An Introduction to Predictive Analytics - Typepad › files › anintroductionto... · An Introduction to Predictive Analytics For Business Rule Developers Mac Belniak Principal Sales

This presentation cannot be reproduced or shared without FICO’s express consent.© 2009 FICO Corporation. 1

An Introduction to Predictive AnalyticsFor Business Rule Developers

Mac Belniak Principal Sales Consultant, Model Builder

Monday, August 24, 2009

© 2009 FICO Corporation.2 © 2009 FICO Corporation..2

Agenda» Decisions & Uncertainty

» Predictive Analytics

» Analytics & Rules

» Model Builder Demonstration

» Decision Management

» Q&A

© 2009 FICO Corporation.3 © 2009 FICO Corporation..3






» Q&A

© 2009 FICO Corporation.4

Uncertainty clouds decision making

Approve/DeclineLoan

DelinquencyHistory

CurrentLimit

CurrentBalance

Profit/Loss

Default?

Dec

isio

n Ti

mel

ine

Odds ofDefault


Predicting if an event will occur parts the clouds

» Binary outcome models» What are the odds an event will occur in a future

time period?» E.g. We can’t know if a borrower will not miss a

payment next year.» But we can predict the odds that they will not miss a

payment next year!» 10:1

» The prediction can help inform future decisions.» Examples

» What are the odds that a prospect will respond to a campaign?

» What is the likelihood that this transaction is fraudulent?» What are the odds that a customer will attrite if I lower

their credit limit?


Predicting amounts brings even more clarity

» Continuous outcome models» What will an amount be in a future time period?» E.g. We can’t know how much we will lose if a

borrower defaults.» But we can predict the amount that will be lost!

» $106,000

» The forecast can help inform future decisions.» Examples

» How many dollars will the prospect spend if they respond to the campaign?

» How much will a cardholder’s utilization change if I increase their credit limit?







» Q&A

© 2009 FICO Corporation..


Predictive models have a natural lifecycle

Access Prepare Explore

Data

Discover Approve Deploy

Model

Rapid scorecard

development Streamlined model review

Fast migration into decision application


Historical data is a key input for predictive modelingPe

rfor

man

ceD

evel

opm

ent

3 M

onth

s24

Mon

ths

History Behavior Measurement

Tenure?Months since account opened.

Delinquency?

30 days delinquent in last year.Search for Credit?

# Inquiries last 6 months.Utilization?

Balance / Credit Limit.

Target:Good or bad?

Was the customer 30, 60, or 90 days delinquent or charged off during the

performance period?

2007

2008

2009


Predictive models learn from historical patterns

Target Variable

Custom

er ID

Mon

ths since accoun

t ope

ned

30 days de

linqu

ent in last year

# inqu

iries last 6 mon

ths

Balance / Cred

it Lim

it

Goo

d or Bad

000000001 9 0 1 30% 0000000002 2 0 0 64% 0000000003 11 0 2 100% 0000000004 7 1 1 84% 1000000005 8 0 1 30% 0000000006 17 0 0 37% 0000000007 3 0 1 83% 0000000008 11 0 1 41% 0000000009 23 0 2 73% 0

Input Variables

Historical Data

Scorecard Algorithm

Model Development Predictive Model


For each variable,understand if and how it predicts historical outcomes

Odd

s of

Goo

d

10:1

5:1

1:1

15:1

20:1

Total Portfolio Odds

# inquires in the last 6 months

None 1 2 3 4


Find the most informative set of variablesO

dds

of G

ood

# inquires in the last 6 months

Odd

s of

Goo

d

Years at current address


Variable Weight# inquires last 6 months

None 531 342 – 10 24

# times 30 days delinquent last yearNone 531 242 633 – Max 33

Months since account opened0 – 36 4537 – Max 24

Variable Weight# inquires last 6 months

None 641 552 – 10 48



Train the predictive model

Target Variable

Custom

er ID

Mon

ths since accoun

t ope

ned

30 days de

linqu

ent in last year

# inqu

iries last 6 mon

ths

Balance / Cred

it Lim

it

Goo

d or Bad

000000001 9 0 1 30% 0000000002 2 0 0 64% 0000000003 11 0 2 100% 0000000004 7 1 1 84% 1000000005 8 0 1 30% 0000000006 17 0 0 37% 0000000007 3 0 1 83% 0000000008 11 0 1 41% 0000000009 23 0 2 73% 0

Input Variables Scorecard Algorithm


Generate scores and rank order the customersCu

stom

er ID

# inqu

iries last 6 mon

ths

30 days de

linqu

ent in last year

Mon

ths since accoun

t ope

ned

000112354 1 0 9…

Input Variables Variable Weight# inquires last 6 months

None 641 552 – 10 48



55

74

+

51

+

= 180

150 175 200 225

GoodBad

Score


Evaluate the quality of the ranking

Customers Sorted by Scores

10%

20%

% B

ad C

usto

mer

s B

elow

the

Scor

e

30%

40%

50%

60%

70%

80%

90%

100%0%

WorstCustomers

BestCustomers

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0%

The 20% of customers with the lowest score cover over

70% of the bads!







» Q&A



Measuring benefitof analytics with champion/challenger

Champion: The rules and models that currently automate decisions.Challenger: Alternative rules and models that are likely more effective.Champion/Challenger: Running the challenger on a small fraction of the population to verify that it is better.

IF account_number ends in 0-8Champion

IF behavior_score > 650&& utilization > 70%…

THEN up limit by 10%

Champion$ Profit / Account

Challenger$ Profit / Account

ELSEChallenger

IF behavior_score_1 > 750&& utilization > 80%…

THEN up limit by 5%







» Q&A



Model Builder supports the entire analytic lifecycle

Access Prepare Explore Model Deploy

» SAS Datasets» Special Values

» Teradata» Oracle» SQL Server» Fixed-Width &

Delimited Text; ASCII & EBCDIC

» MB Native Files

» Replace Missing» Sample

» Stratified» Random

» Partition Train/Test» Sort» Join/Merge» Append» Filter/Where» Variable Creation

» Arrays, Regular Expressions, etc.

» View Table» Statistics

» Summary» Frequency» Correlation» By-Variable

» Dataset Comparison

» Linear Regression» Logistic

Regression» Neural Networks» Reason Codes» K-Means & Bang

Clustering» PCA» Evaluation

Reports» ROC, GINI, KS,

etc.

» Services-Oriented-Architecture» Batch» Transactional

» PMML» Blaze Advisor» Teradata

» Scorecard Tracking» Char. Analysis» Population Stability


Historically, deployment has been expensive

» Traditional technique» Document model» Ask IT to recode» Lengthy testing

» How many € £ $ lost per day?

Time To Deploy Predictive ModelsSelf-Reported Measures, Global Financial Companies

PredictiveModelSpecs

CompiledExecutableIT Software

Development

SQL inDatabase


Quickly deploy via Blaze Advisor

» Imported models are available to rule developers and authorized business users can see and modify them

» PMML integration

ImportDM

RepositoryRule Service

.NET

Rule Service

Java

Code Gen

COBOL

Model Builder


Model Buildersupports scorecard tracking to trigger model refresh

» Easily evaluatepopulation stability

» Track shifts in score distribution

» Analyzecharacteristic-level changes

© 2009 FICO Corporation.23 © 2009 FICO Corporation. 23

DemonstrationBehavior Risk Scorecard

Import Data from SASCheck the DataFind Predictive PatternsSelect Best PredictorsEvaluate Scorecard QualityExport to Blaze Advisor







» Q&A



Fast, Fast, Fast

» Stronger predictive models in less time

» Accelerated analytic review & approval

» Rapid deployment into production systems

Access Prepare Explore

Data

Discover Approve Deploy

Model

Minimizing Time to Better Decisions with Model Builder


Customer Lifecycle SolutionsACQUIRE MANAGE PROTECT

Marketing Origination Customer Management

Collections & Recovery

Fraud

Decision Management Applications

Marketing Solutions

Capstone®

Decision ManagerLiquidCredit®

Service

TRIAD™adaptive control

system

Debt ManagerRecovery

Management System

Falcon®

Fraud Manager

Lifecycle Analytics and Services

PreScore®

ServiceOrigination and Small Business

Scores

Transaction Risk Scores

Debt Placement Services

Collection Scores

CardAlert Service

Across the LifecycleAnalytics Industry Standard and Custom

Credit Bureau Scores: FICO® Score Global FICO® Score FICO® Expansion ScoreInsurance Scores FICO Credit Capacity Index™ MyFICO®

Custom Analytics: Predictive Analytics Optimization Portfolio AnalyticsDecision Management Tools

For Application Developers and In-House Analytics TeamsBusiness Rules Management: FICO Blaze Advisor®

Model Development: Model BuilderOptimization: Decision Optimizer Xpress-MP Suite

Professional Services

Analytic ConsultingBusiness and Technology Consulting

FICO address all stages of the customer lifecycle


AnalyticCapabilities

SharedWorkflows

Data

Create value by connecting decisions

Decision ManagementApplication

Smarter DecisionDecisionLogic

Other Applicationsand

Lifecycle Areas


FICO is the leader in Decision Management

“FICO is where the mathematical approach to problem-solving that is inherent in today’s scores and analytics all began.”

— William Blair & Company

PIONEER Introduced many analytic breakthroughs, including credit scoringIntegrated predictive analytics with rules management

LEADER Products that serve whole industries at high scale» 10 billion FICO scores sold a year» Reviewing 20 billion card transactions a year for fraud

PARTNER Expertise on best practices in Decision Management, including operational, legal and regulatory issuesServing many of the world’s top businesses: » 9 of the top 10 Fortune 500 companies» 2/3 of the top 100 banks in world » 2/3 of top US P&C insurers» 1/2 of the top US retailers

© 2009 FICO Corporation.29 © 2009 Fair Isaac Corporation. Confidential.29

To Learn More…Webinar: Enabling COBIT and IT Governance through Business

RulesSeptember 9

FICO Decision Management Tools User GroupSeptember 16-18

Register at: www.fico.com/events

Webinar: Business Change using BPM and BRMSSeptember 23

Or contact us: [email protected]

+1 888 342 6336+44 (0) 207 940 8718

Visit the Decision Management Community:decisions.fico.com







» Q&A


This presentation cannot be reproduced or shared without FICO’s express consent.© 2009 FICO Corporation. 31

THANK YOU

Mac Belniak [email protected]

An Introduction to Predictive Analytics - Typepad › files › anintroductionto... · An Introduction to Predictive Analytics For Business Rule Developers Mac Belniak Principal Sales

Documents