uberization of insurance! - .net framework
TRANSCRIPT
Copyright © 2018 HCL Technologies | www.hcltech.com
Uberization of Insurance!D.I.C.E – Digital Insurance Claims Experience platform
2 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
PreludeWhat's Next in Insurance?
The Banking, Finance & Insurance services industry is the widely and wildlyconquered sector where digitization and automation has been enabled to the max by IT.
But in recent times, with the advancement of technology and innovations, it is important forthe Insurance industry to explore, pace up, transform, innovate, re-think and live in thepresent.
The technology adoption and re-incarnation of business strategy/priorities to serve thecustomers will play a crucial role in the Insurance Company’s journey in making it stand talland strong in the crowd (as a differentiator).
Storm of Competition
Technology Shift
Regulatory Compliance
Customer Behavior
Key Drivers pushing the Insurance companies to embark on the Digital Journey
▪ As per FBI stats, the insurance industry in US consists of more than 7,000 companies that collect over $1 trillion in premiumsannually.
▪ Additionally, the total cost of insurance fraud (non-health insurance) is estimated to be more than $40 billion annually.▪ That means Insurance Fraud costs the average U.S. family between $400 and $700 annually in the form of increased
premiums. [1]▪ Insurance Fraud is no longer a victim-less crime, instead it has a cascading impact on all the stakeholders bearing the fraud-
loss▪ “Hence there is a great urge & demand to develop a system that can identify potential frauds with accuracy and arrest it at
the very initial stage for detailed scrutinization”
The M
arket Space
[1] Source: https://www.fbi.gov/stats-services/publications/insurance-fraud
3 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Uberization of Insurance!
D.I.C.E – Digital Insurance Claims Experience platform
- insurance claims risks & raffles to circumvent
4 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Problem StatementTypical Challenges in an Auto Insurance Claims Process
This Photo by Unknown Author is licensed under CC BY-ND
Paper-forms that are manual & error-prone
Multiple Handoffs with limited transparency
Limited Input Channel Support
Data stored in pockets across various legacy systems
Reactive Detection of Fraudulent claims
Speed of Settlement is slow, impacting customer experience
Limited personalization or 1:1 engagement
Increased Operational Expenditure / Cost
5 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
High-level SolutionPhase – 1 Implementation
Omniverse Sarah
Claimant using Web
/Mobile App / WhatsApp /
Voice Devices to converse
with Chatbot
Secured Access
Incl. Biometrics
Sending car accident information
and supporting documents /
proofs over Email
RPA Engine to extract information from Email and connect to Pega BPM Engine to create a
case and attach the necessary artefacts / supporting documents
Claim Database
Photo of Number plate
Claim Status Enquiry
Photo taken by
Personal Device
▪ Car Image taken during Policy Initiation
▪ Other Car Data retired for raising a Claim
Car Image taken after
Accident occurs
Up
load
to
Sara
h
En
ter
Cla
im #
AWS Machine Learning
Raise a accidental auto-insurance claims request via IVRS reaching out to a Call Centre Agent
Call Centre Agent
Claim Status Query
Claim Status Information
Insurance Claims
Fraud Detection
Learning & Training the Model
Batch / Agent
Claim Request Submission & attach supporting images / documents
Number Plate image is OCRed to extract the
Vehicle Information Number
Omniverse
Cognitive
Platform
Custom
Vision AICar Image Comparison
for Accident Damage
Information Update Claims Info with
vehicle damage details
6 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Technology Stack
Omniverse Cognitive Platform (NLP, Chatbot, Multilingual,
OCR, Virtual Assistant -Alexa/Google Assistant)
Robotic Process Automation (Automated email parsing &
processing of claims -leveraging UiPath)
AWS Machine Learning (Sagemaker) Library Integration for Fraud Detection / Prediction
Google Map APIs integration for Location & Street View
(Developer Version)
Pega Infinity (8.2.1) for Insurance Claim Case
Management Implementation & Batch Processing for Training
the Model
WhatsApp Integration for fetching claim status info
(leveraging Twilio platform)
7 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Architecture at a Glance
[Initiate Claim] [Claims Review] [Sys Admin]
User Interface
Process & Workflow
[Claim]
[Admin]
[Delete AWS Model]
[Train/Retrain Model]
Agent / Batch Activity to
Generate CSV &
Train/Retrain Models
Pega RULES DB | Claims Database
Predictive
Models
AWS ML Connector
Shared Custom
Component
▪ Currently Pega does not have a connector for AWS ML (Sagemaker). This will helpcustomers planning to leverage AWS ML connectors for the machine learningcapabilities
▪ It may be available in the Q4 2019 by Pegasystems.▪ It is built leveraging custom libraries and config files to manage models in AWS
Sagemaker and creating/maintaining the same
Omniverse Sarah emailSelf-Service
Agents
Input Channel
Process Layer
Data Layer
8 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
▪ Leveraging Machine Learning & Predictive Analytics capabilities for Fraud Detection overcomes the traditional statistics perspective
▪ Enables Real-Time processing
▪ Process a variety of data to the algorithm without being judgmental around the relevance of the data elements.
▪ Leverage complex algorithms that iterate over large data sets and analyze the patterns in data
▪ The ML model works on a 3 stage cycle Train-Test-Predict.
▪ Optimizing the model by continuously adding data & experience makes predictions more accurate
▪ Self-Learning & Self-Healing capabilities help in automated fraud detection (unsupervised ML)
Train
TestPredict
Act
Fraud
Why Auto-Detection of Fraudulent Claims is important?Proble
m S
tate
ment
Solu
tion
Busi
ness
Benefi
ts
▪ Traditional fraud detection techniques are based on heuristic model & checklist driven fraud indicators
▪ Extensive manual intervention & dependency
▪ Inability to create a persona of the customer based on multiple dependent factors like relationship, segment, geo etc.
▪ No “Single-Go-To-Model” works every time for every customer
▪ Manual recalibration of model on a timely basis is challenging and also requires training plans to address such issues in future`
▪ The fraud detection processing involves multiple verification steps and is time-consuming
▪ Primarily Rules based – helpful in arresting obvious fraudulent scenarios but cannot comprehend hidden & implicit correlations in data
Manual Review is Costly
Time-Consuming for
Review & Training
Class Imbalance
Customer needs Quick
Resolution & TAT
▪ Machine learning can evaluate huge numbers of transactions in real time (in microseconds). It can be extended to adopt
advanced models such as neural networks for better accuracy/prediction & faster processing
▪ Machine Learning prediction quality improves with increased in data sets. Caution required in sanitizing the training data. As an
undetected fraud in Training Set can pose a risk for the system to behave adversely
▪ Unlike humans, machines don’t face problems like fatigue or boredom from routine/monotonous/mundane tasks.
▪ Helpful in detecting hidden and non-intuitive patterns for identifying fraudulent transactions
▪ Reduce cost for manual verification, review, and training the personnel
Reduce Cost
Reduction in TAT for
Fraud Detection
Improve Efficiency
Improve Customer
Experience
Reduce Manual Hand-offs
Improve Speed & Scale
Manual Error-Prone
9 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Fraud Detection Use-Case Details (a functional view-point)
Fraudulent Claims Process (further scrutiny)
Claims ProcessingClaim is Genuine
Claim is Fraudulent
Claims
Fraud Inspector *
Insurance Company
Office Staff
Customer
▪ Existing knowledge base of fraudulent claims data is used
by the AI / ML models
▪ The entire process is self healing in nature (i.e the real data
is used on a periodic intervals to train the model)
▪ Based on the prior knowledge a prediction is made.
▪ Based on the accuracy of the model the process can be
fully automated (unsupervised machine learning)
▪ Involves Investigation process by Experts
▪ Need to involve other investigation parties
▪ Increase in cost and a time-consuming process.
▪ Existing Knowledge base of fraudulent claims data is not
used at all
▪ Primarily driven by Rules - cannot comprehend hidden &
implicit correlations in dataFra
udu
lent C
heck p
ow
ere
d b
y A
I / M
L
Tra
ditio
na
l wa
y o
f fraud
ule
nt c
heck
Experts Driven ApproachData Driven Approach
*The Claims Fraud Inspector has the authority to toggle the flag and make the Machine Learning Model
supervised or un-supervised based on the performance and accuracy of data predicted by the model
Toggle Switch
10 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Business Benefits
Improve Productivity & faster time to market
Reduce Manual hand-offs & overheads with STPs
Reduce Operational Cost (OpEx)
Multi-lingual conversation and voice enabled assistants
Eliminate mundane & monotonous tasks with OCR & RPA
Personalized Customer Experience
Providing a human-face to your brand
Elevation of the Nature of Work
Improve Speed & Scale
Proactive detection of Fraudulent Claims with AI/ML techniques
11 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Next Steps…
Google Vision APIs / ML
Enrich Application
Enhance RPA/Chatbot
Blockchain
Augmented Reality (AR)
Cosmos / Ux Kit
CRM Portal
Google Map View APIs
IoT Integration
• Parsing & Extracting data from DrivingLicense & Car Number Plate
• Inspecting and capturing the Damage spotsin the car (compare with old car image)
• Video analysis leveraging Data Science (MLlibraries) to assist in accident examination
• Street View and other Map Views todetect the accident site locationbased on map navigation
• Identify potential & likeliness of thesuch an incident at the site
• Contact Center Application leveraging PegaCPM for the Agent Portal
• Co-Browse Option• Intelligent Assistants• Next best Action & Recommendation• Integrate via Twilio or other platforms that
provide IVRS integration
• Enrich the customer experience & userexperience of the portal leveraging the latestUx kit provided by the Pega Infinity Platform
• Revamp the external UI where the Chatbot isembedded
• Identify & Predict Driving Behavior basedon data patterns captured via IoT devices
• Calculation Logic for Claim Amount basedon driving style
• Alerts & Notification Trace (sent before theincident if ignored by the driver) e.g.Weather Condition, Car Condition etc.
• RPA for parallel legacy system updatetypically done by an agent putting thecustomer on-hold (mimic legacy systems)
• Create a front-office & back-office RPAenabled process
• Enrich Chatbot Training & capability
• Leverage Augmented Reality to raise Claimsat the site location with a device and AR lens
• Identify potential spots that can be claimed• Can also be used by Site/Car Survey
personnel
• Enhance the Application & making it closerto reality:
o Data Properties | Sections | ClaimsFlow
▪ Data Sanity & Cleanup▪ Fine-tune Prediction algorithm▪ Revamp the end-user portal /agent portal
• Decentralized data storage that can befetched by various parties in the valuechain
• It will help in getting transparency andaccess to real-time status/data by anydiscrete system
12 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Application Snapshots – for reference
13 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Application Landing Page (end-user)
14 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Omniverse – Digital Assistant “Sarah”
15 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Claim Processing – Case Type in Pega
16 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
emailClaimBot: Processing the Claim Case creation based on email trigger
Claimant sends email with claim details,
license, car pic
UiPath Robot reads email, downloads pdf claim form and other
attached images
UiPath Robot parses the request details from the
PDF claim form
Prga creates a claims request and saves the
attachment in the request. Returns the ClaimID back to the
UiPath Robot
UiPath Robot notifies claimant about the claim
request details
UiPath Robot posts the claim data and images to
Pega by triggering a service
17 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
emailClaimBot: Snapshots
Emai
l Sen
t b
y C
ust
om
er
RPA
Mo
del
ler
Age
nt
Das
hb
oar
d
Cla
im R
equ
est
Form
| P
aper
→Sc
ann
ed
18 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Digital Assistant: Snapshots
WhatsApp Claim Status Enquiry
Cre
ate
Cla
im R
equ
est
Up
load
Do
cum
ents
Fetc
h C
laim
ant
Dat
a
Mu
lti-
lingu
al f
eatu
re -
Span
ish
Google Assistant Alexa – Virtual Assistant
19 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Pega Agent Desktop for raising Claims & Underwriting
Age
nt
Das
hb
oar
d
Rai
se C
laim
on
beh
alf
of
Cla
iman
tP
roac
tive
Fra
ud
Det
ecti
on
Stre
et V
iew
fo
r d
etai
l scr
uti
ny
$8.6 BILLION ENTERPRISE | 125,000 IDEAPRENEURS | 41 COUNTRIES
21 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Omniverse – An Intro!
22 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Omniverse Core ConstructTechno Functional Platform Constituents
23 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Omniverse Reference ArchitectureConceptual Building Blocks of the Platform
24 Copyright © 2018 HCL Technologies Limited | www.hcltech.com
Omniverse Feature List
SMART BOT FeaturesMulti BOT Ecosystem
• Coexistence of BOT from multiple Technologies
• Live BOT Registration
• Many-to-Many relationship between BOTs and Brains
Multi-Channel Access• Custom Branded Web Chat
• Skype for Business & Skype
• Microsoft Teams
• SMS & Email
• Social like Facebook Messenger, Telegram, Twitter and Slack
• Digital Assistant like Cortana, Alexa and Google Assistant
Multi-Mode Communication• Textual Communication
• Voice Command
• Touch Commands
• Scribble Feed
• Gestural Input
• Behavioural Input for Proactive Action
Integrated Non-BOT Communication• Device Interfaces like MS Cortana, Google Home, Amazon Echo etc.
• Seamless Transfer to / from Live Agents
• Seamless Transition to / from Interactive Voice Responses
State Handling• Caching for Repetitive Data
• Hold User Session in Context for seamless Cross Channel transition
Cultural Diversity• Automated User Language Identification
• Response in User Language
• On Demand Service based Translation
• Predefined Resource based Translation
• Customizable BOT Personality
Power User driven BOT Management• BOT Parameters customizable by Power Users
• On Demand / On Schedule Activation / Deactivation of BOTs through
Registration Console
BOT to BOT Interaction• Cross Boundary BOT to BOT Communication
• Configuration based Interaction with third party BOTs
Non-Functional Boosters• Caching for Optimized Performance
• Secured Access through various Authentication Modes – LDAP
based, Multi Factor, Biometric integrated
• Itemized Auditing
Speech Capabilities• Speaker Recognition
• Speech Recognition
• Text to Speech and reverse
• Multilingual Speech
• Accent Management
SMART Brain FeaturesMulti Brain Ecosystem
• Coexistence of Brains from multiple Technologies
• Coexistence of Cross Functional Brains
• Massive Scale Out of Brain Ecosystem
Natural Language Expertise• Natural Language Processing, Query and Generation
• Strong Natural Language based Conversation
• Spell Check, Synonyms and Regular Expressions
• Content Moderation
Functional Diversity• Generic Query based Brains
• Personalized Query based Brains
• Personalized Transaction based Brains
• Proactive Brains working on Event or Data Correlation
• General Knowledge based Brains
• Casual Small Talk based Brains
Conversational Diversity• Direct Textual Communication
• Dialog or Prompts
• Hyperlink based Information
• Structured Feedback Processing
• Live Validation
• Embedded Content
• Text with Buttons
• Image Bubbles with Buttons
Controlled Learning
• Audit based identification of New / Unknown Utterances
• Auto loading of shortlisted Utterances to Brain(s)
• Automatic / Semi-Automatic Training
• Accuracy Booster Tools and Techniques
Personalization and Contextualization• Authorization on Use Cases
• Persona and User Property based Filter
• Contextual Hold
• Channel based Filter
• Trend based Filter
• User Sentiment driven Response
• User Vocal Tone Analysis
Power User-driven Brain Management• Consistent Brain Definition across base technology platform
• Live Brain Registration
• UI driven Use Case Management
• Multi-Level editing by Power Users
• Time driven Publishing Process
Reporting & Analytics• Integrated Analytics Engine for any Analytical Display
• Monitoring and Analytics with Power User Dashboard and User
Dashboard
• User Data Analytics
SMART Orchestration FeaturesMulti Recipe Ecosystem
• Cross Boundary Backend System Integration across Technologies
• Cross System Integration where cross-system trust is challenged
• Seamless connection with Brains
Power User-driven Orchestration Management• Recipe Registration
• Recipe Definition Management
• Publishing Ecosystem to control Orchestration availability
Multi-Level Orchestration Organization• Recipe Store holding all Recipes
• Cartridges as Building Blocks of each Recipe
• Cartridges comprises of
• Actions
• Connections
• Sequence
Standardized Integration Approach• All Integration through API Gateway only leading to better
management of Integration Definition
• Supporting any REST interface to be connected
• Keeping the core Integration Responsibility out of boundary