Copyright 2019 Daisy Intelligence Corporation All rights reserved

The story of AI and how Daisy fits in Anomalies and the Theory of

About Daisy Intelligence • Delivers AI-powered solutions to retail and insurance sectors; 15 years

Awards and recognition for Daisy Cool Vendor in AI for Retail by Gartner Toronto

What is artificial intelligence (AI)? Artificial intelligence (AI, also machine intelligence, MI) is intelligence

The difference between predictive analytics and reinforcement learning. Copyright © 2019 Daisy Intelligence Corporation.

Predictive modelling • It has a structural false positive issue built in • Artificial

Predictive analytics is less than 10% accurate for rare events • 1, 000 false

The challenge: fraud/risk is getting more pervasive/growing. • Traditional surveillance approaches rely on rules

Daisy in Insurance – Theory of Risk™ Daisy’s Risk Management Platform helps insurance companies

Reinforcement Learning Engine Determining Anomalies and Outliers Daisy’s Theory of Risk™ Copyright © 2019

Five analytic techniques Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private &

Rules Basic Rules with Fuzzy Scoring • Standard If-Then-Else rules which are either True

Probabilistic Rules • IF PC LIKE '42%' AND WITHIN 1 YEAR OF SERVICE AND

Network analysis: connecting individuals with common attributes Copyright © 2019 Daisy Intelligence Corporation. All

Computing requirements Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential.

Example: finding non-obvious matching individuals using A. I. Company: ABC Current Personal Info Cate

Example: using A. I. to find fraudulent root canal claims submitted by a group

Scoring Curve Types Growth (high is suspicious) Decline (low is suspicious) • Scoring curves

How disruptive is AI in insurance? • Unprecedented improvement in risk operations: − Reduce

Case study: Impact of Daisy fraud detection solution False positive rate, labour savings and

How can insurance companies use Daisy’s AI solutions? Transaction Surveillance / Fraud Detection Identify

Investigative capabilities must keep pace with detection otherwise there will be growing backlog of

Why the Time for AI is Now 1. AI is accessible and delivering results

Slides: 25

Download presentation

The story of AI and how Daisy fits in Anomalies and the Theory of Risk™ in Insurance Agenda Results and client success stories Open Discussion Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. |

About Daisy Intelligence • Delivers AI-powered solutions to retail and insurance sectors; 15 years of experience working with organizations in Canada, US, Europe and Asia-Pacific. • Founded by Gary Saarenvirta, who has an M. A. Sc. In Aerospace Engineering from the University of Toronto. • 15+ years assisting companies to make smarter and more profitable operating decisions using math and science delivering 100’s millions in bottom-line results. • We’re a team of 60+ computational scientists and math geeks based in Toronto, Canada. • $10 million Series A funding completed in September 2019. Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. |

Awards and recognition for Daisy Cool Vendor in AI for Retail by Gartner Toronto artificial intelligence software-as-aservice company is one of three vendors designated for 2018. ROB Fastest Growing Company Daisy ranked as Canada’s 39 th fastestgrowing company based on three-year revenue growth of 1, 311% Best AI Startup The Alconics is the world’s only independently-judged awards celebrating the drive, innovation and hard work in the international A. I. community. Daisy wins pitch prize at Elevate The Elevate. R competition allowed 16 startups to pitch in front of judges to receive funding from Espresso Capital. UK – Top 10 AI companies to Watch One of Canada's Leading Tech Companies Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. Best AI Company in Canada 2019 AI done right. | 4

What is artificial intelligence (AI)? Artificial intelligence (AI, also machine intelligence, MI) is intelligence demonstrated by machines, in contrast to the natural intelligence (NI) displayed by humans and other animals. - Wikipedia • Makes decisions without human intervention (i. e. autonomous). • Simulates trillions of alternative decisions to make the best decision and business outcomes. (i. e. reinforcement learning with simulation). • Can recommend decisions that have never been made before. (i. e. does not require labelled training data). • Ability to “sacrifice” to look for longer term gains. (i. e. delayed / non-greedy rewards). • Ability to “learn” and self-adjust its decision-making policy without any human intervention (i. e. machine intelligence). Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 5

Predictive modelling • It has a structural false positive issue built in • Artificial data sets used to train the models exacerbate the problem • Requires a very good screening method to get anywhere close to “lab” accuracy • Is good at rank ordering not record by record prediction / classification − Perfect for direct mail…. segmentation … − Disease (NO) − Fraud (NO) − Individual Risk pricing (NO) − Risk segmentation- based pricing (YES) Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. |

Predictive analytics is less than 10% accurate for rare events • 1, 000 false negatives 1 million claims (1% fraud, 99% not fraud) • 99, 000 false positives = 91. 8% false positive rate • Actual positive predictive accuracy = 9, 000/109, 000 = 8. 2% at identifying fraud 99, 000 wrong not fraud 1, 000 wrong fraud 9, 000 correct fraud Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. 10, 000 fraudulent claims AI done right. | 8

The challenge: fraud/risk is getting more pervasive/growing. • Traditional surveillance approaches rely on rules or predictive based alerts which create many false positives and are not effective at dealing with organized networks. • Data and patterns change dynamically: can’t build millions of models or mine fast enough. • Risk grows as processing moves online/real time. • Building models and/or hiring data scientist(s) is not a scalable way to tackle future challenges. • Need an adaptive self-learning automated intelligent approach to keep pace with fraud. Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 9

Daisy in Insurance – Theory of Risk™ Daisy’s Risk Management Platform helps insurance companies drive dramatically improved business results REDUCE FALSE POSITIVE RATES FROM > 90% TO < 10% *average Daisy clients pre/post Quickly detect and efficiently investigate fraudulent claims, saving significant amounts of time. 80% Our AI-powered Theory of Risk™ makes us unique in the marketplace $$ Millions TIME SAVINGS FOR INVESTIGATORS IN FRAUD AVOIDED Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 10

Rules Basic Rules with Fuzzy Scoring • Standard If-Then-Else rules which are either True or False • Currently implemented 100+ rules • Score individuals on how many rules were broken and/or how many times each rule was broken • Probabilistic Rules Rule Mining • Decompose rules into antecedent • Assign a consequent score based on number of antecedent criteria met. Score based on antecedent correlation to consequent. • Assign a consequent score on how close to the boundary antecedents are • Each new claim affects rule statistics Methods will automatically adapt without human intervention • Use Association Rule Mining techniques to codify healthcare practice • Find procedures that shouldn’t go together or should always go together Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 13

Probabilistic Rules • IF PC LIKE '42%' AND WITHIN 1 YEAR OF SERVICE AND COUNT > 8 AND SAME PATIENT − − − PC LIKE '42%' AND WITHIN 1 YEAR OF SERVICE = 0. 7 COUNT 6 -8 = +0. 25 COUNT 9 = +0. 3 COUNT 10 -15 = +0. 5 COUNT > 15 = +1 • Now rule breaking is not a hard threshold • Can rank rule output Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 14

Network analysis: connecting individuals with common attributes Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. • Same names, addresses, phone numbers, email addresses, bank accounts, credit cards, loyalty cards, etc. • Between individuals who shouldn’t be connected. • Adaptive: Each new personal record or change dynamically alters networks. AI done right. | 15 15

Example: finding non-obvious matching individuals using A. I. Company: ABC Current Personal Info Cate Green 1 Bourne Street Clinton, MA, 01510 Old Personal Info Mary Catherine Green 4909 Battery Lane Bethesda, MD, 20814 DOB: 07/08/1965 CLAIM: September 7, 2010 Provider: 1234 Procedure code: 32221 Fee: $234. 57 Company: ABC Current Personal Info Mrs. Katherine M Greene 25 Old Oak Rd. Little Rock, AR, 72212 Old Personal Info Katherine M. Greundig 1 Bourne St Clinton, MA, 01510 DOB: 07/08/1964 CLAIM: September 8, 2010 Provider: 1234 Procedure code: 32221 Fee: $234. 57 Company: ABC Current Personal Info Mary Kate Greundig P. O. Box 567 Henderson, NV, 89044 Company: ABC Current Personal Info Kate Sinclair 4909 Battery Lane Bethesda, MD 20814 Old Personal Info 52 Old Oak Road Little Rock, AR 72212 DOB: 08/07/1964 Old Personal Info 2500 Bel Air Blvd San Clement, CA, 92407 P. O. Box 567 DOB: 08/07/1965 CLAIM: September 9, 2010 Provider: 1234 Procedure code: 32221 Fee: $234. 57 Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. CLAIM: September 10, 2010 Provider: 1234 Procedure code: 32221 Fee: $234. 57 AI done right. | 17

Example: using A. I. to find fraudulent root canal claims submitted by a group of individuals. Name: Kate Green Address: 1056 Windsor Street Phone: 617 -782 -5312 DL #: F 7382275 Acct #: 584939 Claim: Root Canal $613 Provider: John Doe Name: Kwan Kim Address: #102 -1056 Windsor St. Phone: 978 -365 -5312 DL #: B 9882765 Acct #: 6032 -799949 Claim: Root Canal $623 Provider: John Doe Name: Josh Stevens Address: 65 Flipper Way Phone: 276 -466 -8264 DL #: B 7483029 Acct #: 937105 Claim: Root Canal $554 Provider: John Doe Name: Tom Sinclair Address: 12 Bourne St Phone: 276 -466 -8624 DL #: H 7339986 Acct #: 799949 Claim: Root Canal $716 Provider: John Doe The occurrence of root canals in individuals is 1 per 1, 000 dental visits. Finding a network with 100% of network members having a root canal at the same provider is very suspicious. If you cannot relate the individuals by analyzing 100% of the data, the behaviour appears individually normal. Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 18

Scoring Curve Types Growth (high is suspicious) Decline (low is suspicious) • Scoring curves can be tuned to change the performance of peer and overall DSI • Different curve types can be applied to different metrics Bell (average is suspicious) Inverted Bell (low and high is suspicious) Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 19

How disruptive is AI in insurance? • Unprecedented improvement in risk operations: − Reduce investigative false positives to less than 10% increasing recoveries and avoiding fraudulent payments saving 10’s to 100’s millions of dollars in fraud and money laundering − Align credit decisions and premiums more accurately improving loss ratio by greater than 10% without giving up market share Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. |

Case study: Impact of Daisy fraud detection solution False positive rate, labour savings and recoveries pre-post using Daisy Results Post Daisy client investigative teams experience an order of magnitude gains in labor productivity and fraud avoidance 21 Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 21

How can insurance companies use Daisy’s AI solutions? Transaction Surveillance / Fraud Detection Identify suspicious transactions, transaction source/destination locations, individuals, and social networks. Underwriting 22 Identify the risk associated with new customers and set appropriate premiums or making the right credit decision. Loss Ratio Optimization Choose the optimal level of pricing, fraud detection and claims adjudication automation to achieve the desired loss ratio. Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 22

Investigative capabilities must keep pace with detection otherwise there will be growing backlog of investigative work. Streamline workflow • Ensure tools and processes enable efficient investigative workload Prioritize investigation • Apply investigative effort to those with high return • Tier investigative expertise required to evaluate Automate decisions Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. • Automate those patterns with high confidence and below human investigative priority threshold • Autonomous decision making • Robotic process automation AI done right. | 23

Why the Time for AI is Now 1. AI is accessible and delivering results now with verifiable ROI 2. AI reduces highly repetitive tasks for employees and lets them focus on more strategic priorities 3. AI can better identify risk and prevent fraud 4. AI deliver an invisible advantage Copyright © 2019 Daisy Intelligence Corporation. All rights reserved. Private & Confidential. AI done right. | 24