application of predictive analytics in distribution, underwriting …€¦ · distribution,...

14
SOA Predictive Analytics Seminar – Taiwan 31 Aug. 2018 | Taipei, Taiwan Session 2 Application of Predictive Analytics in Distribution, Underwriting and Claim Management James Lin

Upload: others

Post on 28-May-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

SOA Predictive Analytics Seminar – Taiwan 31 Aug. 2018 | Taipei, Taiwan

Session 2

Application of Predictive Analytics in Distribution, Underwriting and Claim

Management

James Lin

Page 2: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

SOA Predictive Analytics SeminarAugust 2018

Make it WisdomApplication of Predictive Analytics in Distribution, Underwriting and Claim Management

JAMES LIN

Ernst & YoungAugust 2018

Page 3: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

AgendaSection

Distribution Analytics1

2

7

15

What is Going On0 4

Other Applications 

3 22Getting Started

What Is Going On

0

Page 4: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Analytics and Technologies are Changing the Way Insurance Operates 

5

CUSTOMER HAS EVOLVED 

Finger-tip safe process

Easy claim process

Better price

Products meet needs

INDUSTRY IS CATCHING UP 

Mobile Pay/transfer

AI Chat-box

Online risk assessmentAuto product recommendation

Customer management

Mobile payment/transfer

Customer engagement

Big Data ImplementationComputer

Vision

Natural language processing

THE WORLD IS CHANGING

Digitalize

IoT

Artificial Intelligence

Underwriting

Quality assurance

Deep learning

Apply Analytics across Insurance Value Chain

6

Agency Compensation

Agent Behavior and Retention

Group Life Distribution

Intelligent Lead Generation

Up‐sell &Cross‐sell 

Auto‐Underwriting for Life and Medical Insurance 

Health Scoring – EY Digital Life X

Risk Based Pricing

Advanced Member Balance Analytics

Customer Retention

Customer Lapse Study

Inquiries & Complaints Study

Fraud Detection

Anti‐money Laundry Scoring 

Transaction Monitoring

Claims management and claim trends

Analysis of financial movements

IFRS 9 Impairment Modelling

Credit Investment Modelling and Management

Tableau Reports

Analytics

Distribution Underwriting Customer Transaction & Claims Risk & Reporting

Page 5: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Distribution AnalyticsStory behind production, case count and active ratio   

1

Improve Distribution Process through Advanced Distribution Analytics (ADA)

8

Issues

“We don’t know which type of agents to recruit is most effective given product strategy, target client base and market competition” 

Solution

Find the right type of agent and offer them the most relevant and attractive offer

Agent Acquisition

“We want to understand the unique factors drives sales in the market and how to help our agents to be successful in their career” 

Identify qualitative and quantitative insight as well as provide modern IT platform to enhance customer experience and efficiency

Enhance productivity

“How can we optimize bottom‐line performance at same time offer the most optimal compensation to incentivize our agents” 

Optimize agent structure and provide market comparable non‐linear resource allocation and compensation program 

Business Optimization $

“In our efforts to retain agents we are not always taking into account sales quality, ability to influence and value of the sales” 

Take proactive actions to retain valuable agents and extend the agent’s lifetime value

Agent Retention

Page 6: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Do we know who the top sellers are and why?

9

Top Producers

Who are they?

Significant amount of sales comes from the few top agents ‐ 3% of top producers contribute more than 20% of total FYC.

The productivity gap between top producer and low producing agents is large – only 18% of total FYC is from 71% of low producing agents.

Distribution of Manpower, FYC and Bonus by productivity

What do they sell? 

Higher producing agents sells more Long term life policies, as opposed to Accident & Health or MPF. 

Linked and Universal policies are much more prominent in top agents’ product mixes, contributing around 20% of their FYC.

Product mix by personal FYC level  

How productive?

In most instances, single case sizes were not the main driver – Many top agents have modest average case FYC, but high number of cases and consistent performance.

Group with the highest performance as a whole are junior level managers with long service.

Compensation effectiveness on production

Correlation between productivity and service length 

Can we understand the impact of compensation and structure? 

10

63%9%

28%

FYC  from Leavers

FYC from joint leaver (Producer)FYC from joint leaver (Manager)FYC from sole leaver

Adviser Departure Rate by previous Year FYC and 

Production Concentration

Seasoned agents (denoted by yellow dots), in contrast to new joiners (denoted by grey dots), cluster above certain thresholds, providing significant evidence of agents tuning their performance to meet specific compensation requirements.

Presence of extreme low performers suggests inefficiencies in monitoring scheme.

High productivity agents with concentrated production have highest probability to leave (indicated by (dark) red).

Significant whole team departures to competitors are observed while new joiners mostly come solely.

Productivity loss with manager departure can be notablyaggravated through whole team departures. 

Compensation Optimization Leaver Profile

Distribution of production bonus and Simulation of payment under peers’ compensation scheme 

Comparison of FYC from sole leavers and joint leavers

Page 7: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

And who is performing…. And who is leaving?

11

Production concentration by number of good performance years in the last three years

Agency structure benchmarking Productivity distribution by layers of mangers above

Spread of production over the year likely leads to consistently good performance across different years ‐ only a few high producing agents with good performance in three consecutive years have highly concentrated production (production spread less than 2 months), which is much more common case with occasional good performer. 

From benchmarking, the Company currently has more managerial layers than its peers do

There does not appear to be significant differences in average productivity between those agents in a taller tree compared with a more branched‐out situation, indicating additional managerial layers does not lead to more production 

Consistent Performance Agency Structure

Why are top agents top agents ?

12

Performance analysis: quantity drives performance – not pricing

Source: EY

LowPRICING

Average value per policy

QUANTITY VS PRICING MATRIX FOR ALL TIERS OF AGENTS

Low

High

Tier 1

Tier 2

Tier 3

Tier 4

Sales performance is achieved by higher quantity, not by higher pricing

Top performers are effective in: Generating leads Converting clients

Need for comprehensive lead generation & management strategy to help lower performing agents achieve more

High

QUANTITY# of policies per agent per 

annum

SAMPLE

Page 8: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

What is the geographic profile for an area ?

13

Coverage analysis: recruitment targets need realignment towards market potential SAMPLE

Coverage

Market Attractiveness

Region 4

Region 7Region 6Region 1Region 2Region 3Region 5

Bubble size: Actual sales

Low

HighUndercovered

Overcovered

P23

P22

P21

P20

P13

P12

P11P10

P9

P8

P49

P48

P47

P16

P46

P44P45

P19

P18

P17

P15

P14

P7

P6

P5

P54P55

P53

P52

P51

P50

P37

P36P35

P34

P33

P32

P59 P58

P57

P56

P43

P42

P41P40

P39

P38

P4

P3P2

P1

P25

P24

P31

P30

P29

P28

P27

P26

Need to drive recruitmentin these provinces

Note: P = Province; Market attractiveness is defined as the average of scores scaled to 10 for population, population density and population growth. Coverage is # of agents per million people. 

(agents per million inhabitants)Low High

Source: EY

Other ApplicationsAuto Underwriting and Claim Management

2

Page 9: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Key Pain Points under traditional Underwriting Process

15

High Risk Customers

High Operation Cost

Low Sum Assured

“We don’t know precisely who are the high risk customers, when we are looking at an insurance application file”

“Who will claim earlier than others?” 

“We found high proportion of customers requiring medicals are healthy and eligible for  standard premiums”

“We are spending significant resources on Medical checkups.”

“Medical check in some regions is a lengthy processes. Some customers deliberately apply for lower sums assured under relevant thresholds”

“Medical check, manual process, etc. are slowing down our business.”

“Are traditional high risk groups, such as older lives as risky as we envisaged? Can higher acceptance rates be achieved while controlling risk?

Low Business Efficiency

3

2

4

5

1

How models help to find high risk applicants  

16

Single Variable Analysis 

Variable Derivation

Variable selection and interaction

Model Selection

Model Validation

Product

Cover Term

Sum assured

Age & Gender

Annual Income

Marital Status

Indicator of New Customer

保单数 拒保/延期率 保单数 拒保/延期率 保单数 拒保/延期率

<16岁 635                  0.32% 8                       30.77% 643                  0.32%

16‐35岁 360                  0.16% 277                  6.91% 637                  0.28%

36‐40岁 226                  0.23% 185                  9.57% 411                  0.41%

41‐45岁 363                  0.34% 508                  11.12% 871                  0.79%

46‐50岁 468                  0.55% 788                  12.16% 1,256               1.37%

51‐55岁 1,438               6.45% 1,752               15.44% 3,190               9.49%

≥56岁 350                  11.66% 386                  16.06% 736                  13.62%

总计 3,840               0.52% 3,904               12.69% 7,744              1.01%

非体检件 体检件 总计

保单数 拒保/延期率 保单数 拒保/延期率 保单数 拒保/延期率

<16岁 635                  0.32% 8                       30.77% 643                  0.32%

16‐35岁 360                  0.16% 277                  6.91% 637                  0.28%

36‐40岁 226                  0.23% 185                  9.57% 411                  0.41%

41‐45岁 363                  0.34% 508                  11.12% 871                  0.79%

46‐50岁 468                  0.55% 788                  12.16% 1,256               1.37%

51‐55岁 1,438               6.45% 1,752               15.44% 3,190               9.49%

≥56岁 350                  11.66% 386                  16.06% 736                  13.62%

总计 3,840               0.52% 3,904               12.69% 7,744              1.01%

非体检件 体检件 总计

保单数 高风险率 保单数 高风险率 保单数 高风险率

<16岁 743                  0.37% 8                       30.77% 751                  0.37%

16‐35岁 564                  0.25% 284                  7.09% 848                  0.37%

36‐40岁 449                  0.46% 188                  9.72% 637                  0.64%

41‐45岁 740                  0.70% 521                  11.40% 1,261               1.14%

46‐50岁 963                  1.13% 805                  12.43% 1,768               1.93%

51‐55岁 1,589               7.13% 1,800               15.86% 3,389               10.08%

≥56岁 373                  12.43% 392                  16.31% 765                  14.15%

总计 5,421               0.73% 3,998               13.00% 9,419              1.22%

非体检件 体检件 总计

保单数 高风险率 保单数 高风险率 保单数 高风险率

<16岁 743                  0.37% 8                       30.77% 751                  0.37%

16‐35岁 564                  0.25% 284                  7.09% 848                  0.37%

36‐40岁 449                  0.46% 188                  9.72% 637                  0.64%

41‐45岁 740                  0.70% 521                  11.40% 1,261               1.14%

46‐50岁 963                  1.13% 805                  12.43% 1,768               1.93%

51‐55岁 1,589               7.13% 1,800               15.86% 3,389               10.08%

≥56岁 373                  12.43% 392                  16.31% 765                  14.15%

总计 5,421               0.73% 3,998               13.00% 9,419              1.22%

非体检件 体检件 总计

Decision Tree 

Neural Network

SVM

Random Forest

GLM (General Linear Model)

GAM (General Additive Model)

OOT

AUC

Gain Chart

PSI

Page 10: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Auto‐Underwriting Process in Practice 

Exceed thresholds for mandatory underwritingrelevant health disclosures in application

Unhealthy condition in health notice 

Previous application outcome

Regulation Requirement

InsuranceAapplication

Simple Thresholds

Scoring

Random Selection for underwriting(For monitoring and QA)

Manual Underwriting

Accepted at standard rates

Accepted with special conditions / loadings

Declined cover

Underwriters can use the risk scores to help judge how closely to 

examine a case

Predictive Model for early non‐accident claim

Predictive Model for Underwriting decline 

Predictive Model for sub‐standard identification 

Predictive Model generates Scoring

The Questions We Will Ask When Deal With A Claim

18

“Likelihood of acceptance” 

“Estimation of claim amount”

“Pay in partial, standard or more?”

“Rejection reason and involvement of investigator”

“Customer experience”

Claim Analytics

Page 11: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Predictive Models To Help Us Answer These Questions 

19

Model #1: “Am I likely to pay?”Binary classification

The claims at FNOL(before claims 

management/decision)

RejectClaims that we will likely refuse ultimately, or close due to lack of client’s response

ApprovalClaims that we will 

likely pay to the client

Model #2a: “Should I pay all of it? Am I likely to pay more than expected?”Multi‐class classification

Model #2b: “What is the likely refusal reason, or expertise I should involve? ”Multi‐class classification

Claims without client’s follow‐up

Claims likely involving medical expertise / medical refusal reason

Claims likely involving administrative controls / administrative refusal reasons

?

Regular claim paid partially

Regular claim paid on the expected/ requested amount

Severe claim paid exceeding usual scope

Other categories 

%

Model #3a: “What partial ratio of it am I likely to pay?”Percent‐based regression

Model #3b: “How much more am I likely to have to pay?”Amount‐based regression

Model #3c: “Is the insured person likely to file a complaint?”Binary classification

Claims Analytics Drives All‐Round KPI Enhancement 

20

How does Claims Analytics improve your KPI? Delivering tangible results

>5%Margin improvement

4‐8 Weeks to develop business case

2‐4Months to deploy a project

6‐12 Months payback

Page 12: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Getting Started

3

Begins with Asking the Right Questions

22

Perform

 an

alytics

Execution & 

Monitoring

Gather internal and market data Conduct agent survey Apply data analytic tools Visualize analysis outcomes

Resources and matching skillset Design feasible options Scenario and what‐if analysis  ...

Integrated multi‐channel execution Production pool targeting  Agent value management Contact strategies and plans

Prosperity modelling Real‐time decision‐making Retention progress …

Market Insights

What our peers are doing Key trends & Industry performance What works and what doesn’t

Vision & Strategy

What is our target growth Market management How to manage agents Retention strategy Product Strategy

Company Concerns

Compensation cost is too much Agents are not fully motivated by 

current financing scheme Managerial structure is not cost‐

efficient

Questions to be solvedA

sk the righ

t questions

Top agents Product mix Retention  Agency structure

Compensation effectiveness Agent motivation strategies Digitalization  …

To be solved via performing analytics and reshaping strategy

1

2

3

Page 13: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Key Steps in Your Analytics Journey Ahead 

23

Gather Internal Data

Obtain Survey Data and Market Intelligence

1

2

Generate actionable business insights

4

Conduct Data Analysis

3

Data Request ListFinancial and 

behavioral impact assessment

Proposed implementation timeline

Compensation simulation by component 

Market intelligence from industry SMEs

Detailed diagnostics in productivity

High level positioning maps

Interactive dashboards

Key insights from applying statistical modelling and visualization tools

Prioritize and Implement

5

Implementation – make the analytics REAL

24

Interactive data visualization gives you an insight into what’s going on

Data visualization can provides users with a self‐service analytical capability to explore data and answer questions quickly in a intuitive, drag and drop manner, whilst working across multiple platforms and devices. 

What are the Benefits

Quick Turn 

Around

Opportunity to save costs or improve processes through better understanding and insight of data 

Queries multiple data sources without writing code and transform these into interactive graphical visualisations and dashboards.

Real time validation of hypotheses. 

Quick addition of new key metrics to reporting.

Inspect trends and outliers and discover patterns that would not be practical using traditional methods.

Better understanding of data and trends across all business users.

Visualization of Complex Data Sets

Better Informed Decisions 

and Outcomes

Higher Awareness

Insight Discovery

Page 14: Application of Predictive Analytics in Distribution, Underwriting …€¦ · Distribution, Underwriting and Claim Management . James Lin . 9/12/2018 SOA Predictive Analytics Seminar

9/12/2018

Lets explore how visualization can help …

25