Interaction design of financial insurance products under the Era of AIoT

In 2016, insurance regulatory authorities put forward the requirement of adding ‘dual-recording’ in insurance services for the first time, and various financial institutions such as banks, insurance and securities have gradually established business norms of ‘dual-recording’ for financial product sales. ‘Dual-recording’ refers to video and sound recording, which is a very important part of insurance business flow, and it is human and time-consuming at the same time. After the recording of the insurance process, it is necessary to conduct a ‘quality inspection’ on the recording process and results. Quality inspection means the inspection and control of insurance companies and banking insurance agencies on the authenticity, integrity and accuracy of people's identity information and sales process through dual-recorded audio and visual materials. At present, the ‘compliance risk’ involved in dual-recording and quality inspection is generally high. Compliance risk refers to the fact that the behaviours of insurance companies, insurance intermediaries, insurance agents and their employees shall comply with laws and regulations, regulatory provisions, self-regulation rules, market practices, the company's internal management system and the ethical code of honesty and trustworthiness. Failure to comply with these guidelines may result in the risk of legal or regulatory penalties, significant financial or reputational loss.

Now, most financial institutions still adopt the traditional way of manual dual-recording and quality inspection in their dual-recording business processes and products, which are characterised by low efficiency, high cost, poor user experience and high compliance risk. With the development of information technology, major financial and insurance products are seeking more efficiency to solve the problems of long-recording time, low efficiency, high-quality inspection cost and high compliance risk.

Observation, recording and interviews were used in this research to analyse the overall process of ‘dual-recording’ and it is divided into two parts: recording and quality inspection. User roles fall into two categories namely insurance agent and policyholder. During the recording process, the insurance agent needs to listen, introduce the product orally, show the product to the customer, confirm the identity of the customer, operate the software and take pictures for the customer. Policyholders need to listen, check information, answer questions, identify the identity, read text, take pictures, sign the document etc. The whole process is relatively tedious, and the error rate is high.

2.2.1

Recording stage

In this research, the recording process is divided into four stages: (A) ask for willingness, (B) mutual verification, (C) product introduction and (D) policy confirmation (as shown in Figure 2). Each stage is divided into several small steps for analysis.

Fig. 2

Mutual verification is relatively complex and cumbersome, including presenting ID cards and confirming written documents, which involves many steps in checking and confirmation stages. The content to be filled involves personal information which makes policy seekers have vigilance and doubt, thus leading to a decrease in desire to buy.

During the product introduction stage, the insurance agents tend to be unfriendly or emotional as this stage requires them to introduce the policy content repeatedly, which takes a long time. Particularly speaking, during the ‘guarantee scheme description’ and ‘clause introduction,’ agents will have frequent interaction with policy-holders. After listening to various clauses, policyholders need to make a ‘clear or unclear, do not understand’ answer. Some policyholders answer at will in order to go through the process as quickly as possible; Some people will be cautious and will ask for more details, and they tend to doubt the content of the policy and agent; Some people easily will give up the idea of purchasing the policy as they cannot understand the benefits and value of the insurance policy well. Therefore, how to use technology and design to help optimise the overall steps and experience of product introduction is particularly important.

In the order confirmation stage, policyholders need to put a series of signatures, and this is the last step of order completion. After a variety of complex interaction processes between the insurance agent and the policyholder, policyholders will sign immediately if the previous process has been smooth; If they have doubts, they will easily abandon the purchase. Therefore, a pleasant purchase experience for the policyholder is very important at this time.

The earliest concept of Human-Computer Interaction (HCI) originated from the interaction between humans and computers. HCI is defined by the Association for Computing Machinery as the discipline of the design, evaluation, implementation and phenomena associated with computer systems. Professor Alandix from the University of Birmingham believes that human-computer interaction is a discipline that studies human beings, computers and the interaction between them. The purpose of HCI is to make computers better serve human beings. Now, it is generally believed that HCI is a process of exchanging information between humans, hardware and software systems to accomplish some specific tasks. Later, it gradually expanded to a wider range of applications. It is a discipline that studies the interaction between systems and users. A system consists of hardware machines and computerised software. Professor Sun Xiaohua divided HCI into three stages in his paper A Review of Collaborative Research on Human-Computer Intelligence: human-physical system interaction, human-digital system interaction and human-intelligent system interaction [4]. According to Hornbæk et al., HCI research is mainly divided into two aspects: interaction mode and interaction quality. Among them, interaction mode is mainly driven by technology, including interaction mode, information equipment, application field, etc. Interaction quality is related to people's perception and cognition, and its emphasis is on the improvement of usefulness, usability, efficiency, aesthetics and other aspects.

We are in the process of moving from HCI to human-computer intelligent collaboration [4]. The relationship between the intelligent system and humans has changed from the simple ‘tool’ relationship in the past to the mutual ‘collaboration.’ AI is not to completely replace human beings with machines, but to use intelligent computing to improve efficiency and accuracy, make up for human's shortage and reduce manpower wastage, and so machines and humans can work together intelligently. Synergy refers to the fact that the application of Information Communications Technology (ICT) products in the production process can improve compatibility among other elements and improve operational efficiency [6,7,8]. AI product refers to the product or system that uses the theory, method and technology of AI to deal with problems [9]. An intelligent system has four characteristics: situational awareness, adaptive learning, autonomous decision-making and active interaction and collaboration [10]. As designers, we should consider how to design a more rational human-computer collaborative process and experience with the help of intelligent technology through design thinking. ‘Human’ in human-computer collaboration refers to ‘multiple user groups,’ that is, there will be diversified user experiences. Gartner put forward the concept of ‘Total Experience’ (TX) [11] in its Strategic Technology Trends Report of 2021. It is predicted that enterprises that can provide TX in the future will outperform far better than their competitors in terms of satisfaction index, thus it can be concluded that experience is an implicit but important element in human-computer collaboration.

AiCore intelligent dual-recording solution is an intelligent solution for optimising insurance dual-recording developed under the overall framework of AIoT, relying on aiCore AI big data platform. Specifically, as shown in Figure 3, the data of audio and video recordings are acquired through mobile devices will be monitored and stored in real-time. We accomplish this by the comprehensive use of biometric recognition, video image analysis and other AI and big data analysis, and text, voice, face and behaviour detection and recognition; it is done through a well-articulated human-computer interaction design and reasonable insurance process design so that the overall optimisation of user experience in the ‘dual-recording’ during insurance business is realised. Moreover, the dual-recording solution provides installation, configuration, monitoring and operation of a complete set of dual-recording services. Mark Purdy et al. [12] believe that AI is not only a tool to improve productivity, but also a new factor of production. In aiCore intelligent dual-recording solution, AI makes recorded video data more compliant, which is a practical example of using AI as a new factor of production.

Fig. 3

The aiCore intelligent dual-recording solution fully considers the participation rate and experience of different roles (including policyholder, insurance agent, quality inspector and administrator) in the whole recording process, which is the comprehensive experience proposed in Gartner's report, to realise the complete dual-recording interaction process combined with information technology (Figure 4). The design focuses on (1) Intelligent dual-recording: using voice, face recognition and other technologies to improve the accuracy and efficiency of insurance; (2) In-process real-time quality inspection: real-time quality inspection and feedback are carried out for each recording step to reduce post-recording repetition and improve efficiency and customer satisfaction; and (3) AI check at backstage where the recorded video is stored and managed in the backstage to reduce risks.

Fig. 4

3.2.1

Mobile intelligent dual-recording

The mobile intelligent dual-recording is chiefly completed by the interaction between agents and policy-holders. The core demand is to be able to smoothly and finely record the dual-recording video, submit the video to backstage for quality inspection and receive feedback on the effectiveness of quality inspection. The development of big data and intelligent technology presents opportunities for the interactive development of natural language dialogue. Voice replaces the text input and it has gradually become an interactive behaviour of users using intelligent software. The Voice User Interface (VUI) also plays a role in the process of dual-recording.

As shown in Figure 5, the interaction interface of dual-recording system uses not only the visual senses of people, also voice introduction, face recognition and behaviour identification technology into the interaction design process of the insured, and also takes advantage of people's multichannel, such as voice, gestures, eye, face, etc., so that users can complete the above-mentioned human identification processes such as ‘mutual verification, product introduction’ more naturally and harmoniously and save the insurance agent from repetitive work. At the same time, intelligent voice playback can provide more standard dual-recording guidance, ensure process and speech standards and reduce the error rate in oral speech.

Fig. 5

3.2.3

AI check and inspection at the backstage

The main users of the backstage include quality inspectors and managers, and the main function is to conduct a quality inspection of the video and personnel management. The insurance agent uploads the recorded video to backstage through APP, and the backstage system uses AI to quickly review and locate the key nodes of the video. In the interface design of quality inspection, a variety of video screening methods are provided to help quality inspectors quickly find video files that need to be re-checked for quick review (as shown in Figure 6). The core function of managers is to manage the accounts of agents and quality inspectors. Face recognition and other technologies are used to collect the basic information of agents, and through a flexible user management interface, the functions of adding, deleting, modifying and checking user accounts are realised.

Fig. 6

The aiCore smart dual-recording requires the algorithm to focus on facial detection, identification, document detection and recognition, document detection and recognition, motion detection, etc. Based on the content of the detection and recognition, smart dual-recording uses a target detection algorithm and OCR recognition algorithm to detect key information that is to be checked in videos. In the process of text detection and recognition, images will be input before detecting, correcting and recognising text. The text recognition framework is shown in Figure 7.

Fig. 7

Target detection focuses on a specific object, requiring that the category and location (classification + localisation) of this target be obtained at the same time. Therefore, the output of the detection model is a list, where each item uses an array to give the category and position of the detected target (the coordinates of the rectangular detection frame are commonly used). There are three steps involved in the target detection algorithm: extraction of image features, generation of candidate regions and their classification. The image feature is the basis of extracting the detection process. The target detection algorithm based on deep learning has two research branches: (1) a two-stage algorithm represented by R-CNN and (2) a one-stage algorithm represented by SSD and YOLO. The smart dual-recording uses the one-stage algorithm. The block diagram of the target detection process is shown in Figure 8:

Fig. 8