ability has begun to be applied in the underwriting and claims settlement links.
For example, in the field of intelligent underwriting, KEYI.AI, an intelligent underwriting expert launched by Shuidi, relies on the insurance knowledge graph and RAG capabilities to achieve automatic risk identification and underwriting decision - making for complex health insurance. Project data shows that the accuracy rate of underwriting consultations exceeds 99%, the response speed has increased by 260 times, and the proportion of rejected customers being matched with suitable products has increased by 6 times.
In the field of claims settlement, the application of AI has further extended to the entire business process. China Pacific Insurance embedded the AI employee "Lingxi" into multiple links such as claims review, risk early warning, and quality management. After the application, the proportion of manually entered fields has dropped to less than 10%, the team's operation efficiency has increased by more than 30%, and the case quality score has increased by 25%.
On the surface, these achievements are reflected in the improvement of processing efficiency and the reduction of operating costs. However, from a business operation perspective, their deeper significance lies in the fact that AI has begun to have an impact on the core business indicators of insurance companies.
The profitability of insurance companies essentially depends on their risk pricing and risk management capabilities.
Underwriting determines the quality of underwriting risks, and claims settlement determines the level of claim cost control. Together, they affect key business indicators such as the claim ratio, expense ratio, and comprehensive cost ratio.
For a long time in the past, the digital construction in the insurance industry has mainly focused on the onlineization of processes and the computerization of systems, which is essentially an information technology upgrade. Its main function is to improve process efficiency rather than change the risk decision - making mechanism.
The important difference between the current AI application and the previous digital construction is that it has begun to gradually participate in the processes of risk identification, risk judgment, and risk management.
In other words, AI is changing from a process tool to a risk management tool.
This is also an important reason why underwriting and claims settlement have become the first scenarios in the insurance industry to achieve value verification. Compared with front - end links such as sales and marketing, the "two - core" businesses are closer to claim costs, operating expenses, and operating profits, and their application value is easier to measure and verify through business indicators.
From the currently disclosed practices, the first area where insurance AI forms business value is not traffic acquisition but risk management. This may also be one of the important characteristics that distinguish the insurance industry from most consumer Internet industries.
Agent intelligent agent "new infrastructure": Accumulating "digital labor force"
If large models improve information understanding and knowledge processing capabilities, then the changes brought about by Agent intelligent agents are more like a reconstruction at the organizational capacity level.
From the current industry development trend, Agents are promoting insurance AI to move from single - point tool applications to process - collaborative applications. This may be one of the most notable changes in the insurance industry since 2025.
In the past few years, most AI applications in the insurance industry have mainly been for auxiliary decision - making. Whether it is intelligent customer service, knowledge Q&A, or office assistants, they essentially belong to the "Copilot" mode, that is, AI provides suggestions, and humans ultimately complete task execution and process advancement.
In this mode, AI mainly plays the role of an efficiency tool. However, the development of Agents has begun to break through this boundary.
Compared with traditional AI applications, Agents can not only understand information but also complete task disassembly, process planning, system calls, and result feedback according to established goals and collaborate with other Agents to complete complex tasks.
For the insurance industry, this ability has special significance.
Insurance companies are essentially typical process - driven organizations. There is a large amount of cross - departmental collaboration among links such as sales, underwriting, claims settlement, customer service, risk control, and compliance, and a large number of business processes require multiple approvals, verifications, and system interactions.
In the past, even if AI had certain analysis capabilities, it was difficult to truly enter the complete business process. The emergence of Agents has enabled AI to have the ability to participate in the operation of business processes.
Currently, some insurance institutions have begun to explore relevant practices. For example, Ping An has formed a large - scale intelligent agent application ecosystem. Public data shows that employees have developed more than 70,000 intelligent agent applications in total, and the annual model call volume has reached 3.65 billion times; Allianz Life Insurance cooperated with Volcengine to build an intelligent marketing system, covering multiple scenarios such as intelligent customer service, intelligent training, intelligent outbound calls, and conversation analysis. After the project was launched, the sales training cycle was shortened by 3 times, and the rate of customer consultations transferred to manual service decreased by 10%; China Re Property Insurance built an intelligent bill processing platform based on AI Agents, realizing full - process automation of bill identification, information extraction, data verification, and system write - back. Currently, it covers hundreds of bill formats, and the business coverage exceeds half of the company's annual bill processing volume.
From these practices, the insurance industry is gradually taking shape of a "digital labor force".
Compared with simple automation tools, the greatest value of this kind of digital labor force is not just to replace some repetitive labor but to gradually precipitate the knowledge and capabilities originally scattered in individual experience into organizational capabilities.
In recent years, more and more insurance institutions have begun to try to disassemble professional knowledge such as clause understanding, underwriting rules, risk identification, and sales compliance into reusable capability modules and embed them into the Agent system to achieve the precipitation and reuse of knowledge and capabilities.
From the perspective of knowledge management, this means that a large amount of implicit experience accumulated by the insurance industry for a long time is being gradually digitized, structured, and systematized.
In the past, the professional capabilities of an excellent underwriter or claims adjuster often required years of practical accumulation. In the future, some experience and capabilities are expected to be precipitated through the way of "knowledge base + model + process" and replicated on a large scale within the organization.
The significance of this change may be no less important than the improvement of efficiency itself.
For a long time, the insurance industry has essentially been a typical labor - intensive knowledge industry. Many core capabilities are in the hands of individuals rather than being precipitated in the organizational system. Therefore, business expansion often accompanies personnel expansion, and the organizational scale is highly related to the manpower scale.
The development of Agents provides another possibility for the insurance industry.
When more and more professional knowledge and business processes can be digitized, precipitated, and reused continuously, the ability boundary of insurance companies will no longer depend entirely on the number of employees but more on the accumulation level of their digital capability assets.
From this perspective, the value of Agents lies not only in improving efficiency but also in promoting the transformation of insurance companies from "relying on individual experience" to "relying on system capabilities".
This may also be an important sign for the insurance industry to move towards the next stage of intelligent development.
Competition in the next stage: Competing in the business operation system
For a long time in the past, discussions about AI in the insurance industry have mainly focused on the model capabilities themselves.
Who accesses large models first, who has industry models, whose parameter scale is larger, and whose response speed is faster have often become the focus of industry attention.
However, from the practices in the past two years, the evaluation criteria for AI in the insurance industry are changing.
With the rapid development of open - source large models, model capabilities are being rapidly popularized. Especially since 2025, domestic open - source models represented by DeepSeek have promoted large models to enter the stage of "technological equal rights". The cost of model acquisition and the deployment threshold have significantly decreased, and local deployment has gradually become the standard configuration for insurance institutions. More and more insurance companies are starting to carry out scenario fine - tuning and application development based on open - source models. The large model capabilities themselves are gradually evolving from a differential competition factor to a general infrastructure.
In this context, the focus of industry competition has also begun to change. Compared with model parameter scale or reasoning ability, insurance institutions are more concerned about how to build a high - quality industry knowledge base, precipitate reusable business rules, connect data and business processes, and transform long - term accumulated risk experience into sustainable iterative digital capabilities. The model provides general intelligence, and what really determines the application effect is increasingly the insurance industry's own knowledge system, data assets, and business scenarios.
This is particularly important for the insurance industry.
The insurance industry naturally has strong regulatory, compliance, and risk - control attributes. In key links such as underwriting and claims settlement, AI not only needs to give results but also needs to explain the basis for the results and meet audit and regulatory requirements. Therefore, the insurance industry's demand for AI has never been just the model capabilities themselves but a set of capability systems that can be deeply integrated with the business operation system.
In the past, the core capabilities of the insurance industry were mainly precipitated in experienced professional talents and organizational processes. In the future, more and more industry knowledge, risk experience, and business logic may be continuously accumulated, reused, and iterated in the form of "data + knowledge + model".
In this sense, the focus of insurance AI applications is gradually shifting from model capability construction to knowledge system construction and business operation system construction. How to transform industry knowledge, risk experience, and data assets into sustainable accumulated digital capabilities and truly embed them into the core business processes of insurance companies may become an important direction for the industry's development in the next stage.
(Author Wang Yan is the founder of Beijing Shuimu Fanglue Consulting Co., Ltd.; Editor: Yang Rui)
This article is from the WeChat official account “Caijing Mayflower” (ID: Caijing - MayFlower), author: Wang Yan, published by 36Kr with authorization.