Introduction To Predictive Analytics: FICO Solutions Education

Introduction to Predictive Analytics
FICO Solutions Education
Confidential. The material in this presentation is the property of Fair Isaac Corporation, is provided for the recipient only, and shall not be used, reproduced, or disclosed without Fair Isaac Corporation's express consent. 2010 Fair Isaac Corporation.
Nic Orton VP, Product Management FICO
What will you learn?

The purpose of this session is to teach you about the Predictive Analytic capabilities that are part of FICO solutions.
By the end of this session you will be able to: Understand and articulate the role that Analytics play in FICO Solutions and how it adds to FICOs Value Propositions Understand the primary types of Analytics offered by FICO From Predictive models; through to Decision Modelling & Optimization
that are the central part of the FICO triangle

Predictive Analytics
Tools
2010 Fair Isaac Corporation. Confidential.
Agenda
1. FICO Decision Management Portfolio and the Decision Engine 2. Predictive Analytics Terminology
3. Predictive Models
4. Decisioning Analytics / Decision Modelling & Optimization 5. Key Points to Remember
Tools
Decision Management Portfolio Key Products

Customer Lifecycle Solutions Acquire
Marketing Decision Management Applications Origination
Manage
Customer Management FICO TRIAD Collections & Recovery FICO Debt Manager FICO Recovery Management System
Protect
Fraud
FICO Precision Marketing Manager FICO Retail Action Manager
FICO Origination Manager FICO Capstone FICO LiquidCredit
FICO Falcon Fraud Manager FICO Insurance Fraud Manager
Across the Lifecycle & Embedded in Applications
Scores & Analytics

Decision Management Tools
Scores: FICO Score FICO Insurance Risk Scores myFICO

Analytics: Models ( Predictive Analytics ) Decision Modelling & Optimization Portfolio Analytics Business Rules Management & Decision Services: FICO Blaze Advisor Predictive Analytics: FICO Model Builder Optimization: FICO Xpress Optimization Suite FICO Decision Optimizer
Services
Analytic Services Business and Solution Consulting

4
Basic Components of a Decision Management Solution
Data In
Decision Engine
Decisions Out
Business Processes
Data
Data
Inside the Decision Engine
Benefits: Decision Engines automate decision/s to improve them through

Consistency Efficiency Use of analytics to render better (more profitable) decisions
Predictions
Rules
Optimization
Adaptive Control
Are you thinking to yourself.
How can I show Analytics make the decision smarter?
What is the Value of Analytics
Value of Analytics = Better Business Results!
3. Achieve better 2. Make better

business results
decisions on interactions with their customers and prospects
1. Leverage all
available data, insights and expertise

Example Decision: Should I give this customer a credit line increase?
1.Leverage all available data,

insights and expertise
Customer data available:

2. Make better decisions on their

interactions with their prospects and customers Based on the customer data, what should the action / treatment be?

3. Achieve better business

results If an organization executes the best action / treatment, they will experience:
Master file data Credit card transaction data Credit bureau data
No increase $500 $1000 $2500 $5000
Increased share of wallet Revenue Reduced losses Maintained / improved attrition rates Profit!
9

Case Study: Credit Line Optimization
Challenge: Increase profitability while controlling losses through optimal assignment of credit line increases
Incremental Profit Growth
$10.00 $9.00 $8.00 $7.00 $6.00 $5.00 $4.00 $3.00 $2.00 $1.00 $0.00
Results: Profitability per account increased by $10
10
11
12
13
14
15 M
on th
on th
on th
on th
on th
on th
on th
on th
on th
on th
on th
on th
on th
on th
on th
Insight Bank
on th
16
10

Case Study: Loss Ratio Models for Insurance
Challenge: Increase profitability for auto insurance underwriting for Brazilian insurer, in a competitive environment with high attrition and low penetration.
Results: Profits increased 130% with a reduction in the number of unprofitable policies. Claims shrank by 25%, Loss ratio dropped by 4.5%
11
Predictive Analytics within The Decision Management Portfolio

FICOs predictive analytics help businesses gain deep consumer insight and make smarter decisions. FICO: Make Every Decision Count
Tools
12
Predictive Analytics Terminology
13
Terminology
Predictive Analytics is an area of statistical analysis that deals with extracting information from data and using it to predict future trends and behavior patterns. -- Wikipedia
Better Decisions:
Who should I market this product to? Should I accept this applicant? How can I upsell this customer? Should I call this delinquent account early? Should I reject this transaction for fraud?
Better Decisions!
14
Terminology
Predictive Models, Scorecards:
A general term for an algorithm developed to classify or rank-order entities, such as applicants, account holders, or customers, based on data inputs and built to predict a specific outcome .
Example: Application risk model, behavior model, transaction model, propensity to buy model
Scores:
Scores are the output of the mathematical algorithm (Predictive Model / Scorecard) that can be related to a specific level of Risk / odds. Example: 720 is a possible score for a FICO Credit Bureau Risk Score
Economic Impact Solutions:
Predicts how the risk levels observed at each score band will change based on different economic conditions. Example: A person who has a 720 FICO Score today will behave more like a 700 under a continued recession
15
Terminology
Strategies:
A set of business rules or decision logic that is used to assign a certain action or treatment to a specific type of customer. Example: Credit line increase strategy
Action, Treatment:
The end result of a business decision Example: Give customer a $1000 credit line increase
Predictions vs. Decision Strategies:
Predictions: One way to improve decisions is to improve the prediction of a future event at the time the decision is made. Example: predict risk at time of origination Decision strategies: Another way to improve decisions is to improve the decision logic, or strategy used to apply an action to the customer. Example: Credit line strategy
16
Predictive Models
17
Predictive Models
Building Predictive Models
To build Predictive Models, we first collect two snapshots of historical data:
1.
Observation Data: Capture all data that was known at the time to make the historical decisions.
2.
Performance Data (for example, 1-2 years later): Capture the outcomes of those decisions.
PAST
Observation Period Performance Period Build the Model
FUTURE
DEVELOPMENT PERIOD
18
Predictive Models
Next, we process the historical information and compare the characteristics of the observation data with the performance of the accounts (for example, 1-2 years later). Whats the difference?
What patterns exist within the good customers? What patterns exist within the bad customers?
19
Predictive Models
Next, we process the historical information and compare the characteristics of the observation data with the performance of the accounts (for example, 1-2 years later). Whats the difference?
What patterns exist within the good customers? What patterns exist within the bad customers?
OWN/ RENT
% GOOD
% BAD
INFO ODDS
TIME WITH EMPLOYER
% GOOD
% BAD
INFO ODDS
Own Rent
Live w/ Parents
60
20
3:1 1:1 3:7
<3 3-5 >5
10 30 60
40 30 30
1:4 1:1 2:1

20
10
30
10
70
Predictive Models
After we evaluate all of the different variables, weve determined which past behavior predicts future behavior. These variables / characteristics come together in the form of a scorecard.
Characteristics
Assigned Score Weights for given Customer
Attributes
21
Predictive Models
Evaluating Predictive Models
Goal: Separate the distribution of the goods from the distribution of the bads as much as possible
The farther away the goods are from the top of the bads, the better your model K-S is similar to Divergence the bigger the number, the better the model
DIVERGENCE
# OF APPLICANTS
GOODS
BADS
80 100 120 140 160 180 200 220 240 260 280
22
SCORE
Predictive Models
Rank Ordering
Scores rank order consumers, prospects, accounts or

transactions
Less Likely 20:1 odds

Paying as agreed
. by the likelihood or propensity to do what it is we are trying to predict
For example, a Credit Risk Score answers Which

consumers are more likely to go 90 days past due on their credit obligation?
680
More Likely 3:1 odds

Paying as agreed
660 640
620
23
Predictive Models
Custom Predictive Scores - Originations for Card Portfolio Benefits Reduction Profit % in Loss Receivables 24% 1.3% 19% 36% 28% 24% 19% 36% 38% 1% 1.9% 1.5% 5.1% 4% 7.7% 6%
Region
Previous Population Scorecard Type Quality Custom Empirical Prime Subprime Prime Subprime Prime Subprime Prime Subprime
Profit / applicant $12 $9 $18 $14 $16 $13 $24 $19
N. America (NA) Pooled or bureau model

Latin America (LAC) Custom Empirical Pooled or bureau model
Custom Predictive Scores allows lenders to better differentiate their customers and improve decisions
24
Pop Quiz!
Risk Models are designed to:
a) b) c) Tell us what someone did in their past Predict which individuals will be bad Rank order individuals by risk levels
25
Predictive Model
Capabilities Highlight: Economic Impact Analytics / Solutions
26
Economic Impact Analytics / Solutions

Lenders must also understand how the odds
of default in their customer
segments & portfolios shift as market conditions change.
What are the odds of default for risk score 670 in an economic downturn?
What are the odds of default for risk score 670 in an economic upturn?
Gross Domestic Product (GDP) United States 2008-2010

27

FICO Economic Impact Analytics Original Risk Score
(i.e. FICO Score, Origination Score, Behavior Score)
Customer & Portfolio Segment

e.g.. Location
Historical performance
+ Actual economic + Macroeconomic

variables
Anticipated
e.g.. Delinquency Cycle e.g.. Loan/Product Type Etc., Etc.
tells our clients:
If [ current & projected economic conditions are Y ]
then [ your risk scores will represent these odds of default ]

28

Client Benefits FICO Economic Impact
By using economic impact analytics to translate market directions into actionable portfolio or customer decisions, our clients will be able to:
1. Limit losses
Tighten up credit policies sooner and for the right populations during economic downturns
2. Grow portfolio responsibly
Determine when and how to proactively loosen up credit policies as markets recover
3. Better prepare for the future
Simulate and quantify the impact of future macroeconomic conditions on their score and portfolio performance to make sure strategies are aligned with their risk preferences.
29

Case Study Results
Lifecycle Originations Benefit $11.50 profit lift / applicant 15% reduction in loss 0.2% lift in profit % receivables Notes Region: Eastern European Product Type: Personal Loan Benefit gained from using EIS to guide cutoff decisions Region: US Product Type: Cards Benefit based on using EIS score vs. traditional behavior score in CLD and Bucket 1 collection strategies Results derived from back validation in an out of time sample
Account $3.50 yearly profit lift / Management active /Early Stage 0.1% lift in yearly profit Collections % receivables
Benefit of EIS comes from the ability to more accurately predict risk and make better decisions, in expectation of future economic conditions
30
Predictive Model
Capabilities Highlight: Transaction Data, Profiles and Scores
31
Transaction Data, Profiles and Scores

Transaction data is everywhere! But underutilized
Credit card Debit card and deposit Click stream Purchase history
Captures lots of detail about a Transaction Data, customers Scores Profiles and behavior Transaction analytics can help you make more accurate and timely decisions across

Credit Facilities Collections Pre-delinquency Authorizations Early-life management Marketing More

32

Transaction Data Contain a Wealth of Information
Looking only at summarized data, these two cardholders appear to have similar risk
Summarized Data
Cardholder #1
Cash Merchandise
Cardholder #2
33

Transaction Data Contain a Wealth of Information
The transaction data shows a difference in the timing of activities which can indicates that the Cardholders have different risk profiles
Summarized Data
Cardholder #1 Cash Merchandise
Cardholder #2
34

Finding Early Signs of Risk
35
Challenges in Using Transaction Data

Modeling
Cleaning the data Capturing the full power of the data Finding the right characteristics
Operational Implementation
How to process and return decisions rapidly with so much data? Not only a storage issue, but processing for real time decisions
FICO has been modeling and operationalizing transaction-based credit risk scores since 1996
36

Patented Profiling Technology
Profiles
Latest patterning technology captures all relevant data for any event Combines all information about an account in one integrated representation
Each account profile includes statistics describing:

Type, frequency and size of transactions Patterns of behavior Recent trends Significant events
The account profile is the key enabling technology for transaction-level models predicting fraud risk, credit risk, attrition, marketing purchase propensity, etc.
37

Application of Adaptive Analytics in Fraud
Adaptive Models allow the system to learn from fraud case dispositions as they show up Enabling changes to be captured more quickly into Predictive Models Adaptive Analytic models augment robust consortium models with real time updates
Example: Improved performance with shifting fraud patterns like Card-NotPresent fraud detection How does it happen? A new blended score is created from the scores of the base and adaptive models
38
Decisioning Analytics / Decision Modelling & Optimization
39
What Is Decision Optimization? An analytic technology that helps banks choose the optimal decisions on each customer in their portfolio What is the optimum amount to lend customers to maximize interest income but are significant but conflicting business Especially when therecontrol portfolio bad debt below $M ? goals and constraints.
Balance Growth
Maximize Profit
Capital Early Repayment Control Losses
Increase Debt Product Repayment Cross Sell
Regulatory Resources Requirements

40
Actions Impact Future Profitability

Decision Models can increase accuracy of those predictions
Account Profitability
KEY Profitability Prediction (No Action)
Profitability
Potential Actions
Today
Action A Action B Action C
Break-Even
Current profit trajectory
Time Decision Models predict the different reactions from the actions a lender can take
41
Decision Modelling
Evaluates customers reactions to your actions
The decision model - captures customers reactions to your actions
Action Customer
Offer #1
Reaction
E(Bal. change) = $150 E(Inc. loss) = $6 E(Inc. profit) = $10 ...
Risk score = 680 Revenue score = 720 CB Balance = 2,250 CB Utilization = 61% Segment = A
Line Increase Option 1: $1,000
Offer #2
E(Bal. change) = $210 E(Inc. loss) = $9 E(Inc. profit) = $15
Line Increase Option 2: $2,000
42
Decision Modelling
How does Decision Optimization work?
Action - Effect modeling is more easily seen in a graphical representation, an Influence Diagram, that captures the interactions between actions and effects.
Current Balance Credit Bureau Score Good/Bad $ Rev Credit Line Change $ Loss Current Limit Utilization
Behavior Score
Profit
The ability to model customers reactions to your actions is a key FICO advantage
43
FICO Decision Optimization Methodology
Accelerated Learning
Test efficiently to learn beyond historical operating regions to further increase future performance.
Decision Modeling
Establish mathematical relationships between customer treatment options, their reactions and profitability.
Interpretation
Gain insight into key profit drivers and opportunity pockets through diagnostics and final strategy engineering.
Simulation and Optimization

Identify optimal strategy scenarios subject to your multiple goals, business constraints and forecasts for the future.
44
FICO Decision Optimization Methodology

Provides Insight Into Trade-offs Among Competing Business Objectives Credit Line Management Efficient Frontier
Projected Profit and Loss Consequences for Multiple Optimization Scenarios allows users to simulate multiple what-if scenarios and see what is possible to achieve while exploring tradeoffs between business goals and constraints 1,200 Scenario 1 Projected PROFIT per Account 1,150
Without any incremental losses, profits can be increased by more than 30 per account
G F E D J
H I
1,100
Efficient Frontier
1,050
B
Scenario 2
Baseline / BAU
Avg Revenue = 2,000 Avg Loss = 970 Avg Profit = 1,030
1,000
A
If we accept 10% higher losses, profits can be increased by $100 per account
950 -10%
0%
10% 20% Projected CHANGE IN LOSS over "Baseline"
30%
45
Decision Modeling and Optimization

Client Results Lifecycle Originations Pricing & Amount Benefit 26 profit lift / applicant 1.5% lift in profit % receivables Notes Region: UK Product Type: Personal Loan
Originations Loan Amount

Account Management
$30 profit lift / applicant 5.6% lift in profit % receivables

Typically $8 12 yearly profit lift / active account
Region: Europe Product Type: Personal Loan

Regions: UK & US Product Type: Cards Results vary depending on historical operating point
46
Decision Modeling and Optimization

Client Results Lifecycle Collections (Cycle 1) Benefit $5 yearly profit lift / cycle 1 account 0.5% lift in yearly profit % receivables $10 yearly profit lift / cycle 2 account 0.1% lift in yearly profit % receivables Notes Region: Asia Product Type: Cards Projected results using decision model Region: US Product Type: Cards Projected results using decision model
Collections (Cycle 2)
Fraud
$0.5 profit lift / total account 3-4 bp of sales
Region: US Product Type: Debit Results derived from back validation
Decision Modeling & Optimization allows lenders to maximize profit by finding the optimal decisions to make on their customers.
47
Pop Quiz!
What are the benefits of optimizing decision strategies
a) b) Better decisions Ability to explicitly manage trade offs between risk and reward c) d) Improved profitability All of the above
48
Key Points to Remember
49
Key Points to Remember

1. Analytics is the brain that improves decisions 2. There are two key types of analytics: Predictive models: providing insight through predicting future trends and behavior patterns Decision Analytics / Decision Modelling & Optimization: defining the best / optimal action to a given customer or account in line with the overall business operating goals & constraints 3. Analytics within FICOs Decision Management Portfolio adds Value to the Solutions helping businesses gain deep consumer insight to make smarter decisions. And so making very decision count
Tools
50
Thank You!
51

Introduction To Predictive Analytics: FICO Solutions Education

Uploaded by

Copyright:

Available Formats

Introduction To Predictive Analytics: FICO Solutions Education

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Predictive Analytics: FICO Solutions Education

Uploaded by

Copyright:

Available Formats

Introduction to Predictive Analytics

FICO Solutions Education

Nic Orton VP, Product Management FICO

What will you learn?

that are the central part of the FICO triangle

Decision Management Portfolio Key Products

FICO Precision Marketing Manager FICO Retail Action Manager

FICO Origination Manager FICO Capstone FICO LiquidCredit

FICO Falcon Fraud Manager FICO Insurance Fraud Manager

Across the Lifecycle & Embedded in Applications

Scores & Analytics

Scores: FICO Score FICO Insurance Risk Scores myFICO

Analytic Services Business and Solution Consulting

2010 Fair Isaac Corporation. Confidential.

Basic Components of a Decision Management Solution

2010 Fair Isaac Corporation. Confidential.

Inside the Decision Engine

Benefits: Decision Engines automate decision/s to improve them through

Are you thinking to yourself.

How can I show Analytics make the decision smarter?

What is the Value of Analytics

2010 Fair Isaac Corporation. Confidential.

Value of Analytics = Better Business Results!

3. Achieve better 2. Make better

Value of Analytics = Better Business Results!

1.Leverage all available data,

2. Make better decisions on their

3. Achieve better business

No increase $500 $1000 $2500 $5000

2010 Fair Isaac Corporation. Confidential.

Value of Analytics = Better Business Results!

Results: Profitability per account increased by $10

Value of Analytics = Better Business Results!

2010 Fair Isaac Corporation. Confidential.

Predictive Analytics within The Decision Management Portfolio

Predictive Analytics Terminology

2010 Fair Isaac Corporation. Confidential.

Economic Impact Solutions:

2010 Fair Isaac Corporation. Confidential.

Predictions vs. Decision Strategies:

2010 Fair Isaac Corporation. Confidential.

2010 Fair Isaac Corporation. Confidential.

2010 Fair Isaac Corporation. Confidential.

2010 Fair Isaac Corporation. Confidential.

TIME WITH EMPLOYER

3:1 1:1 3:7

<3 3-5 >5

1:4 1:1 2:1

2010 Fair Isaac Corporation. Confidential.

Assigned Score Weights for given Customer

2010 Fair Isaac Corporation. Confidential.

Scores rank order consumers, prospects, accounts or

Less Likely 20:1 odds

. by the likelihood or propensity to do what it is we are trying to predict

For example, a Credit Risk Score answers Which

More Likely 3:1 odds

Profit / applicant $12 $9 $18 $14 $16 $13 $24 $19

N. America (NA) Pooled or bureau model

2010 Fair Isaac Corporation. Confidential.