Improving the Way We Ask What You Do? An Enabler of Self-Serve for Commercial Lines Property/Casualty Insurance

Transcription

1 Improving the Way We Ask What You Do? An Enabler of Self-Serve for Commercial Lines Property/Casualty Insurance Prepared by Marc-André Desrosiers, MBA, FCAS, Actuarial Expert Marion Grégoire-Duclos, FCAS, Actuarial Consultant Guillaume Couture-Piché, MSc, R&I Analyst

2 Agenda 1. Who We Are 2. Motivation 3. Applied example 4. Why It s Relevant 5. Current Situation 6. Goal This Week 7. Quantitative Aspects 8. Materials Provided/Available 9. Potential Approaches 10.Support 2

3 Motivation Speciality Lines clients need to feel they can tell us about what they do to move to self-serve Some advisors lack confidence or are over-confident in activity code selection Underwriters spend a material amount of time reviewing IBC code selection 3

4 Who We Are 5

5 Who We Are Business Interruption Property Liability Other 6

6 Applied Example Take a look at the quick quote functionality of one of our competitor: 7

7 Why It s Relevant Importance of Variables 8

8 Why It s Relevant Importance of Variables 9

9 Current Situation Suppose we look for the word food using Excel filters There s an over-abundance of riches that leads to Garbage In Garbage Out, inefficiencies and bad experiences for clients and staff 10

10 Goal This Week Create a Decision-Tree (Taxonomy) Dynamic/Contextual Optimized Create Hazard Flags Risk-relevant Objectively Assessable 11

11 Quantitative Aspects Industry codes Flags 1. Hazard flags Analytical methods are needed to find the most informative flags. Examples of analytical methods: Text mining Web scrapping Important attributes: Predictive of the risks Optimized partition of the clients Partial example of a solution: found flags using analytical methods that describes the industry codes, in an Excel format. 12

12 Quantitative Aspects 2. Decision tree Analytical methods are needed to generate the most efficient tree. [ ] Yes Sell things? No [ ] Important attributes: Optimized splits Adaptable to changes in the portfolio Automatic formulation of questions using flags 5811a10 Yes No 7321a00 Partial example of a solution: using flags found in the previous step, build a flow chart that helps decide which industry code is associated to a certain business. 13

13 Internal documents Our taxonomy: 1. Industry codes (URSG) All our commercial insureds are categorized using these codes. a) Actual code (7 digits) b) Groups and subgroups c) Brief text description of IBC codes a b c 14

14 Internal documents Our expert info: 2. Expert documents (AM Best) These documents give our underwriters crucial information about the risks related to classes of insureds. Clean and structured text Risks classification (1-10) Explanation of risks written by experts Related documents Different taxonomy than URSG 15

15 Potential Approaches Text mining: Word2vec Doc2Vec Words predicting the hazard scores Lemmatization Part-of-speech tagging Concept mining Network analysis: Community detection Centrality measures Clustering: K-means Hierarchical Nearest neighbor Dimension reduction 16

16 Support 17

17 Appendix 18

18 Materials Provided/Available Internal documents: URSG Groups and subgroups Brief text description of IBC codes AM Best documents Risks classification (1-10) Explanation of risks written by experts Clean Related documents Websites worthy of web scraping: US Bureau of Labor Statistics Occupation title by NAICS siccode.com Description Cross references Examples naics.com Synonyms Number of US Businesses Tables of links: IBC split codes to AM Best codes IBC to NAICS (North American Industry Classification System) IBC to SIC (Standard Industrial Classification) 19