Technology Empowers PAL Ping An Life

Transcription

1 Technology Empowers PAL Ping An Life Shenzhen, China

2 Technological application in PAL has gradually upgraded to 3.0 Traditional technology Mobile internet Artificial inteligence Agent recruitment Online recruitment test Full-process online recruitment management AI recruitment Training Online training management Online distance training AI-based training Agent management Online management of activity plan and performance Full-process online activity record AI assistant online activity management Sales model Electronic proposal MIT online sales MIT mobile-end sales SAT sales Services PC-end online services Mobile-end Jin Guan Jia APP AI customer service AI risk control 2

3 Technology empowers PAL to obtain remarkable achievement in 1.0 and Traditional technology 2.0 Mobile internet Achievement 1 Paper saved 310 mn sheets/year Time for underwriting 5 days shortened to 15 mins 2 Online training 91% Proportion of online services 90% 3 4 Services manpower reduced 70% Productivity improved 32% 6 5 1Total paper saved in preparing electronic proposals and insurance policies 2Time taken from application for insurance, insurance underwriting to completing payment 3Proportion of online learning courses to all courses for 4Proportion of online services to overall services for customers 5Number of manpower reduced under Smart Customer Service model 6Growth of FYP per from 2014 to

4 Multiple business pain points lead to 3.0 solutions Pain-points Solutions 1.Agent recruitment Hard to screen candidates massively AI recruitment 2.Training Hard to tailor-made tutoring AI-based training 3.Agent Management Hard to do real-time management AI assistant 4. Sale model Hard to do precise marketing SAT 5. Services High service cost of manpower AI customer service AI risk control 4

5 1. Agent recruitment: AI recruitment and interview to achieve better screening, resources allocation and career development Accurate screening Efficient financial resources allocation Optimal career path preset On board Terminated 13-month retained coverage ratio 1 95% FNA Qualified Talent scheme High potential NA Low potential Agent High-performance Supervisor 1 Proportion of retained selected by the model to the total retained 5

6 AI recruitment: Design screening process, collect information online based on characters and build model Big data to portray characteristics Select Agents Collect information online throughout the entire process Build model to select Mining historical data Expert experience Agent characters EPASS data Career ambition Training data Customer network + Features of retained + Visit record Adaptability Performance data Sales ability Features of lapsed Design recruitment process Massive information collection EPASS test Reference check AI Work Training interview experience Information collection Personal info Training status Career ambition Prospective s info Behavioral habits Customer service Customer network Retained Resources allocation labeling Optimal career path preset Algorithm 6,400+ layers which improves the node factors of decision trees 10+ hidden layer network 76,800+ neural network Model Integration of multiple algorithms including GBDT, FM, DNN, etc. 40 mn + parameters computing 1,000 + CPU&GPU cluster computing 6

7 AI interview: AI-empowered interviewers conduct real-time online smart interviews, to achieve information collection and confirm willingness Online smart interviews in high concurrency Attract prospective s to interview & decisional interview Housewives Professional sales person Generate questions Collect information Q Question pool Collect information of prospective s Office workers SME owners AI-empowered interviewers Questions feedback A Precision matching Answer pool Expert experience Confirm willingness to join Interactive technology Facial recognition Voice identification Automated Q&A Deep neural network Smart Q&A Emotional understanding engine QA engine Q&A Facial outline identification Voice synthesis Smart guidance Knowledge reasoning engine Intent identification engine Decision engine 7

8 AI recruitment and AI interview have been fully launched in agency channel and have gradually delivered achievements 2019 Target 2018 Fully launched in agency channel Cumulative time length of online AI-interview mn hours/year 13-month retained identification ratio% 2 95% Saved financial resources RMB 630 mn 3 1Estimation of the time length of candidates online interviews 2Proportion of retained s selected by the model to the total retained s 3Total saved costs including office expense, training allowance and trainings 8

9 2. Training: AI-based training to achieve skills improvement and 1 high-performing rapid replication Agent skills improvement High-performing rapid replication benchmark before after A Goal and ambition Agent with resources H Background B Habit G Team environment C Learning ability Agent Potential highperforming Agent with social skills Hardworking F Sales capability D Customer resources Agents with good sales skills E Customer development 1Agents with over RMB 3,000 first-year commission and over two insurance policies in the current month 9

10 Agent skills improvement: Weakness identification, smart allocation of courses, and training results tracking Weakness identification Smart allocation of courses Training results tracking A G H F B D C brenchmark before Course label Course + match + rule Engine allocation Learning results evaluation + Capability assessment E Strength Weakness H. Recognition of life insurance C. Familiar with technology application. E. Customer relationship F. Multi-product recommendation Course 1 Course 2 Course content + Course form + Exam Pass Move to next step Fail Restudy + supervisor mentor Algorithm Model Online course pool 12 Abiility categories scores 44 Ability scores 97 Indicator scores Score card model K-MEANS algorithm Category forecast model 900+ courses Cover all ability categories 10

11 Rapid replication of talent : Big data to build high-performing profile model and design training plan Big data to explore talent characters Talent training plan Mining Historical data Expert experience Visits Working behavior Training course Learning ability + Pocket E usage Customer relations ability Profile of with networks Training plan Profile of diligent Plan A Plan B Profile of socialoriented Profile of s with good sales skill Target Method Time Agent profile model Agent with networks Agent with social skills Diligent Agent with sales skills Algorithm Model Online management system 450+ recoding features 2 layers model Potential label fusion GBDT, LR algorithm integrated model K-MEANS algorithm Label management rules Personalized target set up Smart allocation of tasks Online monitoring 11

12 AI-based training is about to be fully launched and will significantly improve efficiency of nurturing high-performing Shorten the time taken of training highperforming 1 Expand the scale of high-performing Ordinary months 2019 Target 15 months High-performing ,000 20% 2019 Target 459,000 1The average time from on-boarding to first achievement of high-performing 12

13 3. Agent management: AI assistant realizes smart allocation of tasks, and provides online sales assistance Smart allocation of tasks Online sales assistance Insert company plan in advance Business target split + smart schedule plan 1v1 smart mock training 100% participation Managers allocate tasks real-time Generate new schedule based on managers instruction AI assistant Assistance related to work Q&A 7*24 online and answer questions Self-made and customised plan Agent can adjust plan based on individual situations 13

14 AI assistant: Already developed three capabilities: smart task management, smart mock training, and smart Q&A Smart task management Smart mock training Smart Q&A Task A Task B Task type Importance Effective period Source Manmachine practice Manmachine exam + + Ability evaluation Massive information collection + Professional answers Task label Task allocation Task engine Huge script pool Smart allocation engine Question Pool Answer Pool Accurate matching Q&A update Technology support Task engine Text mining Dynamic task split Intention comprehension Decision generation Deep semantic matching model Natural language understanding Label expansion Text processing Knowledge map calculation 14

15 4. Sales model: SAT enables real-time connection, high-frequency interaction, precise marketing and sale process management Real-time connection High-frequency interaction Precise marketing Company Activity Tool Product name Customer classification Importance of lead Customers A S T A Agents Agent News Product Car owners Childcare Insurance Health Recommendation Relationship lead Sales lead Service lead E-Kit Agent sales process management Generation of compliant content Real-time monitoring for communication Behavior evaluation Risk evaluation of Technology support JS-SDK SDK QR CODE OCR Cloud marketing Cloud recognition Fingerprint payment Facial recognition AI smart recommendation Smart allocation Deeplink Big data LBS Voice interaction 15

16 Sales process management: Full-process tracking to reduce behavioral risks of s Generation of compliant contents Real-time monitoring for communication Behavior evaluation Risk evaluation of Expert teams to generate content Communication Repost frequency Repost time Repost target Fake Information communication Customer complaints E Risk High Product Information Centralize content publication Customer voice Credit status Customer visit Consumption record CS call Loan record Online media Asset status Agents Mis-selling D C System review Manual review Zhan E Bao Offline behavior Attendance record Complaint record Training record Zhan E Bao + Wechat + Jin Guan Jia Poor consumption record Low attendance Back-end support B A Low Technology support Sales behavior evaluation model Consensus monitoring system Multi-media interaction Text recognition Cloud recognition Big data Voice recognition 16

17 SAT model had been deeply applied in PAL, with remarkable results Reached customers 1 Interaction times 2 Sales lead 3 conversion ratio 220mn 1.3bn 10.8% 1Number of readers who read the contents forwarded by online 2Number of time users had read the content forwarded by online 3 Proportion of successful conversion of sales lead provided to agency channel 17

18 5. Customer service: 24h-online virtual-human services to achieve precise and efficient service 7*24 online virtual-human service Precise and efficient service Business type identification Smart processing Product recommendation Instant ID identification Instant identification ID + Needs Customers Capture of needs precisely Online document collection RPA Machine 1 processing Online interaction Automatic review Capture of scenario leads Product matching one sentence recommendation One-click solution Inquiries + Process services Precise recommendation Technology support H5 Text interaction Multi-media interaction Biometric identification OCR Image recognition Knowledge map NLP Decision tree RPA robots Scenario pool Timeline Customer profiles Service tips Services + Products Micro facial expression Hospital network NBA Optimal routes Recommendation engine 1Robotic Process Automation 18

19 AI risk control: Big data builds risk factors pool, identifies risks precisely, provides layered strategies and visualizes management Big data builds risk factors pool Identifies and grades risk precisely Visualizes risk management Data mining Basic data Health data + Wealth data Behavior data Layered strategies High Precise Real-time data presentation Expert experience Underwriting rules Claim rules Case features Indicator features Medium Efficient Nation-wide risk map Risk factors Rules factor Blacklist factor Model factor KRI (key risk indicator) factor Low Simple Automatic classification and monitor Algorithm Model 300mn+ data for screening, risk fields 42mn+ historical data 20mn+ case studies Health risk forecast model Customer wealth evaluation model Customer fraud behavior model Related transaction behavior model 19

20 AI customer service and AI risk control have been fully launched and will generate significant impact Fully launched in Target 99% RMB 350mn RMB 2bn RMB 2.8bn Proportion of online AI services Anti-fraud reduces losses Superfast underwriting reduces premium lapsed Service drives sales revenue Proportion of AI-enabled services to online service 2 Total sum of claim rejection of high-risk claiming cases identified by Smart Risk Management System 3 Improvement in (estimated) direct underwriting approval * estimated underwriting premium * 14% reduction in lapse 4 Total sum of sales revenue achieved by successful product recommendation in the service process 20

21 Thank you!