Recommended
More Related Content
What's hot
What's hot (17)
Viewers also liked
Viewers also liked (9)
Similar to Effective Commercial Underwriting using Big Data and Risk Analytics
Similar to Effective Commercial Underwriting using Big Data and Risk Analytics (20)
Recently uploaded
Recently uploaded (20)
Effective Commercial Underwriting using Big Data and Risk Analytics
- 1. Lakshan De Silva Lakshan.desilva@intellectdesign.com Effective Commercial Lines Underwriting using Big Data and Risk Analytics
- 2. WHO IS THE WRITER? WHERE IS THE DATA FROM? What is so unusual about this? CLUE: Is it not in the content 2
- 3. A quality first quarter earnings announcement could push shares of H.J. Heinz (HNZ) to a new 52-week high as the price is just 49 cents off the milestone heading into the company’s earnings release on Wednesday, August 29, 2012. The Wall Street consensus is 80 cents per share, up 2.6% from a year ago when H.J reported earnings of 78 cents per share. The consensus estimate remains unchanged over the past month, but it has decreased from three months ago when it was 82 cents. Analysts are expecting earnings of $3.52 per share for the fiscal year. Analysts project revenue to fall 0.3% year-over-year to $2.84 billion for the quarter, after being $2.85 billion a year ago. For the year, revenue is projected to roll in at $11.82 billion. The company has seen steady earnings for the last eight quarters, but income has been sliding year- over-year by an average of 7.3% over the last four quarters. The biggest drop came in the most recent quarter, when profit dipped by 21.7%. Most analysts think investors should stand pat on H.J, with eight of 16 analysts rating it hold. Analysts have grown increasingly optimistic about the stock in the last three months. The stock price has increased from $53.59 on May 29, 2012 to $55.51 over the past quarter. " Source: http://www.forbes.com/sites/narrativescience/2012/08/24/forbes-earnings-preview-h-j-heinz-3/ 3
- 4. WRITER=MACHINE, DATA=PUBLIC • The writer is a specialized Natural Language Generation Software • The data is publically available • It was “written” over 2.5 years ago! - Natural Language Generation has progressed since then. 4
- 5. AGENDA 1. Concepts – “Big Data” approach 2. Processes – Comparing traditional to Big Data 3. People – Getting the right skills and experience 4. Technology – What are the choices and what to consider? 5
- 7. *Includes machine learning and deep learning Descriptive Analytics What is happening Confirm Hindsight Analytical Technique Outcome Process Traditional Approach Form Theory Observe Data Diagnostic Analytics Why did it happen Predictive Analytics What is likely to happen Form Theory Foresight Big Data Approach Observe Data *Identify Pattern Prescriptive Analytics What should I do about it 7
- 8. BIG DATA APPROACH IS ADDITIVE TO TRADITIONAL APPROACHES 8
- 9. Traditional Approach Big Data Approach Volume Fewer sources – just store everything we need to make the decisions of today Large number of sources – store everything for the questions and decisions you are yet to think of Velocity Managed changes done infrequently Changes done constantly – crowd sourced Variety Structured – internal/external Structured – internal/external, Unstructured – external Veracity Source Level – e.g. this is the “source of truth” for FTE, perils, crime etc. Data Level – e.g. number of sources of FTE data with certainty scores for each Feedback Manual cleansing from fewer people – scholarly “encyclopedia approach” Crowd sourced or via machine learning algorithms – “Wikipedia Approach” Visualizing Graphs, tables, one way reports Alerts, graphs, interactive applications Examples Policies, Quote, Claims, Billing RiskMeter, D&B, Pitney Bowes, Lexis Nexis, Verisk All of traditional data + LinkedIn company profile, Facebook, Twitter, Yahoo BOSS, Google Search, Urbanspoon, Yelp, Glassdoor 9
- 11. Traditional Approach • Subjective • Non-auditable, paper based • Time consuming • Ineffective • No learning Big Data Approach • Objective • Auditable, electronic • Fast • Effective • Learning Underwriters manually enter multiple queries to search dozens of sources to begin the diligence process 30-60 min 30-60 min 16 min 30 min 20 min 20 min OFAC PEP RCA Media Internet SEARCH AGGREGATE SELECT REVIEW RATE QUOTE Low Value High Value Underwriters compile content associated with the subject of investigation Underwriters subjectively select which associated information to review, articles to read, etc Underwriters read articles and learn about the subject of investigation, Identifying potential risk associated with the subject Underwriters designate certain Information / findings as risky based on their review and draw risk-based conclusion about the subject Underwriters report on their findings using document templates and boiler plate language Culling internet search results Misspelled names Subjective decision making necessarily results in incomplete reviews >70% of Underwriting Time <30% of Underwriting Time 30 min 10 min <1 min <5% of UnderwritingTime <95% of UnderwritingTime 10 min SEARCH AGGREGATE SELECT REVIEW RATE QUOTE Underwriters designate certain information/ findings as risky based on their review and draw risk-based conclusion about the subject Underwriters read articles and learn about the subject of investigation, identifying potential risk associated with the subject OFAC PEP RCA Media Internet Culling internet search results Misspelled names Subjective decision Making necessarily results in incomplete reviews Low Value High Value Configurable report Generation and Integration 11
- 12. TURNING UNDERWRITING TO BE BOTH DEDUCTIVE + INDUCTIVE Underwriters usually do not have the time or do not see value in re-checking assumptions 12
- 13. SEARCH AGGREGATE SELECT REVIEW RATE QUOTE Media Structured Data Unstructured Data internet Universeofavailableinformation Red flags missed due to human limitations 13
- 15. Traditional Approach Big Data Approach Hypothesis If I get a view of all the drivers, all the trucks and all the equipment hauled, I will have the right information to underwrite and eventually rate. Fatigue is the leading cause of fatal crashes for long haul trucks. Vibrations transmitted from engine beneath the seat is the main reason for this. Preventing fatigue is the key to a lower loss ratio and improved driver safety. Leading Questions 1. Have you ever been declined or had your insurance coverage cancelled or non- renewed in the past three years? 2. Is there a vehicle maintenance program in operation? 3. Does the applicant obtain MVR? 1. Have you ever been declined or had your insurance coverage cancelled or non-renewed in the past three years? 2. Are all trucks fitted with seat dampeners? 3. Does the applicant obtain MVR verification on drivers? Outcomes Dismiss good risk and missed chance to institute loss prevention programs Uncover good risk and drive good risk management within clients A lot of manufacturers publish the vibration data for the type of trucks. With big data search you can actually get this once you know the vehicle schedule!!! 15
- 16. LOOK TO OTHER INDUSTRIES TO FIND THE RIGHT SKILLS 16
- 17. Traditional Approach Big Data Approach Skills Actuarial, Business Analysis, Accounting, Defining Data, Defining Data Relationships Natural Language Processing, Artificial Intelligence, Software Engineering, Search, Experimenting with Data Experience Common to find with 10+ years in insurance Rare to find professional with more than 3 years of experience 17 BI / EDW Professional Data Scientist
- 18. BIG DATA TOOLS ARE MATURING - THE TIME IS RIGHT TO EXPERIMENT Tools that were traditionally IT related are now available to mainstream business users – remember making webpage used to be an IT job! 18
- 19. Traditional Approach Big Data Approach Statistical Packages R, Julia, Matlab Is your team based on a statistical background or IT/data science background? Databases Hbase, Cassandra, MongoDB, Google BigTable Are you looking at transaction (Billing) vs. documents (Underwriting) Do you want it real time? Do want it in-premise or in the cloud? Do you want to pay licenses or maintenance? Languages* Pig, Hive Is your team familiar with SQL (working with databases) or writing procedural (Excel Macros)? Visualization Pentaho, Tableau, Google Do want it in-premise or in the Cloud? Do you have a insurance DWH? How structured is your data? Do you want to geo-overlays? * Assume Hadoop/ HDFS has been installed 19
- 22. KEY LEARNINGS • Big Data means re-thinking your approach to data processing - You may only know about a part of the problem you will solve up front - The data may show additional problems and answers • Re-look at how and when data is used in the underwriting process - With Big Data tools you can speed up and uncover additional risks • Are you asking the right underwriting questions? - Ask less and more pertinent questions and pre-fill information • Look outside of insurance to build your Big Data team • The technology is evolving rapidly – start experimenting early before the competition gets far too ahead 22
- 23. © 2015, Intellect Design Arena Limited. All rights reserved. These materials are confidential and proprietary to Intellect and no part of these materials should be reproduced, published in any form by any means, electronic or mechanical including photocopy or any information storage or retrieval system nor should the materials be disclosed to third parties without the express written authorization of Intellect Design Arena Limited.
- 24. © 2015, Intellect Design Arena Limited. All rights reserved. These materials are confidential and proprietary to Intellect and no part of these materials should be reproduced, published in any form by any means, electronic or mechanical including photocopy or any information storage or retrieval system nor should the materials be disclosed to third parties without the express written authorization of Intellect Design Arena Limited.