Currently, the insurance industry is facing a deep transformation. With the introduction of the “Ten Rules of the New Country” policy for the insurance industry, stricter standards and requirements have been put forward for improving the insurance capabilities and service levels of the personal insurance industry. This means that the insurance industry needs to leave the expansive development model and move towards higher quality and more standardized services. In this transformation process, insurance institutions need to continuously improve the quality of insurance services to meet new market and regulatory requirements.

In order to better standardize insurance service behavior, under the power of the big model, Waterdrop launched the “AI Big Model Insurance Quality Inspection Solution”, which greatly expanded the scope of quality inspection, improved work efficiency, and effectively controlled operating costs.

The traditional quality inspection method uses manual sampling. Quality inspectors sample recordings of the entire customer service process and mark risk content. A quality inspector completes an average of 10-15 single-recording inspections every day, and the quality inspection channel is relatively single, which cannot cover the company's WeChat, public account, circle of friends, etc. It is a job that consumes a lot of manpower but has a ceiling.

In recent years, many technology companies have tried to introduce intelligent quality inspection technology into the insurance industry, and small-model quality inspection is difficult to understand complex changes in insurance service logic and language expressions. The accuracy rate is relatively low, and the risk of missing inspections is high.

“Waterdrop AI Quality Inspection” is based on large model contextual semantic understanding and long-text reasoning ability. It can thoroughly understand complex conversations, user intentions, and emotional attitudes, and identify more hidden and complex quality inspection rules. According to internal data, Waterdrop AI quality inspection can achieve 100% full coverage, including voice, corporate WeChat chat history, friend circle and other channel operation behavior. The cost of quality inspection has also been drastically reduced. Including the cost of manual review, AI quality inspection costs can be reduced by more than 50% compared to manual quality inspection costs (not considering early development costs).

Behind the “AI quality inspection” capability, it is inseparable from the deep empowerment of Shuidi Company's self-developed “Waterdrop water protection model”. In the past, due to the variety of insurance products, relatively complicated terms, and lack of massive data for model training, it was difficult to break through the accuracy of AI quality inspection.

Since 2019, Waterdrop has invested in AI construction. Under the premise of compliance, it has accumulated a large amount of industry segmented data, including massive vertical insurance service corpus samples, more than 7,000 current or past popular insurance product data, and is proficient in 10,000+ professional issues in the medical insurance field. Problems such as irregular service attitudes, excessive sales promises, misinterpretation of product coverage, and inadequate purchase insurance reminders can be identified as soon as possible. After continuous training, the accuracy rate of AI quality inspection is close to the level of manual inspection based on 100% manual detection accuracy.

According to reports, Waterdrop's “AI Big Model Insurance Quality Inspection Solution” has formed a solution for external export and can be opened to the industry. The next step will be to continue to strengthen quality inspection accuracy training, help more insurance institutions upgrade service efficiency and service quality, and help the high-quality development of the industry.