Application of Decision Sciences

to Solve Business Problems

Retail Banking Industry

Analytics

for Retail

Banks

A Strong P&L Discipline to all

Analytics

New Accounts Acquired

Accounts Closed

Account Activation rate

Payment Rate

Total Ending Receivables

Interest

Cost of Funds

Net Interest Margin

Risk-based Fees

Interchange

Affinity Rebates

Cross-Sell

Annual Fees

Net Credit Losses

Net Credit Margin

Operating Expenses

Loan Loss reserve

Net Income

RE

VE

NU

ES

EX

PE

NS

ES

Bank P&L

Acquire New Customers- Segments X Products X Channel- Mailbase Expansion- Pricing

Reduce Customer Attrition- Voluntary / Involuntary

- Retention Strategies- Winback

Increasing Activation Rates- Deepening Engagement- Inactive Customer Treatment

Improve Profitability of Assets

- Balance Transfer- Credit Line Strategies

- Pricing

Maximizing Interest Revenue- Product Pricing- Customer Behavior – Revolvers, transactors, etc

Maximizing Fee Revenue- Over Credit limit

- Delinquency- Bad Check

Reduce Net Credit Losses

- Credit Line strategies

- Pricing strategies - Collections

Increasing Cross-sell Revenues- Revenue Enhancing Products- Breadth of relationships

Marketelligent brings a top-down approach to all Analytics. Our analytics expertise impacts all line items of a Business P&L

Strategic Reporting

Strategic MIS & Reporting

Having a multi-dimensional view of critical business metrics available in real-time is key to effectivemanagement of a Business. Marketelligent helps banks in putting together Strategic MIS across Product andFunctions without having to invest heavily in bespoke IT systems and BI packages. Our capabilities includeETL, development of analytical databases, identification of key metrics and views, SAS-based datamanipulation and report delivery on visually-rich platforms.

Sample Reports include: Acquisitions and profile of New Customers Portfolio & Vintage performance Credit lines and Utilization Delinquency, Risk, Loss and Collections Product P&L’s and trends across time

Marketing Analytics

Marketing Analytics

Marketing Analytics covers all functions of a Bank that help the business get a better understanding of itsCustomers thereby leading to a deeper Customer Engagement and enhanced revenues. Marketelligentbrings unique expertise in key functions across the Customer Lifecycle:

Customer Segmentation Customer Lifetime Value (CLV) Increasing Usage and building profitable balances Campaign design, management and tracking Measurement of marketing effectiveness Cross-sell and up-sell Customer Retention and win-back Marketing Response Scorecard development, validation and maintenance

Acquisition Usage & Loyalty

Val

ue

Time

Retention

Activation

Marketing Analytics

Customer Segmentation

In today’s competitive business scenario with customers having a multitude of options, their preferences andbuying patterns have been constantly evolving. For retaining the profitable and loyal customers, it istherefore necessary to keep track of changing customer trends and accordingly tailor the offerings.

Segmentation is the practice of identifying homogenous groups of customers based on their needs, attitudes,interests and purchase behavior. It enables identifying profitable customer segments and customizingproduct and service offerings and marketing campaigns to target them effectively. It is typically done using acombination of transaction data, demographic data and psychographic information.

It aids in answering critical business questions like: Which are the most profitable and loyal customer segments and how do we have tailored offerings for

these segments? How do we have special promotion campaigns, specifically to reach the high value customer segments? What are the revenue and profit contributions by different customer segments?

Platinum: Current investment > 50KGold: Current investment > 5K; < 50KSilver: Current investment < 5K

Tenure<12mo

All Customers1,889

1,637 MM EAD87k AED/Customer

New Customers4,568 (24%)

433 MM EAD (27%)95k AED/Customer

Existing Customers11,573 (76%)

1,203 MM EAD (73%)

84k AED/Customer

Savers2,944 (16%)

39 MM EAD (2%)13k AED/Customer

Investors7,316 (38%)

812 MM EAD (50%)111k

AED/Customer

Redeemers871 (5%)

60 MM EAD (4%)69k AED/Customer

Revolvers3,190 (17%)

292 MM EAD (18%)92k AED/Customer

Platinum

Gold

Silver

Platinum

Gold

Silver

Platinum

Gold

Silver

Platinum

Gold

Silver

Platinum

Gold

Silver

Marketing Analytics

Customer Lifetime Value

Customer Lifetime Value (CLV) represents how much a customer is worth in monetary terms and is based oncustomer’s expected retention and spending rate. It can be defined as the present value of the total profitexpected from the customers during the entire period they do business with the company. CLV analysis usescustomers’ past transaction data and employs predictive modelling techniques to forecast how much eachcustomer would contribute to the company’s revenues and profits till they remain with the company and donot attrite. The analysis can also be extended to estimate the lifetime values of new customers. CLV analysistakes into account estimated annual profits from customers, duration of business relation of the customer,and the discount rate to assess the net present value of the customers.

CLV analysis is used for: Forecasting the expected revenue from new customers and weighing it against the acquisition and

retention cost for them Deciding how much to spend on marketing programs for different customers Identifying the high value customer segments that can contribute the maximum to company’s revenue

and have special offers for them Identify the prospects who can become profitable for the company

MonthlyExpenses

MonthlyNet Revenues

CustomerTenure

Net Margin

AccumulatedMargin

AcquisitionCosts

Customer Lifetime Value

Predict monthly Revenues

Predict Customer Attrition

Predict Response RatesFrom existing P&L’s

Marketing Analytics

Profitability & Loyalty analysis

For the sustainable growth of any enterprise, it is very important to identify the most profitable and loyalcustomers. Having special schemes for these customers in form of offers and discounts, can help in realizingthe long term goals of increasing profits and expanding customer base.Organizations use customer profitability and loyalty analysis to identify the most valuable customer segmentsto prioritize marketing, sales and service investments. Transactional behaviour is analysed for creating aCustomer Value Score (C-score) for each customer, which explains their engagement levels. The C-Score canbe leveraged for proactive action to defend, retain and grow the customer base.

This can help answer key business questions like: Which are the most profitable and loyal customer segments and how much they contribute to the firm’s

profit? Which are the customer segments to be targeted for marketing programs and special offers? Which are the customer segments that can have a negative impact on the company’s profitability?

0%

10%

20%

30%

50%

Pro

fit

Co

ntr

ibu

tio

n

10% 20% 30% 40% 50% 60% 70% 80% 90% 100%Watch Out:Customers accounting

for losses.

Focus:Customers who

have growth potential

Sustain:Highly profitable customers

Invest & Sustain:Profitable customers

%Customers

Invest :Highly profitable

customers40%

Cross Selling & Upselling Strategies

It is not just enough to retain the profitable and loyal customers, but it has become a necessity to increasethe revenue contribution from the existing customer base. Cross Sell involves the sale of additional items inorder to increase the wallet share from the customers.

Market basket analysis is the technique used to evaluate customers’ purchasing behaviour and to identify thedifferent items bought together in the same shopping session. It uses transactional data and employspredictive modelling techniques to identify customers’ preferences based on the associations between theproducts recently purchased. It helps to determine which products are to be offered and which are thecustomer segments most likely to be receptive to these cross selling propositions.

It aids in strengthening relations with customers by: Customizing layouts, product assortments and pricing so that it appeals to the customers Designing effective affinity promotions

• Stimulating trials and increase customer awareness during launch of new products and variants• Handling excess stock by designing offers among associated products

Marketing Analytics

CONFIDENCE Product 1 Product 2 Product 3 Product 4 Product 5 Product 6 Product 7 Product8

Product 1 100% 25% 9% 6% 18% 2% 28% 31%

Product 2 42% 100% 7% 8% 22% 6% 29% 22%

Product 3 31% 16% 100% 5% 10% 4% 18% 17%

Product 4 35% 29% 8% 100% 28% 7% 26% 12%

Product 5 47% 35% 8% 12% 100% 3% 37% 24%

Product 6 37% 66% 18% 19% 21% 100% 25% 21%

Product 7 45% 28% 8% 7% 23% 2% 100% 25%

Product 8 57% 24% 9% 3% 17% 2% 29% 100%

Probability that Product 8 is purchased given that Product 1 is bought is 31%

Probability that Product 1 is purchased given that Product 8 is bought is 57%

Increasing sales by creating cross-selling opportunities using MBA

Marketing Analytics

Customer Retention

To retain customers, it is very essential to keep tracking customers’ activity regularly — their frequency ofshopping, evolution of their shopping patterns, how often do they shop and so on. Customers attrite on adefinite path to inactivity which can be identified and therefore managed. Also, acquiring new customers hasbecome far more expensive than retaining existing ones and hence customer retention has become a majorcorporate priority. By employing attrition analysis, customers whose engagement levels have lowered andwho are likely to attrite can be identified and appropriate retention strategies can be formulated.

Churn analysis helps answer key business questions like: Which are the customer segments, to be targeted for retention programs? How do we identify the factors which are most likely to drive customers to remain:

• Creating segments based on preferences and buying patterns so that right offers can be made to theright people

• Understand the variables that make high-value customers most likely to purchase and offerincentives and personalized service

Attritio

n R

ate %

0%

10%

20%

30%

40%

50%

60%

70%

-

1

2

3

4

5

0 - 0.5 year 0.5 - 1 year 1 - 3 years 3 - 5 years > 5 years

# C

ust

om

ers

(MM

)

Risk & Fraud Analytics

Risk & Fraud Analytics

Risk Analytics covers all areas of the business that directly impact loan losses and charge-off’s.Marketelligent brings deep expertise in designing and implementing profit-based strategies that enable abusiness to limit its losses; at the same time not compromising on revenue opportunities.

Acquisition Credit Policy Pricing and Credit Line Management Collections and Recoveries Fraud detection and management Scorecard development, validation and maintenance across functions : approval, delinquency,

collections, etc

Behavioral Models Revenue and Cost Drivers

Optimal Line Determination Optimal Line Drivers

Balance Model

Revolve Model

Risk Model

CMV

LOCOptimal LOC

Revenue

LOC

Cost

LOC

Predicted V/s Actual Inactivity

0.00

50.00

100.00

150.00

200.00

250.00

300.00

0 50 100 150 200 250 300 350 400

Predicted Inactivity

Ideal

Actual

Predicted

Illustrative process for assigning Optimal Line of Credit (LOC)

Other Models

Business Situation :The insurance provider generated leads through cross-selling. Potential customers were targeted in a 4-stage process, and they generallydisplayed 4 possible outcomes:

Resources were being wasted on pursuing unlikely Prospects classified in red boxes above. The insurance provider wanted to determine whichmembers were more likely to complete at each stage , and then fast track the application through the approval process. By reducing theproportion of declined and incomplete applications, operating costs could be optimized.

The Task :- Develop a framework of predictive models to calculate the probability of a prospect purchasing the insurance product

- Get a more targeted base of Prospects, and hence reduce costs by removing prospects with least probability of buying the Product

Analytical Framework :- Historical data , which contained information from both, internal and external sources, was analyzed

- Logistic models were built for identifying separate probabilities for each stage of the approval process

- Testing was done if Oversampling or Undersampling would improve the performance of the predictive models

- All the models were then combined to identify the ‘best’ leads

- The models were validated and implemented as SAS Macros to enable real-time scoring

The Result :• The framework of 3 models provided useful insights on probabilities associated with approvals and important factors affecting it

• Up to 25% of total applications were removed with a loss of just 5% of Paid customers

• With costs going down by 25%, we were able to achieve an increment 14% net profits.

Client : An Insurance Provider in the US

Targeted

ProspectsApproved

Completed Forms

Paid Premium

Didn’t PayDid notDeclined

Target : 100

Approved : 85

Completed: 68 Paid : 58

Predictive Models

Target : 75

Approved : 70

Completed: 62 Paid : 55

Analytics in ActionTargeted Prospecting. Increasing Profits by 14%.

Business Situation:

The client, a South-east Asia based Retail Bank encountered a significant increase in customer churn on their Card Portfolio despite having a

tried and tested loyalty program in place. This resulted in a 4.6 % drop in Balances during Q2 2012. The business wanted to monitor and

control Customer churn at regular intervals.

The Task:

To develop and implement a program that monitors Customer engagement levels and attrition risk, measure business impact from Customer

churn, and develop actionable strategies to manage Customer Attrition.

Analytical Framework:

High value customers that left the business impacted Balances significantly. Segments were developed to slot each high-value Customer on

the basis of recent purchase patterns. Movement of Customers across segments and over time was used to identify the level of ‘engagement’

the Customer had with the Business. Segment-specific offers and campaigns were implemented to manage customer attrition. Results from

the campaigns were used to continuously refine targeting and messaging.

The Result:

• Based on the analysis, the Business was able to identify high value Customers at risk of attrition. Suitable Retention programs were

designed and implemented for these Customers.

• Business was able to more efficiently utilize its Retention budget as targeted customers consisted of only 15% of the overall customer base

• Balances in Q4 2012 were up by an average of 2.1 % as compared to the previous two quarters.

Analytics in ActionProactively Retaining your most Valuable Customers

Client: A Leading South-east Asian Bank

10%

39%21%

32%39%

24%30%

6%

# of customers Revenue contribution

0%

25%

50%

75%

100%

High Medium-High Medium-Low Low

Business Situation :The Client - a provider of short-term payday loans - was experiencing high default rates in its loan portfolio. Payday loans are instant smallvalue, short term loans. These loans are received by the borrower and are to be returned back on their next payday along with requisite feesto the lender.

Default rates were highly seasonal and geography specific.

The Task :- Develop and implement a ‘First Payday Default (FPD)’ Risk Model. The model will be used at the point of Acquisition to screen out Prospects

with a high probability of first payment default.

- Refine Acquisition Business strategy across key functions – pricing, loan amount and loan term.

Analytical Framework :A 3-step analytical process was used:

1. Segmentation of Customer Base: Segmented Customers based on loan performance, demographics and credit bureau profile. Identifiedand tagged High/Medium/Low FPD Customer Segments

2. Building a Default Scorecard: Identified Customers that have defaulted by Segment. Built predictive models for each segment. Modelidentified Customers most at risk of defaulting in the first payment cycle

3. Building a holistic Strategy for execution: Created a strategy grid for executing business strategies. Recommend appropriate treatments byStrategy grid

Segmentation Predictive Modeling Business Strategy

The Result :The New FPD model was implemented in a robust test-control mode and results tracked:

• Geographic location of the Prospect was identified as a significant segmentation variable. 6 geographic segments were created and a FPDmodel built for each segment.

• Seven credit bureau variables were identified as significant in identifying potential high-risk Prospects. Of these, ‘# inquiries in past 30 days’and “# loans given in past 12 months’ were identified as most significant.

• Once implemented, the new predictive FPD models helped lower portfolio loan losses by 19%.

Analytics in ActionLowering Portfolio Defaults by 19%

Client : A Consumer Finance Company making short-term Unsecured Loans

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%Random New Model

% D

efa

ulters

Cap

ture

d

% Customers

Approval Model

Conversion Model

FPD Model

5+ cycle Risk Model

180 day Revenues

Reactivation Model

Acquisition

Risk

Revenue

• Individual Scores

• Strategy Matrix• Joint Scores

Old Model

10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Business Situation:

Being a service provider of online remittances, the client faced significant problems with fraud and money laundering. A lot oftime was being invested in manual review of all transactions, which in turn delayed the overall transaction processing time,and also impacted customer satisfaction.

The Task:

To develop a process where only high risk transactions are sent for manual review, and rest are approved automatically.

The Analysis:

• Transaction history, compliance, service data along with external data sources like SSN responses and AMLOCK (Anti-Money Laundering Database) data for high risk Customers was leveraged

• Critical variables defined that would serve as inputs to the Risk Score were divided into 5 broad categories - Geographic,Historic, Identity, Transaction and Demographic - to make it easier to understand the reason for high risk during manualreview process

• Along with these variables some rules were also created based on compliance policy and Government regulations, whichwhen triggered automatically moved transaction to manual review irrespective of the risk score

• Multiple linear regression was performed to arrive at a transaction-level Risk score, and 4 Risk classes werecorrespondingly defined (<25: Low, 25-50: Moderate, 50-75: High, 75-100: Extremely High)

The Implementation:

• A schedule of data extraction was setup to generate the Risk score before the KYC forms were prepared.

• The process was streamlined to ensure that all KYC forms carried a Risk score

• The Risk score and Risk class were populated on the KYC sheets

• Compliance committee continues to give it’s opinion on each KYC form (Low Risk to Extremely High Risk) and thepredictive model continues to be bootstrapped

The Result:

• The new Process reduced number of transactions reviewed manually from 700-800/day to 200/day, and also reduced theaverage turnaround time for transactions from from 3.0 to 1.5 days.

• The process also helped the compliance team to identify which documents/clarifications to get from customer to processthe transaction.

Analytics in ActionTransaction-level Risk Assessment

Client : A Leading Service Provider of Online Remittances

Business Situation :The Credit Card issuer was facing significant losses due to fraud in spite of having a real-time transaction scoring application. It flaggedsuspicious transactions and declined them with a large false positive ratio, leading to bad customer experience and attrition. There were alsogaps in the process of identifying fraud with good accuracy due to constraints within the Fraud Operations group.

The Task :- Develop optimized authorization rules that efficiently capture fraud with minimum impact on genuine transactions.

- Revisit the transaction scoring mechanism and suggest a methodology that is more customized for certain type of transactions. Retire non-performing authorization rules.

- Develop new strategies/rules to better identify fraudulent transactions.

- Measure ‘agent performance’ in Fraud Operations Queues and suggest areas of improvement to Fraud Operations.

Analytical Framework :A 5-step analytical process was used:

1. Historical card transaction logs, data on confirmed fraud cases for the past two years, Credit Bureau data and Card Association notificationswere leveraged for the analysis.

2. For Customer attrition analysis, accounts with a minimum 12 months-on-book at the time they were wrongly queued/declined were used,their attrition rates, were then measured using a test-control approach.

3. Apart from an existing real-time fraud scoring engine, further analysis was conducted to identify domestic and international fraud hotspots.

4. Detailed segmentation was carried out on recent transactions to segregate fraud population with least impact on non-fraud population –this led to new authorization rules.

5. Developed automated MIS reporting measuring existing rule performance, agent performance in Operation queues and a framework toretire non-performing rules.

The Result :• Fraud Detection rates improves by 70 basis points Year-on-year after implementing the new scoring layer incorporating localised scoring.

• Fraud Operations agents false positives and false negative ratios (identifying true fraud and non-fraud respectively) improved significantlywithin 2 months of implementing Decision Quality framework. Missed dollar opportunity due to not identifying true fraud reduced by 15%Year-on-year.

Analytics in ActionCombating Credit Card Transaction Fraud

Client : A Credit Card Issuer facing high Transaction Fraud

MANAGEMENT TEAMGLOBAL EXPERIENCE.

PROVEN RESULTS.

Roy K. CherianCEORoy has over 20 years of rich experience in marketing, advertising and mediain organizations like Nestle India, United Breweries, FCB and FeedbackVentures. He holds an MBA from IIM Ahmedabad.

Anunay Gupta, PhDCOO & Head of AnalyticsAnunay has over 15 years of experience, with a significant portion focusedon Analytics in Consumer Finance. In his last assignment at Citigroup, he wasresponsible for all Decision Management functions for the US Cardsportfolio of Citigroup, covering approx $150B in assets. Anunay holds anMBA in Finance from NYU Stern School of Business.

Greg FerdinandEVP, Business DevelopmentGreg has over 20 years of experience in global marketing, strategic planning,business development and analytics at Dell, Capital One and AT&T. He hassuccessfully developed and embedded analytic-driven programs into avariety of go-to-market, customer and operational functions. Greg holds anMBA from NYU Stern School of Business

Kakul PaulBusiness Head, CPG & RetailKakul has over 8 years of experience within the CPG industry. She waspreviously part of the Analytics practice as WNS, leading analytic initiativesfor top Fortune 50 clients globally. She has extensive experience in whatdrives Consumer purchase behavior, market mix modeling, pricing &promotion analytics, etc. Kakul has an MBA from IIM Ahmedabad.

ADVANCED ANALYTICAL SOLUTIONS

MARKETELLIGENT, INC.80 Broad Street, 5th Floor, New York, NY 10004

1.212.837.7827 (o) 1.208.439.5551 (fax) info@marketelligent.com

CONTACT www.marketelligent.com

Industry Business Focus Tools and Techniques

Consumer Finance Investment Optimization SAS, SPSS, R, VBA

Credit Cards Revenue Maximization Cluster analysis

Loans and Mortgages Cost and Process Efficiencies Factor analysis

Retail Banking & Insurance Forecasting Structural Equation Modeling

Wealth Management Predictive Modeling Conjoint analysis

Consumer Goods and Retail Risk Management Perceptual maps

CPG & Retail Pricing Optimization Neural Networks

Consumer Durables Customer Segmentation Chaid / CART

Manufacturing and Supply Chain Drivers Analysis Genetic Algorithms

High Tech OEM’s Supply Chain Management Support Vector Machines

Automotive Sentiment Analysis

Logistics & Distribution

YOUR PARTNER FOR

DATA ANALYTICS SERVICES