Exploring AIGC Applications in Insurance: Insights from ZhongAn Insurance CTO Jiang Jiyun

Interview Guest | Jiang Jiyun, CTO of ZhongAn Insurance

Author | Luo Yanshan

Different models have distinct advantages in various scenarios; enterprises must fine‑tune large models to meet customized needs, and implementation varies by company, becoming a core competitive advantage.

Each wave of technology reshapes industries, and forward‑looking practitioners adopt new tech to optimize processes and improve products, gaining a competitive edge.

ZhongAn Insurance is a proactive pioneer in the AIGC (AI‑generated Content) wave, having started researching ChatGPT in 2022 and exploring its business applications.

"With the rise of AIGC, collaborative intelligence will become the scientific frontier of digital‑to‑intelligent transformation, especially for data‑intensive, human‑service industries like finance and insurance, which will see rapid breakthroughs," said Jiang Jiyun in a recent InfoQ interview.

Exploring AIGC in Insurance Business: Dual‑Track Technology and Business

As an internet‑based insurer, ZhongAn uses technology to break traditional industry barriers. Currently, AIGC is applied across many internal processes, especially in marketing services, where it helps staff quickly generate copy, visual designs, and optimize tweets.

For copywriting, AIGC tools generate brand‑aligned text, titles, descriptions, and slogans, boosting efficiency and enabling bulk creation of educational articles.

In visual design, AIGC platforms produce brand‑consistent graphics in hours instead of days, covering posters, live‑stream backgrounds, icons, and animations, accelerating product and campaign launches.

AIGC’s text‑analysis capabilities also optimize marketing content by analyzing social‑media data to craft more engaging tweet titles, bodies, and tags, increasing click‑through and exposure.

Beyond content generation, AIGC powers intelligent customer service, automatically drafting responses during calls and summarizing sessions afterward, extracting new leads from semantics.

Traditional chatbots rely on NLP; ZhongAn’s new LLM‑based assistant improved accuracy by 25% in A/B tests, delivering higher‑level intelligence.

These applications have received positive feedback from business units, and staff are increasingly learning AI techniques, fostering a two‑way sprint between technology and business teams.

Targeted AIGC Applications

Using tools like ChatGPT or Midjourney may seem simple, but real‑world vertical applications are far more complex.

Companies adopt strategies such as building their own models, fine‑tuning, or using embeddings. ZhongAn recommends embeddings for mid‑size insurers because building large models demands massive data, resources, and labeling effort, and raises compliance concerns about using customer data.

Since ChatGPT‑3.5 does not support fine‑tuning, ZhongAn built an embedding‑based AIGC platform called Lingxi, offering a standardized development paradigm that abstracts text embedding, knowledge‑base slicing, and vectorization complexities.

Jiang notes challenges in applying AIGC to insurance scenarios.

Finding Value Points – Conversation Is Not the Only Form

First, ZhongAn must productize best practices, fine‑tune internally, and improve the model.

Although AIGC excels at natural language and logical reasoning, it may falter on highly emotional or complex questions, highlighting a limitation of general AI.

Thus, the team abstracts complex problems into forms solvable by large models, creating prompt‑engineering and knowledge‑base solutions to aid users beyond generic conversational interfaces.

Practical advice: stay value‑driven, identify where AIGC adds business value, and avoid using it merely to prove feasibility.

Value analysis should consider workflow simplification, user‑experience improvement, throughput increase, and cost reduction; if benefits are unclear, resources should not be invested.

Conversation Is an Option, Not the Only Option

Relying solely on dialogue is inefficient; ZhongAn’s engineers accumulate prompt patterns for specific problems and design graphical interfaces that hide complexity, letting business users focus on their expertise.

In short, ZhongAn encapsulates and engineers prompts to make AIGC responses more efficient and purpose‑focused.

Ensuring Reliability and Security

Beyond scenario‑specific customization, AIGC raises sensitive‑information security concerns.

Since large‑model training involves massive data, prompting often requires sharing context with vendors, risking privacy leaks.

ZhongAn addresses this by implementing platform‑level safeguards: input/output encapsulation, audit capabilities, sensitive‑information detection and masking, and strict API authentication.

The Lingxi architecture separates Model‑as‑a‑Service (MaaS), application framework, and platform layers, enabling controlled interception of inputs and outputs.

Audit logs track user inputs and model outputs for risk tracing.

Sensitive‑information detection uses regex and a keyword knowledge base to block illegal inputs/outputs, preventing political, data, or personal leaks.

All model APIs require authentication and robust access control to ensure legitimate, traceable usage.

How the Insurance Industry Responds to Technological Development

Historically, insurers relied on outdated machine‑learning pipelines with fragmented models built from scratch, limiting efficiency.

While leading insurers now adopt ML platforms and feature‑engineering tools, overall efficiency remains low.

Since 2010, ZhongAn has used multimodal models (text + structured data) to improve health‑insurance claim risk models, and from 2022, Transformer‑based user‑behavior sequence models have outperformed traditional methods in real‑time ad‑placement, simplifying feature engineering for sparse data.

These new model strategies help the company keep pace with rapid tech evolution.

Jiang emphasizes that AIGC is not a universal solution; each scenario must be evaluated for suitability.

Different models excel in different contexts; enterprises can fine‑tune or engineer prompts to meet custom needs, and implementation varies per company, becoming a core future competency.

Currently, ZhongAn’s AIGC services are internal exploratory projects, not yet commercialized.

In the short term, AIGC will be applied to intelligent marketing, core insurance systems, DevOps platforms, and data products; long‑term, it will underpin differentiated core competitiveness, essential for standing out in China’s competitive insurance market.

ZhongAn may later incubate products and share outcomes with the broader insurance sector, narrowing the AI capability gap between small‑ and large‑scale insurers and offering “curve‑overtaking” opportunities.

-END-

Followers of the "ZhongAn Technology Team" are encouraged to star the public account to stay updated.