Artificial intelligence-based robotic automation process method and system for independent insurance sales agents

Abstract

Methods and systems for AI-based robotic automation for persuasive referrals are provided. In a novel aspect, a robotic persuasive referral is generated based on a potential customer product-service P_PS matrix, which is generated based on predictive analysis using a DNN model and dynamically obtained feedback. In one embodiment, the DNN model is trained using customer personal profiles for associated product and service PS revenue for each customer dataset. In one embodiment, the predictive analysis uses a decision tree classifier. In one embodiment, a computer system detects one or more predefined trigger events including feedback information for the robotic persuasive referral and one or more predefined life event, updates the P_PS matrix and the robotic persuasive referral accordingly. In one embodiment, the feedback information is sentiment analysis of responses from potential customers. In another embodiment, proximate cause, frequency, and page view analysis are performed based on one or more detected life events.

Description

[0001] This application is a divisional application of Rocco Technology (Beijing) Co., Ltd., filed on October 22, 2021, with application number 202111233531.1 and invention title "AI-based Robotic Process Automation for Independent Insurance Sales Agents". Technical Field

[0002] This invention generally relates to artificial intelligence (AI) automation, and more particularly to AI-based robotic process automation for independent insurance sales agents. Background Technology

[0003] There are three different ways to buy insurance. These include buying directly through the internet, or through a dedicated agent or independent agent. Buying directly through the internet is becoming increasingly popular. In fact, many insurance contracts for similar products, such as car insurance, property insurance, and some health insurance, can be purchased directly online. This is especially true for the younger generation who are accustomed to online shopping. This is particularly true when the insurance product is complex or involves long-term financial investments such as life insurance or retirement insurance. This makes it even more important to cultivate these potential customers by combining a large online digital user experience with one-on-one consultation and persuasion. Therefore, insurance sales agents remain a major driver of acquiring new customers. Dedicated agents work for only one insurance company and are paid by that company. Customers deal directly with the insurance company. On the other hand, independent agents do not work exclusively for one insurance company. They can work for multiple companies, offering a wide range of products. Compared to dedicated agents, they have more freedom to offer plans that are more suitable for their clients. This flexibility makes these independent agents strong competitors in the insurance industry. However, the insurance industry faces the problem of high turnover and a shortage of skilled insurance sales agents. The shortage of skilled insurance sales agents is due to many agents retiring or about to retire. The high turnover rate is due to the low income level of entry-level sales jobs.

[0004] Tools that help independent insurance agents work more effectively and efficiently can attract and retain entry-level sales agents. Unfortunately, insurance software applications are primarily developed for insurance companies. We believe that independent insurance agents can thrive and be rewarded if tools can help them and increase their chances of success. Therefore, we developed a personalized AI-powered Robotic Process Automation (RPA) to support independent insurance sales agents (IISA bots). RPA is fundamentally changing how independent insurance agents operate by automating and optimizing the end-to-end sales process. The IISA bot also provides agents with AI-based decision support to maximize their revenue. Summary of the Invention

[0005] A method and system for AI-based robot automation for persuasive references are provided. In one novel aspect, robot persuasive references are generated based on a potential customer product-service (P_PS) matrix, which is generated based on predictive analytics using a deep neural network (DNN) model and dynamically acquired feedback. A computer system obtains one or more potential customer input datasets for one or more corresponding potential customers, where each potential customer input dataset includes multiple predefined potential customer attributes. Predictive analytics is performed on the one or more potential customer input datasets using a deep neural network (DNN) model, where the DNN model is trained on a previously existing large dataset containing multiple customer datasets. Robot persuasive references identifying one or more matching products and services (PS) for the potential customers are generated based on the P_PS matrix. In one embodiment, the DNN model is trained using a customer profile for the associated product and service (PS) revenue for each customer dataset. In one embodiment, the predictive analytics uses a decision tree classifier to recommend potential customers and potential PSs. In another embodiment, the predictive analytics is also based on a PS knowledge base and one or more agent profiles. In yet another embodiment, one or more potential customer input datasets are augmented datasets, each of which includes one or more related datasets based on one or more predefined relational rules.

[0006] In one embodiment, the computer system detects feedback information for the robot's persuasive reference and one or more predefined trigger events of one or more predefined lifecycle events, updates the P_PS matrix based on the detected predefined trigger events, and updates the robot's persuasive reference based on the updated P_PS matrix. In one embodiment, the feedback information for the robot's persuasive reference is sentiment analysis of responses from potential customers, such as their comments on persuasive content. In another embodiment, the sentiment analysis is based on audio input analysis using a speech sentiment classifier. In yet another embodiment, the feedback information for the robot's persuasive reference is based on public behavior analysis of one or more detected public actions of potential customers. In one embodiment, recency and frequency (RF) analysis is performed based on one or more detected public behaviors and lifecycle events. This analysis provides recommendations for potential customers and their associated products and services.

[0007] Other embodiments and advantages are described in the following detailed description. This overview is not intended to define the invention. The invention is defined by the claims. Attached Figure Description

[0008] The accompanying drawings illustrate embodiments of the invention, wherein the same reference numerals denote the same components.

[0009] Figure 1 An exemplary diagram of the IISA robot system is shown.

[0010] Figure 2 Example diagrams showing an exemplary DNN model and a top-level IISA classifier are provided.

[0011] Figure 3 An exemplary diagram is shown for applying a DNN learning model used for predictive analytics to generate potential customer entries to the P_PS matrix for persuasive reference.

[0012] Figure 4 An exemplary schematic diagram illustrates the detailed process of predictive analysis using a decision tree classifier to generate a PPS matrix.

[0013] Figure 5 An exemplary diagram is shown showing the dynamically updated P_PS matrix based on predefined triggering events, including persuasive references and other lifetime events.

[0014] Figure 6 An exemplary diagram is shown of the sentiment analysis process used to determine readiness.

[0015] Figure 7 An exemplary block diagram is shown of a machine in the form of a computer system that performs AI-based robotic process automation for IISA.

[0016] Figure 8 This shows an exemplary flowchart of IISA's AI-based robotic process automation. Specific Implementation

[0018] Reference will now be made in detail to some embodiments of the invention, examples of which are shown in the accompanying drawings.

[0019] Figure 1An exemplary diagram for a robot IIISA support system is shown. The IISA robot system 100 provides end-to-end assistance based on AI technologies such as deep neural networks (DNNs), decision tree classifiers, sentiment analysis, dynamic feedback, and event-driven updates. The IISA robot system 100 interacts with customers and / or prospects 150, agents 160, products and services (PS) 170, and a network / Internet 110. In one embodiment, one or more network interfaces connect the system to a network. The IISA robot system 100 includes a network interface, a predictive analytics module 120, a matrix module 130, a robot module 140, and a feedback module 150. The network interface connects the IIISA robot system 100 to a network. The input module obtains one or more corresponding prospect input datasets for one or more prospects, where each prospect input dataset includes multiple predefined prospect attributes. The analytics module 120 performs predictive analytics on the one or more prospect input datasets using a deep neural network (DNN) model, where the DNN model is trained on a previously existing large dataset containing multiple customer datasets. Matrix module 130 generates a prospect product-service (P_PS) matrix for each prospect based on predictive analytics and feedback attributes, where feedback attributes are obtained from the corresponding prospect's response to persuasive content. Robot module 150 generates a robot persuasive reference that identifies one or more matching PSs for the prospect based on the prospect product-service (P_PS) matrix. Feedback module 150 detects one or more predefined trigger events, including feedback information to the robot persuasive reference and one or more predefined lifecycle events; updates the P_PS matrix based on the detected predefined trigger events; and updates the robot persuasive reference based on the updated P_PS matrix.

[0020] In one embodiment, predictive analytics 120 uses an ensemble of decision trees 122 and a DNN model 121. In one embodiment, the DNN model 121 is trained using personal profiles for the associated PSs of each customer dataset in the P_PS large dataset 125. The customer large dataset contains a large amount of customer data to train and validate the DNN model 121. In one embodiment, the DNN model 121 is used to discover relationships between customers and products and services (PSs). In other embodiments, the P_PS large dataset 125 also includes a PS large dataset used to train and validate the DNN model 121. In one embodiment, a matrix module 130 maintains a P_PS matrix, which includes potential customer entries 131, agent entries 132, and PS entries 133. Potential customer entries 131 are generated by predictive analytics 120. In one embodiment, the IISA robot system 100 obtains agent entries 132 and PS entries 133 via a network interface. In another embodiment, the IISA robot system 100 obtains agent entries 132, and PS entries 133 are generated by predictive analytics 120. Robot module 140 generates a robot persuasive reference for use by the agent. This feedback is obtained from potential customers. In one embodiment, feedback entry 134 is obtained through a feedback process and included in the P_PS matrix. The robot module generates / updates the robot persuasive reference based on the updated P_PS matrix.

[0021] Figure 2 Example diagrams show an exemplary DNN model and a top-level IIISA classifier. The IISA bot provides end-to-end assistance based on AI technologies such as deep neural networks (DNNs), decision tree classifiers, and sentiment analysis. Process 230 is the DNN, trained using customer demographics and purchasing behavior. The DNN is used to discover potential customers from the agent's family, friends, and fans (FFF) network and what products and services (PSs) they will primarily purchase, and how much revenue these transactions will generate. Process 210 automates personalized persuasion campaigns for potential customers who are focused on the PSs they will primarily purchase. Based on the personalized persuasion campaigns, process 220 predicts potential customers who are now most likely to purchase the suggested PSs. IISA classifier 200 includes a set of potential customer classifiers associated with the predicted potential customer information, such as classifier-1 201 for who, classifier-2 202 for what, classifier-3 203 for how much, and classifier-n 205 for one or more other classifiers. The set of classifiers such as classifiers 201, 202, 203, and 205 is generated by DNN 230. Other classifiers are time classifier-5 207 and classifier-m 208 of one or more other classifiers. The set of classifiers such as classifiers 207 and 208 is generated by processes 210 and / or 220.

[0022] DNN 230 has an input layer 231, a hidden layer 232, and an output layer 233. DNN 230 has multiple inputs, including customer attribute -1 251, customer attribute -2 252, and customer attribute -n 255. DNN 230 has multiple outputs, including output -1 261, output -2 262, and output -m 265. The assumption for developing a DNN model about customer buying behavior is that customers with similar backgrounds will exhibit similar behavior, i.e., they will purchase similar insurance products and services. Figure 2 A supervised deep neural network (DNN) is described to learn patterns in customer purchasing behavior. The input is customer attributes (demographics), and the output is product and service revenue (PS revenue). The training set contains attributes such as the customer's gender, age, income, occupation, education, marital status, homeownership, residence, mortgage, investment profile, parents' age, and parents' residence. If the customer is married, the dataset is expanded to include the same attributes for their spouse, including the same attributes for their parents. If the customer is married and has children, the dataset is further expanded to include the children's data. The DNN is trained by inputting the customer's attributes onto the insurance policies they have purchased. The number of nodes (x) and the number of hidden layers (y) of the DNN are parameters to be fine-tuned during training. A modified linear activation function (ReLU) is used for the hidden layers 232, and a SoftMax activation function is used for the output layers 233. The SoftMax activation function used in the output layers produces the associated probabilities of each product and service (PS) revenue, e.g., prob(PS_1)…prob(PS_n). Given the DNN output, the following predictive metrics will be derived, which can be used to recommend who the customers are, what products and services they are likely to buy, and how much revenue these customers will generate.

[0023] • Expected product and service (PS) revenue per customer:

[0024] • Maximum potential PS revenue that can be generated per customer:

[0025] • Products and services that generate the highest revenue:

[0026] • An ordered list of product and service revenues based on their associated probabilities:

[0027] ROL={ rps_1,rps_2…rps_(n-1),rps_n} where prob(rps_(n-1)) <prob(rps_n and rps_n=PS×prob (PS_n)

[0028] To train and validate the model, a large dataset of customer attributes is used. For example, a collection of purchase histories from customers who have participated in an insurance company for more than five years is used.

[0029] Figure 3 An exemplary diagram is shown for applying a DNN learning model used for predictive analytics to generate potential customer entries to a P_PS matrix for use as a persuasive reference. In step 301, potential customer attributes are obtained for one or more potential customers. These potential customer attributes are input to DNN predictive analytics 310, which outputs potential customer information 311. Output information 311 includes exemplary predictive metrics such as expected product and service (PS) revenue E(Rev), maximum possible PS revenue Max(Rev), products and services that generate the highest revenue argMax(Rev), an ordered list (ROL) of product and service revenues based on their associated probabilities, etc. Output 311, optionally along with one or more other pieces of information including agent profiles 331, commission plans 332, and a PS knowledge base 333, is used as input to predictive analytics 320, which includes a P_PS matrix generator 321. The P_PS matrix generator 321 interacts with a decision tree 322 to generate a P_PS matrix 340. Process 350 generates a robot persuasive reference and obtains feedback on the robot persuasive reference. Process 360 analyzes the feedback and determines if an update is needed. Process 360 also monitors other events, such as the prospect lifetime event 361, to determine if an update is required. If Process 360 determines an update is needed, it sends a message to the P_PS generator 321 for the update. For an independent sales agent, his initial leads are essentially his family, friends, and his social network followers (FFF). As his FFF network grows, his customer base expands. Once the DNN model is trained and validated, the agent gathers attributes of his prospects from his FFF network.

[0030] As an example, in step 301, the attributes of people in the FFF network are collected as input to the deep neural network (DNN) model 310. Process 321 takes into account predictive analytics from the DNN 310, agent profiles (331), commission plans and other monetary bonuses for PSs (332), and a knowledge base (333) about insurance companies and the PSs they offer. Decision tree classifier 322 works in conjunction with 321 to recommend a set of potential customers and a list of their associated PSs in the P_PS matrix 340. Process 350 initiates a persuasion campaign using personalized persuasive content. The persuasion campaign aims to determine the level of interest, attitude, and readiness of potential customers. Upon detecting a triggering event, process 360 updates the P_PS matrix 340 based on the results of the persuasion campaign.

[0031] Figure 4This is an exemplary schematic diagram illustrating the detailed process of predictive analytics using a decision tree classifier to generate a P_PS matrix. The P_PS matrix generator ingests the results of a DNN predictive analytics study and optionally one or more inputs, such as the commission plans and other monetary bonuses of associated PSs, the agent's background and preferences, performance metrics from an insurance company and their PSs from a knowledge base. Company and PS performance metrics include, for example, the PS's sales support (ranking), the insurance company's reputation and history. Metrics from the agent profile include the agent's PS preferences, experience, and knowledge of a particular PS. Summary Table 410 depicts recommendations for potential customers and their potential PSs, including one or more potential customer IDs and other entries such as expected income E(Rev), total income, best PS, and ROL. For example, the agent's best potential customer is potential customer ID #2001, followed by potential customer ID #0500, and so on. The best PS for potential customer #2001 (based on a decision tree classifier with a set of three inputs) is product AIA P1, with a total income of $8,200. Details of potential customer #2001 are shown in Table 440. Table 440 contains relevant predictive analytics from the DNN, along with other inputs such as agent profiles, commissions, PS operational metrics, and insurance company metrics. A more detailed Table 440 for Potential Customer #2001 includes exemplary items such as ROL, PS, probability of success, revenue, E(Rev), commission, total revenue, turnaround time, commission payout terms, sales support, after-sales support, etc.

[0032] In one embodiment, one or more potential customer input datasets are augmented datasets, each augmented dataset comprising one or more related datasets based on one or more predefined relational rules. The augmented datasets become input 435 to a decision tree classifier 431 already trained with a DNN. For example, even if potential customer #2001's second PS, i.e., product AIA P2, generates $10,000, a higher total revenue, it is not considered the agent's best choice when considering other factors. This process is applied to each potential customer to create recommendations. The decision tree results for each potential customer are collected by 435 and processed by yet another decision tree classifier. A first decision tree, 431, is used to rank the potential PSs of the potential customers. A second decision tree, 432, is used to rank the potential customers. Process 420 merges the outputs from these decision tree classifiers, including 431 of table 440 and 432 of the other inputs. For each potential customer, a detailed table similar to 440 will exist. Process 420 creates an interactive panel for the agent, allowing them to delve into a detailed level and change their priorities to suit their needs.

[0033] Figure 5An exemplary diagram illustrates a dynamically updated P_PS matrix based on feedback to a persuasive reference and other predefined triggering events throughout the lifecycle. In one embodiment, the P_PS matrix is ​​dynamically updated. The IISA robotic system detects one or more predefined triggering events, including feedback to the robotic persuasive reference and one or more predefined lifecycle events; updates the P_PS matrix based on one or more detected predefined triggering events; and updates the robotic persuasive reference based on the updated P_PS matrix. The persuasion campaign is fully automated through a collection of software components, including a sentiment analysis engine and a rule-based expert system, to create personalized content. The content can be simple information, touching stories and testimonials, or even a GoFundMe request supporting someone in need who lacks insurance coverage.

[0034] The rule-based expert system 520 creates a robot persuasive reference based on the P_PS matrix 519 and also on knowledge base input 510. The knowledge base 510 includes one or more attributes, including attribute-1 511, attribute-2 512, and attribute-n 515. For example, an insurance knowledge base includes attributes such as property and casualty insurance (e.g., car, homeowner, tenant, earthquake, flood, etc.), health, dental, and vision insurance, life insurance, disability, retirement insurance, pet insurance, etc. Content is created by the expert system 520 based on the potential customer's background (i.e., attributes) and their response to a specific "probing" (i.e., content sent to the potential customer). Process 530 disseminates relevant content, such as insurance-related content, and pushes personalized content to potential customers. The robot persuasive reference is delivered to the corresponding potential customer, and feedback is collected. In one embodiment, process 540 detects triggering events by collecting and analyzing feedback and other events. Triggering events include output from sentiment analysis 551, lifecycle events 552, and other events 555. In one embodiment, the process captures potential customers' behaviors to obtain their responses, such as clicking links, commenting on news, discussing and / or sharing evidence, responding to content sharing of their own stories, or requesting donations to GoFundMe. These responses are analyzed using a sentiment analysis engine 551 to determine potential customers' attitudes and readiness. Process 560 determines potential customers' readiness based on trigger event analysis. If process 560 determines no, then at step 571, the state is updated to the P_PS matrix 519. If process 560 determines yes, then at step 572, recency and frequency (RF) analysis is performed.

[0035] Figure 6An exemplary diagram is shown illustrating the sentiment analysis process used to determine readiness. The sentiment analysis engine monitors changes in potential customers' attitudes and updates the corresponding attributes in the P_PS matrix. Stimuli (persuasive content) 610 serve as independent variables that change a person's attitude. These stimuli are personalized content created by the IISA bot. Stimuli can be delivered via chatbot dialogue 611 and other activities such as face-to-face conversations, video conversations, social media posts, etc. In one embodiment, a response to stimulus 610 is directly identified as attitude 630. In another embodiment, the analysis process 620 assigns a classifier based on responses to one or more stimuli 610. In one embodiment, a stimulus response detector 621 detects stimulus responses as input. A response classifier processes the detected stimulus responses captured by 621 and identifies an input classification for each input. Input classifications include audio clips, video clips, public action responses to social media posts, and text input responses. A sentiment classifier 623 classifies the input responses based on the classification from 622. In one embodiment, the sentiment classifier 623 is a 1D CNN LSTM process of video / audio clips to generate a speech sentiment classifier result. For text input, sentiment classifier 623 is a text analysis engine that generates predefined response classifiers for the responses. Public actions include responding to social media posts, such as clicking, reposting, and forwarding. Sentiment classifier 623 assigns predefined sentiment classifiers based on different public actions. The generated response classifiers are mapped to sentiment classifier 625. Optionally, the sentiment classifiers are generated directly from response classifier 623. In another embodiment (not shown), a 2D CNN LSTM can be used to acquire stimulus responses, such as audio / video clips, and generate sentiment classifiers for feedback. In other embodiments, additional AI-based analysis processes are performed to obtain attitude 630. Attitude 630 sorts the responses and provides feedback information for updates.

[0036] In one embodiment, attitude 630 has three possible outcomes: influence 631, cognition 632, and behavior 633. Influence 631 occurs when a person develops and exhibits a positive attitude towards the insurance policy (PS). Cognition 632 occurs when a person begins to purchase and believes in the PS. Behavior 633 occurs when a person demonstrates public behavior, such as actively seeking out and sharing insurance information (e.g., clicking / forwarding links, attending PS webinars, organizing insurance-related events, or making donation requests related to GoFundMe insurance). When a person demonstrates these types of behavior, they are considering purchasing the insurance PS. A person's readiness to purchase insurance develops from influence to behavior. However, this framework does not assume that there is always a linear progression from influence to behavior.

[0037] Another way to determine a potential customer's readiness is to identify life events. Life events are significant events in a person's life, such as buying a first car, starting or graduating from college, getting a job or promotion, renting an apartment, buying a house, getting a pet, being hospitalized, getting married, having a child, retiring, etc. Life events are essentially milestones in a person's life. Each life milestone presents a unique opportunity to sell insurance (PS). The IISA robot continuously mines the potential customer's social space, for example in process 540, to look for life events. With the help of a sentiment analysis engine, life events will immediately trigger the sales process. When a potential customer exhibits public behavior in response to personalized content or detects a life event, process 572 recency and frequency (RF) analysis will determine when the potential customer is likely to make a purchase.

[0038] In one embodiment, feedback information for bot persuasive references is based on public behavior analysis of one or more detected public actions of potential customers. Recency and frequency (RF) analysis is performed based on one or more detected public behaviors. The three main behavioral responses from collected potential customers are link clicks (page visits), page view time, and link forwarding. RF analysis collects behavioral data from potential customers and categorizes them into two types: recency and frequency. Consider the example RF analysis results for an agent:

[0039] Table 1 RF Analysis

[0040]

[0041] As shown in Table 1, the R-quartiles and F-quartiles are numbered 1 to 4, where 4 = best and 1 = worst. Quartile = 4 refers to the top 25%. In one embodiment, the RF% score is a weighted average. The weights are predetermined based on the Analytic Hierarchy Process (AHP) of domain experts.

[0042] Recency is the recentity of a potential customer's public action. Publicity measures the number of clicks or shares and page view time. Frequency is the total count of public actions. Page view time is the total number of minutes spent viewing the page on the clicked link (note that personalized content is generated for each potential customer per PS identified by the decision tree classifier). Intuitively, a high recency score indicates a strong potential customer interest in a PS. When two potential customers have comparable recency scores, the frequency score becomes the deciding factor. The recency and frequency scores are a weighted average of clicked links, shared links, and page view time. The weights are also predefined by domain experts using the Analytic Hierarchy Process (AHP).

[0043] For all identified lead-PS pairs, the quartiles of each lead's count are calculated, with quartile = 4 representing the top 25% and quartile = 1 representing the bottom 25%. Weighted average scores for the R-quartiles and F-quartiles are calculated using predefined weights from the AHP (Advanced Personal Hierarchy Process) from domain experts. Finally, the weighted averages are normalized to produce the RF% (Free Rate). The RF table is the final panel created by the IISA bot for the agent. In this example, the bot identifies four leads from the agent's FFF (Free Rate) network. The agent should focus on leads 2001 and 2010, who will soon show signs of a potential purchase decision. The RF table also shows the potential spending of these leads, allowing the agent to make intelligent decisions.

[0044] Figure 7An exemplary block diagram is shown of a machine in the form of a computer system performing AI-based robotic process automation (IISA). In one embodiment, the device 700 has a set of instructions that cause the device to perform any one or more methods for speech emotion recognition of interview questions. In another embodiment, the device operates as a standalone device or can be connected to other devices via a network. The device 700, taking the form of a computer system, includes one or more processors 701, main memory 702, and static memory unit 703 that communicates with other components via bus 711. A network interface 712 connects the device 700 to a network 720. The device 700 also includes a user interface and I / O component 713, a controller 731, a driver unit 732, and a signal and / or sensor unit 733. The driver unit 732 includes a machine-readable medium on which one or more sets of instructions and data structures are stored, such as software implementing one or more methods for speech emotion recognition functionality or software utilized by one or more methods for speech emotion recognition functionality. The software may also reside wholly or partially within the main memory 702 and the one or more processors 701 during execution. In one embodiment, one or more processors 701 are configured to obtain one or more potential customer input datasets corresponding to one or more potential customers, wherein each potential customer input dataset includes multiple predefined potential customer attributes; perform predictive analytics on the one or more potential customer input datasets using a deep neural network (DNN) model, wherein the DNN model is trained on a previously existing large dataset containing multiple customer datasets; generate a potential customer product-service (P_PS) matrix based on the predictive analytics and feedback attributes, wherein the feedback attributes are obtained from the responses of the corresponding potential customers; and generate a robot persuasive reference for identifying one or more matching PSs of potential customers based on the potential customer product-service (P_PS) matrix. In one embodiment, the software components running one or more processors 701 run on different networked devices and communicate with each other via predefined network messages. In another embodiment, these functions may be implemented in software, firmware, hardware, or any combination thereof.

[0045] Figure 8An exemplary flowchart of IISA's AI-based robotic process automation is shown. In step 801, the computer system obtains one or more corresponding potential customer input datasets for one or more potential customers, where each potential customer input dataset includes multiple predefined potential customer attributes. In step 802, the computer system performs predictive analytics on the one or more potential customer input datasets using a deep neural network (DNN) model, where the DNN model is trained on a previously existing large dataset containing multiple customer datasets. In step 803, the computer system generates a potential customer product-service (P_PS) matrix based on the predictive analytics and feedback attributes, where the feedback attributes are obtained from the responses of the corresponding potential customers. In step 804, the computer system generates a robotic persuasive reference based on the P_PS matrix, identifying one or more matching PSs for potential customers.

[0046] Although the invention has been described in conjunction with certain specific embodiments for illustrative purposes, the invention is not limited thereto. Therefore, various modifications, adaptations, and combinations of the various features of the described embodiments may be practiced without departing from the scope of the invention as set forth in the claims.

Claims

1. An AI-based robotic automation process method for independent insurance sales agents, comprising: An independent insurance sales agent IISA robotic computer system, having one or more processors coupled to at least one memory unit, obtains one or more potential customer input datasets corresponding to one or more potential customers, wherein each potential customer input dataset includes multiple predefined potential customer attributes; Predictive analysis is performed on the one or more potential customer input datasets using a deep neural network (DNN) model, wherein the DNN model is trained on a previously existing large dataset containing multiple customer datasets; wherein the DNN model of customer purchasing behavior assumes that customers with similar backgrounds will exhibit similar behavior, i.e., purchase similar insurance products and services. The potential customer product-service P_PS matrix is ​​generated based on predictive analysis using the deep neural network (DNN) model and dynamically obtained feedback attributes. Based on the predictive analysis and feedback attributes, a potential customer product-service P_PS matrix is ​​generated, wherein the feedback attributes are obtained from the responses of the corresponding potential customers; Based on the P_PS matrix, a robot persuasive reference is generated to identify one or more matching product and service PSs for the potential customer; The detection includes feedback information to the robot's persuasive reference and one or more predefined triggering events for one or more predefined lifecycle events, wherein the lifecycle events refer to milestone events experienced by a person; The P_PS matrix is ​​updated based on one or more detected predefined trigger events; and The robot persuasion reference is updated based on the updated P_PS matrix.

2. The method according to claim 1, wherein, The DNN model is trained on the associated PS revenue and customer profile for each customer dataset.

3. The method according to claim 1, wherein, The predictive analysis also uses one or more decision tree classifiers.

4. The method according to claim 3, wherein, The predictive analytics are also based on the PS knowledge base and one or more agent profiles.

5. The method according to claim 3, wherein, One or more potential customer input datasets are augmented datasets, each of which includes one or more related datasets based on one or more predefined relational rules.

6. The method according to claim 1, wherein, The feedback information for the robot's persuasive reference is a sentiment analysis of the response from the potential customer.

7. The method according to claim 6, wherein, The sentiment analysis is based on audio input analysis using a sentiment classifier.

8. The method according to claim 6, wherein, The sentiment analysis is based on text input obtained from the potential customers using a set of predefined sentiment classifiers.

9. The method according to claim 1, wherein, The feedback information for the robot's persuasive reference is based on public behavior analysis of one or more detected public actions of the potential customer.

10. The method according to claim 1, wherein, Recency and frequency (RF) analysis is performed based on one or more triggers, said one or more triggers including one or more detected public behaviors and one or more detected lifetime events, said recency being the recentity of a potential customer's performance of said public behavior.

11. An Independent Insurance Sales Agent (IISA) robot system, comprising: One or more network interfaces that connect the system to a network; Memory; as well as One or more processors, said one or more processors coupled to one or more memory units, said one or more processors being configured to: Obtain one or more potential customer input datasets corresponding to one or more potential customers, wherein each potential customer input dataset includes multiple predefined potential customer attributes; Predictive analysis is performed on the one or more potential customer input datasets using a deep neural network (DNN) model, wherein the DNN model is trained on a previously existing large dataset containing multiple customer datasets; The DNN model of customer buying behavior assumes that customers with similar backgrounds will exhibit similar behaviors, i.e., they will purchase similar insurance products and services. The potential customer product-service P_PS matrix is ​​generated based on predictive analysis using the deep neural network (DNN) model and dynamically obtained feedback attributes. Based on the predictive analysis and feedback attributes, a potential customer product-service P_PS matrix is ​​generated, wherein the feedback attributes are obtained from the responses of the corresponding potential customers; Based on the P_PS matrix, generate a robot persuasive reference that identifies one or more matching product and service PSs for the potential customer; The detection includes feedback information to the robot's persuasive reference and one or more predefined triggering events for one or more predefined lifecycle events, wherein the lifecycle events refer to milestone events experienced by a person; The P_PS matrix is ​​updated based on one or more detected predefined trigger events; and The robot persuasion reference is updated based on the updated P_PS matrix.

12. The IISA robot system according to claim 11, wherein, The DNN model is trained on the associated PS revenue and customer profile for each customer dataset.

13. The IISA robot system according to claim 11, wherein, The predictive analysis also uses one or more decision tree classifiers.

14. The IISA robot system according to claim 13, wherein, The predictive analytics are also based on the PS knowledge base and one or more agent profiles.

15. The IISA robot system according to claim 13, wherein, One or more potential customer input datasets are augmented datasets, each of which includes one or more related datasets based on one or more predefined relational rules.

16. The IISA robot system according to claim 11, wherein, The feedback information for the robot's persuasive reference is a sentiment analysis of the response from the potential customer.

17. The IISA robot system according to claim 11, wherein, The feedback information for the robot's persuasive reference is based on public behavior analysis of one or more detected public actions of the potential customer.

18. The IISA robot system according to claim 11, wherein, Recency and frequency (RF) analysis is performed based on one or more triggers, said one or more triggers including one or more detected public behaviors and one or more detected lifetime events, said recency being the recentity of a potential customer's performance of said public behavior.

19. The IISA robot system according to claim 16, wherein, The sentiment analysis is based on audio input analysis using a sentiment classifier.

20. The IISA robot system according to claim 16, wherein, The sentiment analysis is based on text input obtained from the potential customers using a set of predefined sentiment classifiers.

Images