1 Introduction
Underwriting is the insurance company's decision whether to underwrite the insurance for the client, at what rate, and the conditions for underwriting based on the information provided by the client and obtained by the insurance company. For insurance companies, underwriting is a means to control costs and risks, so as to ensure the stability of the company's operations and business. For customers, underwriting makes different underwriting results according to the physical condition of different customers, which can maintain fairness among insured individuals. As an Internet insurance company, ZhongAn conducts most of its underwriting automatically online. Faced with the online underwriting scenario on the Internet, the information of the insured, the region and channel of the insured all increase the uncertainty of underwriting, which means that it is necessary to reserve and continuously introduce a large number of underwriting risk control methods. Market conditions, medical development, etc. are constantly adjusting the rules. Since ZhongAn launched health insurance products in 2015, it has launched more than 2,000 products, including the "Exclusive Series" and other star products that have sold over 10 billion premiums. In recent years, health insurance has developed rapidly, taking the lead in the entire insurance industry, and the difficulty of underwriting risk control has also increased, which in turn has led to the demand for intelligent underwriting engines.
2. Thinking and Challenges
2.1 Business level
1. Fine management of policies
The risks faced by different product types are different, and there will be obvious differences in underwriting. Medical insurance is very strict, while accident insurance and life insurance are relatively loose. Critical illness insurance can be used as a reference. For example, medical insurance needs to pay attention to the customer's age, gender, social security, health status, disease history, moral hazard, etc. Compared with medical insurance, Accident insurance pays more attention to occupation, living environment, etc. In different periods of the insurance company's operation, the benefits and purposes brought by the products to the company are different. Under normal circumstances, when insurance companies hope that this product will rush out to seize the market and attract more premiums, the underwriting requirements for the product will be relaxed in policy, such as: new product promotion period, good start period, certain periods of premium shortage . Even for the same product, there are differences in policies between new insurance and renewal, and most of the underwriting rules are reusable.
Therefore, policy management requires flexible arrangement and real-time adjustment for various underwriting rules, and also supports rule set management. In addition, the following points need to be paid attention to in the implementation of underwriting:
• Stability
Due to the situation of calling external third-party services, out of consideration for the overall stability of underwriting, it is necessary to have the ability to downgrade and remove the rules.
• Dealing with timeliness and rule dependencies
Excessive underwriting time will affect the conversion of insurance premiums. In order to ensure the experience of online customers, the execution of underwriting rules needs to have the ability to execute in parallel. At the same time, data dependencies between rules need to be considered, such as the checksum of health risk levels The verification of the cumulative risk insurance amount and the upper limit setting of the user's cumulative risk insurance amount depend on the evaluation results of the health risk level.
• cost
Third-party risk control rules are generally fee-related interfaces. When the rules are executed in parallel, if a branch has already given an interception conclusion, there is no need to execute other branches. At the same time, the rules also need to support caching to reduce additional costs.
2. Data index management
Multiple specific underwriting rules constitute an underwriting strategy. Sometimes it is necessary to analyze the execution results of each underwriting rule, and adjust the underwriting strategy according to the distribution of execution results according to the current market demand or the company's business needs.
For example, different risk control rules have been introduced for medical care, accidents, and serious illnesses. Based on the pass rate and rejection rate of relevant risk control rules, the impact of factors such as risk, cost, and premium growth on the company's business will be comprehensively assessed. The result determines the entry conditions of the risk control rules or the adjustment direction of the interception strategy; therefore, the management of data indicators needs to conduct statistics and analysis on the distribution of the execution results of each rule.
3. Policy log management
With the increasing number of Internet insurance users, online insurance underwriting has also faced new challenges, and efficient solutions to insurance consultation issues have also become an important part. Facing the insurance issues of different users, the underwriting post needs to give the most suitable solutions, such as whitelist, product recommendation, underwriting policy adjustment, etc.; the choice of solutions requires specific data support, including the implementation of the underwriting policy Specific rules, the execution order of the rules, the execution status of each rule, etc., detailed execution logs become the most powerful basis for supporting these decisions.
4. Rights Management
The formulation of the underwriting policy needs to be based on the characteristics of the product itself and the actuarial conclusions. The result of the policy configuration needs to be strictly reviewed. Therefore, authority management needs to be strictly controlled for the launch and adjustment of the policy.
5. Version management
随着产品的快速迭代，核保策略也会随着时间的推移按照当下业务需求做出相应的调整，每次调整都会衍生出不同的版本，因此版本管理需要能够保留核保策略每个版本的历史状态，以便于查看和核实核保策略的制定历程，同时也能够支持版本之间切换。例如，保险产品上线时制定了针对出生30天至65周岁的被保人进行医疗风控校验的核保策略，为了提升核保时效并降低成本，针对30周岁以内的被保人则无需进行医疗风控校验，经过一段时间后根据投保结果分析评估，调整放宽年龄段。
6. 面向业务人员的设计
一款保险产品主要包括条款、责任、费率、保障期限等，核保策略实际上也是围绕着这些元素来设置的。涉及医疗险条款，核保策略就很严格，针对意外险则会相对宽松。这也意味着业务人员配置核保策略时，也应该是能按照保险产品的结构进行的。在互联网的大背景下，互联网保险公司的产品上线和迭代较之传统保司频率更高，而配置产品核保策略又是产品上线流程中的重要一环，如何保证业务配置人员在几千款产品几千个渠道的不同营销活动中快速配置，并且能够跟外部系统打通流程也是核保引擎需要特别关注的，这也是智能化核保规则引擎最需要解决的问题。
2.2 实施层面
• 灵活的部署方案
决策引擎的部署一般分为Saas模式和私有化部署模式。不同的业务场景可根据自身业务特点选择不同的集成方式。
Saas模式： 决策引擎产品部署在供应商提供的云服务器上，由供应商负责部署和运维决策引擎相关系统，客户策略人员需要在供应商提供的SaaS管理后台进行策略配置，客户接入系统通过HTTP、RPC等形式调用供应商提供的决策服务开放接口。客户所有的策略及业务数据均沉淀在供应商的数据环境中，存在一定策略及业务数据的安全性问题。
私有化部署模式： 供应商将决策引擎相关系统部署在客户公司内部服务器上，客户可根据业务情况动态灵活选配资源，日常运维工作由供应商运维人员负责。客户策略业务数据及系统数据均在客户公司内部沉淀，数据安全性更高。同时，公司内部系统通过内网接入，成本更低。
3. XD决策平台的引入
3.1 XD的介绍
X-Decision决策系统（简称XD）是由众安自主研发的一款实时规则引擎，承载着策略编排和计算的相关任务， XD的规则设计器采用纯浏览器编辑模式，无须安装任何工具，打开浏览器即可完成复杂规则的设计、测试和发布。 XD决策引擎助力金融、保险等多场景构建专业、强大、灵活的决策平台。全场景覆盖，让决策更高效。
3.2 XD在保险领域的思考
随着保险业务的迅速扩张，产品形态千变万化，营销渠道不断丰富，所对应的核保策略管理要求也越来越精细化。保险公司都在进行重大的数字化转型，消费金融领域也在逐渐渗透到保险行业，随之核保策略维度不断细分，业务人员对保险产品的核保策略管理也越发棘手，亟需智能化决策引擎对核保策略进行精细化统一管理。传统决策引擎作为工具型决策引擎，提供了通用规则的灵活配置能力。但面对核保领域的业务，如何同时解决互联网核保领域产品数量众多、业务场景策略差异大等痛点，作为通用化的决策引擎，一时也难以满足核保策略运营需求。
X-Decision决策系统作为产品化的决策引擎，不仅提供了通用的规则配置能力，还提供了两核领域的智能化解决方案，满足了互联网核保诉求。在早期，X-Decision决策系统就开始调研决策引擎在互联网核保领域的痛点，保险业务的痛点复杂点梳理：
• 核保规则统一管控
核保人员基于核保维度差异化将日常的核保规则管理进行区分，大体分为通用性公共规则、产品大类规则、产品级细分规则、条款规则、责任及渠道级别规则。同一条通用的核保规则（如同一产品可以在若干个不同的渠道销售，针对地域的风控限制有所差异）挂靠在不同的产品层级及营销渠道下，策略阈值限制有差异化表现。同时不同产品、渠道层级下使用规则也各有不同。
产品规则：
投保人年龄为18-55周岁;
姓名合法性校验;
医疗风控校验;（规则进入条件:30-55周岁且被保人所属地区为XX省XX市）
针对通用规则的修改，核保人员期望可视化展现规则调整影响匹配的产品、渠道、条款及责任等。不同的保司，其产品结构灵活多样，需要对规则管理基于细分条件进行精细化配置管理，其规则复杂度不胜枚举，而通用的决策引擎不具备产品属性，如为满足此类规则管理，规则数不计其数，核保人员恐对核保规则管理望而却步。
• 产品中心与策略映射
保司在数字化转型过程中，会建设产品中心，维护产品大类、产品细分、产品属性、营销活动等信息，其中规则配置属性限制需要基于产品属性区分。 随着外部环境多变，同时应对新形式下的监管要求，核保政策也需要实时动态调整。如通用决策引擎要支撑此类业务场景，满足产品规则差异化阈值，就需要配置维护此类数据（如同一产品，在A渠道投保时需要校验A机构风控校验，在B渠道投保时需要校验B机构风控校验）。如单独在决策引擎维护此类信息，会导致决策引擎过多的侵入核保业务，可能造成产品中心与决策引擎两边配置数据差异化，对数据映射维护成本增高。此时需要动态参数引用来解决同产品不同渠道参数差异化问题。纵观通用决策引擎，暂无此类产品参数配置能力。
• 产品规则业务视图
核保人员在策略管理过程中，为增加业务透明度，需核保人员维护产品与规则的关系。如要高效对策略管控，规则可视化为重中之重，需直观呈现产品大类下的不同产品、渠道、责任、条款下对应规则。通用决策引擎无产品属性，无法与产品中心进行规则关联绑定，无法让业务人员以既有的业务视图配置规则。另外，针对线上核保数据，规则执行轨迹可视化也尤为重要，核保数据被规则拦截，规则回溯、执行轨迹呈现是核保人员对策略是否有调整空间最直观的体现，如何将业务模型融入通用决策引擎成为难点。
3.3 XD在保险领域的实践
X-Decision决策系统本身是一个开放的、策略内容完全自定义的决策平台，具有超高的灵活度等特点。对于医学专业知识储备极为丰富核保专家， X-Decision对于核保业务人员极不友好。为此，X-Decision需解决的是如何将业务模型融入到策略配置中，让核保业务人员能够按照既有的业务视图配置规则。
1. 业务模型组件：
业务模型组件在X-Decision中应运而生，X-Decision 抽象各产品渠道配置参数，形成业务模型。例如针对产品在不同渠道下风控校验的差异化，通过产品中心将风控因子同步至决策引擎，为风控因子在规则中动态配置所引用。由此，X-Decision依靠业务模型与业务领域连接，通过单向同步等方式打通业务模型结构、数据与X-Decision之间的通道，形成了产品组合模型、条款模型、责任模型、渠道模型、营销活动模型等业务人员熟知的业务视图，不需要在多系统中反复调整，从而保障规则变更操作的高效性、安全性。
此外，多个模型之间可以通过组合，形成复杂的树型结构。业务模型组件可以根据各保司下产品结构不同进行灵活动态配置，满足保司规则业务模型差异化配置等特点。
从规则的角度出发，相同规则在不同产品组合下在校验阈值上稍有差异。由此，XD把规则校验条件以及阈值映射到业务模型上，形成同一维度下不同的数据表现。
2. 产品策略配置模块
为解决产品与规则的绑定关系，使得核保人员对产品和规则管理更加敏捷、高效，提升规则维护可视化，X-Decision新创产品配置策略模块。首先X-Decision提供资源库抽象规则，供多个产品组合引用。在执行具体产品组合规则时，由抽象规则结合具体产品组合的属性数据，得出不同的校验结果；其次，为便于规则管理，规则编写完成后，X-Decision先对规则进行业务分组、形成规则组；最后，X-Decision提供产品配置策略模块，可配置公共规则集、业务模型策略，主要解决在不同产品、渠道下差异化规则配置需求，同时解决产品规则可视化痛点。核保业务人员可在产品配置策略模块配置不同节点，如公共规则集模块、产品大类通用规则模块、产品细分规则模块等。
在相应模块节点，核保人员可快速选择事先已基于不同业务分组的规则或规则组与模块建立绑定关系。当产品中心上架新产品或渠道时，会与X-Decision进行模型数据同步。同步完成后，核保人员可在产品模型条件中快速选中新产品，并对产品下各渠道配置规则绑定关系。产品配置策略模块降低了业务与技术人员的沟通成本，减少信息偏差，节省技术开发资源。
3. 产品策略执行轨迹：
X-Decision为产品策略配置模块针对核保数据提供了树型执行轨迹追踪功能，便于业务人员追踪回溯策略执行链路，以便快速排查产品在各渠道售卖中的出单转化等问题。
3.4 健康险个险在线核保对接XD核保规则引擎
健康险在线核保对于系统稳定性要求极高，最终采用了Saas模式对接，并设计了对应的降级策略。由于部分风控规则依赖于第三方机构且校验逻辑复杂度较高，包含缓存、白名单等，健康险系统内已经开发实现，出于系统整体的稳定性和开发工作量的考量，选择由健康险业务运营系统侧根据XD接口规范提供回调接口，支持XD在执行到相关依赖第三方的规则时调用健康险业务运营系统，由健康险系统执行既有校验逻辑和功能，将校验结果返回给XD，以保证既有功能的完整性，且通过该接口支持多个规则合并回调，减少交互，避免不必要的网络开销。
4. 总结和展望
X-Decision决策系统在核保领域取得重大创新突破，实现核保运营对策略精细化、差异化、可视化、智能化管理，提供核保领域行业解决方案，加快了保险运营能力蜕变速度，填补决策引擎在核保领域的产品化空白。健康险产品核保从2022年7月首款产品接入XD至今已经历了半年验证，执行准确且性能稳定，新产品上线效率提升了60%以上。实际上健康险除了在线核保场景上使用XD规则引擎外，在核保产品推荐、保全复核流程、理赔自动理算中都已经全面使用XD，并取得良好的业务效果。
在未来，XD将在产品化道路上持续沉淀，以解决用户和业务痛点为出发点，提供更加轻量、易用、可插拔的决策组件。此外，XD还将从用户体验着手，优化策略配置方式，简化系统接入流程，让无论是开发人员还是技术小白的使用者都能灵活上手。同时，在具体的业务领域上不断深耕，依靠灵活多样的决策组件，快速搭建适用于业务的决策平台，提高XD在特定决策领域下功能和产品形态的多样性，助力其向智能化决策发展。
5. 本文作者
葛超、夏兆旭，来自众安保险-金融事业部-金融科技中心
袁江、王光，来自众安保险-健康险事业部-产研中心
6. 参考文献
-
核保篇 | 核保的概念和影响因素 zhuanlan.zhihu.com/p/153055753
-
核保为什么是购买保险的重要环节？zhuanlan.zhihu.com/p/14