Reinventing Fraud Prevention & Underwriting with Machine - - PowerPoint PPT Presentation

▶
reinventing fraud prevention underwriting with machine
SMART_READER_LITE
LIVE PREVIEW

Reinventing Fraud Prevention & Underwriting with Machine - - PowerPoint PPT Presentation

Jun 18, 2023 175 likes •516 views

Reinventing Fraud Prevention & Underwriting with Machine Learning Ido Lustig VP Risk Lendit April 2016 Propriety and Confidential BlueVine flexible business lines of credit and invoice factoring 08/2013 03/2014 06/2014 12/2014

slide-1
SLIDE 1

Propriety and Confidential

Lendit

April 2016

Reinventing Fraud Prevention & Underwriting with Machine Learning Ido Lustig – VP Risk

slide-2
SLIDE 2

Propriety and Confidential

BlueVine – flexible business lines of credit and invoice factoring

Tel Aviv, Israel

  • R&D, Risk
  • 32 Employees

Palo Alto, CA

  • Biz, Ops, Sales
  • 34 Employees

$64M in equity and debt financing to date

06/2014 12/2014 08/2013 03/2014 12/2015

Founded + Seed Beta launch Full launch Series B Series C

slide-3
SLIDE 3

Propriety and Confidential

3 QUESTIONS

underlie our underwriting process

slide-4
SLIDE 4

Propriety and Confidential

?

slide-5
SLIDE 5

Propriety and Confidential

We need to ask the right questions 
 and answer them like (smart) humans would have.

Machine-human interaction is the key for scale and accuracy

slide-6
SLIDE 6

Propriety and Confidential

Machine-learning capabilities continually advancing

http://www.bloomberg.com/news/articles/2016-01-03/after-winning-at-chess-this-computer-may-help-decide-on-loans

slide-7
SLIDE 7

Propriety and Confidential

But it’s 
 still not perfect

slide-8
SLIDE 8

Propriety and Confidential

  • First and last name correlation with loss
  • Number of letters in each word
  • Total number of letters
  • Number of times each letter appears
  • Order of letters
  • …..
slide-9
SLIDE 9

Propriety and Confidential

Problem #1 - overfitting

Observations

https://shapeofdata.wordpress.com/2013/03/26/general-regression-and-over-fitting/

slide-10
SLIDE 10

Propriety and Confidential

Problem #2 – Equal Credit Opportunity Act: things you can’t use (and end up using…)

https://www.washingtonpost.com/news/wonk/wp/2015/05/26/what-your-name-says-about-your-age-state-job-and-political-leanings/

slide-11
SLIDE 11

Propriety and Confidential

https://www.washingtonpost.com/news/wonk/wp/2015/05/26/what-your-name-says-about-your-age-state-job-and-political-leanings/

Problem #2 – Equal Credit Opportunity Act: things you can’t use (and end up using…)

slide-12
SLIDE 12

Propriety and Confidential

http://fivethirtyeight.com/features/how-to-tell-someones-age-when-all-you-know-is-her-name/

Problem #2 – Equal Credit Opportunity Act: things you can’t use (and end up using…)

slide-13
SLIDE 13

Propriety and Confidential

https://www.washingtonpost.com/news/wonk/wp/2015/05/26/what-your-name-says-about-your-age-state-job-and-political-leanings/

Problem #2 – Equal Credit Opportunity Act: things you can’t use (and end up using…)

slide-14
SLIDE 14

Propriety and Confidential

Problem #2 – Equal Credit Opportunity Act: things you can’t use (and end up using…)

So what’s OK to ask?

And what would we be better off not asking at all?

slide-15
SLIDE 15

Propriety and Confidential

Problem #3 – understanding the outcome

  • Clear rejection reasoning (ECOA)
  • Debrief and improve your policies
slide-16
SLIDE 16

Propriety and Confidential

Our 2¢

Insight driven and data backed automation process

slide-17
SLIDE 17

Propriety and Confidential

Map Questions Automate Answers Expose to analysts Get feedback Fine tune features Retrain models

slide-18
SLIDE 18

Propriety and Confidential

Are documents doctored? Is the person who she claims she is? Does the business exist? Any evidence of criminal activity? Does the business have a decent website?

Ask the right questions (fraud example)


Does the activity match the client’s industry?

slide-19
SLIDE 19

Propriety and Confidential

Does the business have a decent website? (automation)

Guess URL

✓ Use user provided (www.idosbiz.com) ✓ Use email domain (sales@idosbiz.com) ✓ Use search engine API (search ido (AND biz OR business)

Crawl Website

✓ Download website ✓ Classify using internal model ✓ Use Industry as a standard

Website score

➢ Down ➢ Not found ➢ Weak ➢ Medium ➢ High

slide-20
SLIDE 20

Propriety and Confidential

http://glo4led.com/

slide-21
SLIDE 21

Propriety and Confidential

http://www.valleyisleaquatics.com/

slide-22
SLIDE 22

Propriety and Confidential

http://www.royalgranitesandgems.com/

Does the business have a decent website?
 Does the business have a decent website given the industry?

slide-23
SLIDE 23

Propriety and Confidential

  • Not self-derived from the data
  • Answer critical questions
  • Fine tuned, highly accurate
  • High coverage
slide-24
SLIDE 24

Propriety and Confidential

Map Questions Automate Answers Expose to analysts Get feedback Fine tune features Retrain models

slide-25
SLIDE 25

Propriety and Confidential

slide-26
SLIDE 26

Propriety and Confidential

slide-27
SLIDE 27

Propriety and Confidential

  • Hold both calculated and analyst values
  • Auto-retrain low performing variables
slide-28
SLIDE 28

Propriety and Confidential

Map Questions Automate Answers Expose to analysts Get feedback Fine tune features Retrain models

slide-29
SLIDE 29

Propriety and Confidential

  • Same process for features, models, and decisions
  • For high level models – use tagging (fully automated)
slide-30
SLIDE 30

Propriety and Confidential

Featur e Deploy Human Feedb ack Auto Retrain Featur e Deploy Reality Feedb ack Auto Retrain

slide-31
SLIDE 31

Propriety and Confidential

Automated Decision Rate and Accuracy

0% 25% 50% 75% 100% Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Coverage Accuracy

slide-32
SLIDE 32

Propriety and Confidential

Thank You