6. Our People: Empirically Driven Consultants Our Portfolio : Full BI & Analytics-chain
Led by experienced technocrats Extract – Data capture forms, Data marts & Data
Team with deep business experience reconciliation
Sourcing is based on stringent 7i filters Monitor – Performance monitoring tools,
Continuous learning Dashboards, Data cubes, Basic analytics
Development centers in Bangalore and Hyderabad Predict – predictive and forecasting models for
strategic planning
Our Processes : Flexible & Efficient Our Platforms : Technology agnostic
Ability to balance analytical prowess with pragmatic Simple, Scalable and Robust
business application Capable of handling offline Excel files to complex
Driven by customers’ business objectives databases
Proven record of delivering step-up change in business Experience of working on varied source systems
KPIs and P&L lines Our platforms integrate seamlessly with existing
Demonstrated ability of high-octane delivery infrastructure
6
11. Our Delivery Model allows Clients to Flex both Quality and Cost Levers, while sourcing globally
Quality Lever (On-site/In-house)
Requires senior analytics staff/domain
experts
Quality Lever
6 1 Advanced education required
Deploy Define Ability to interact with stakeholders
Results Business
Objectives
5
ONSITE Evaluate &
Iterate Results
OFFSHORE
CLIENT
2
Analytical
Design & Data
Selection
4
Modeling & 3
Analysis Data
Preparation
Cost Lever
Cost Lever (Offshore)
Requires large volume of data work
Repetitive tasks, easily productionalised
Rules based
Time Consuming (60-75% of total time spent)
11
12. Our analytical process is driven by customers’ business objectives
Service Delivery Model ‘Key Steps at Each Stage
1
Define • Develop clear problem statements and
Business performance metrics
Quality Lever Objectives • Understand larger business and market context
6 1
Deploy Define 2 • Develop analytical solution and establish key
Results Business Analytical hypothesis
Objectives Design & Data • Identify data sources and validation sources
Selection • Establish availability of key data
5
3 • Clean and merge data using visual and statistical
Data methods
Evaluate &
• Create Meta data ,construct new variables
Iterate Results Preparation
CLIENT • Perform high level reconciliation of the data
2
Analytical
Design & Data 4 • Iterate different modelling alternatives and
Selection Modeling & evaluate fir vs. objectives
Analysis • Recommend one model with key assumptions
4
Modeling & 3
5 • Pressure test and validate the model
Analysis Data
Evaluate & • Iterate results to improve accuracy
Preparation
Iterate Results • Agree new analysis requirements / priorities
• Translate model results into tangible business
Cost Lever impact
6
• Identify process changes required to implement
Deploy • Implement the model/solution
ONSITE Results • Monitor for accuracy and performance
OFFSHORE
12
16. Presentation Layer
Customizable dashboards 1. High degree of user customization
Privilege on the Presentation Layer
Manager permissible
Report Automation Engine Administrator 2. Information privilege mirrors
Generates reports basis rules set Control
organization structure
3. High level of administrator rights –
Analytical Engine on rules, formats and access
Define business rules and triggers for monitoring and
reporting
4. Post implementation, our
involvement needed only if data
Benefits
sets or rule dimensions need to be
Analytical Data Mart altered
Consolidating & reconciling data from disparate
sources and mapping onto Logical Data Model
5. Cost of scaling up for more data
Data Source sources is marginal & proportional
Auto Auto Auto Manual
16
17. Onsite Offshore
Define Business Analytical design & Development &
Data preparation Evaluate & Deploy
Objectives Data selection Coding
…. and governed by 3 pillars of strength
Prescriptive requirements Pre-built modules Embed thru Technology
• Ability to understand business • Fundamental Sciences - • Tools & Applications to hardwire
requirement and context, quickly Statistics, Econometrics, Ops analytics in day-to-day ops
• Proactively think through cascading Research • Tools : SAS, SQL, VB, .NET, C++
and x-functional impact • Techniques - Predictive • Dbases: Oracle, MS SQL, MS
• Quick solutions to issues Modeling, Forecasting / Access, DB2
Simulation, Optimization
• Min time from client on briefing • ETL Tools : SQL server, SAS
• Pre built data adapters ETL, Informatica
• Pre-configured KPIs , dashboards &
reports - rapidly customizable
17
18. Business Solutions developed with business needs as focus
Solutions Addresses functional issues and operational challenges
Pre built data adapters to crunch time and cost
Quick Delivery Pre-packaged metrics & dashboard templates
Well defined Requirements documents
Data format agnostic – works with data dumps from core systems &
Simple and other offline sources
Scalable Light and low-cost IT infrastructure
Fully customizable
Senior management has a “dashboard” view
Cuts across org. Functional executives have “drill-down” view
structure Business analysts have a “scratch-pad” view
Automated generation, transmission and distribution
One version of Detailed reconciliation across GL, Risk and business
the TRUTH Well defined sign off processes
18
25. 1. Wealth Modeling 2. Customer Lifetime Value
Customer Lifetime Value (CLV) is long term and dynamic value that
•Identify the customers with the potential to be
can help you optimize your decisions for long term profitability
upwardly mobile (to migrate) through this segment
scheme to help drive product development and Average
Profit
portfolio actions Optimum Short Term Strategy
• The Near Term Wealth Score will assess how
close a customer is to the target wealth profile Long Term Strategy
within their given life stage. The higher the Customer B
score, the more closely they resemble the target Over a longer period customer B
wealth profile. is more profitable than customer
A
• The Lifetime Wealth Score will assess how close
Customer A
a customer is to the ideal target wealth profile
across all life stages. The higher the score, the
more closely they resemble the target wealth
profile.
Today Time
120%
100%
80%
3. Retention Modelling
60%
The models capture 40%
Models that identify those customers most likely to close their 40%
of the potential attriters in
accounts and Triggered Based Retention strategies 20% the first two deciles.
0%
0 1 2 3 4 5 6 7 8 9
Random% Closed %
25
26. Objectives Results
• Identify Upwardly Mobile Customers: Identify the customers • Developed five Near Term Wealth Score models, one for each of the life
with the potential to be upwardly mobile (to migrate) through this stages
segment scheme to help drive product development and portfolio • Model for Life stage 1 has a maximum KS of 80%
actions • Life stage 4 is chosen as the ideal “wealthy” profile for the entire
• Improve Value Understanding: To develop a value profile for portfolio
the different customers and segments. • Model for Life stage 4 has a maximum KS of 78%
• Develop a Reusable Segmentation: Develop a segmentation • The top two deciles in Value model capture 87% of total value which
that can be reused globally gives good separation
• There are sixteen actionable segments based on Wealth score and Value
score
Approach Business Impact
• We developed two different scores:
1. A Near Term Wealth Score and
2. A Lifetime Wealth Score • Targeted marketing based on Wealth profile and Value profile
• The Near Term Wealth Score will access how close a customer is to
the target wealth profile within their given life stage. The higher • Clear strategies can be drawn to move customers from “Mass” to
the score, the more closely they resemble the target wealth “Advanced” and “Premier” segments based on scores
profile.
• The Lifetime Wealth Score will access how close a customer is to • Plug and Play SAS codes
the ideal target wealth profile across all life stages. The higher the
score, the more closely they resemble the target wealth profile.
• Developed historical data model that provides monthly account
level profit estimates. These estimates are then converted into
Value Score
• Developed a two dimensional segmentation based on Wealth
Score and Value Score
26
27. Designed an approach that will measure the wealth potential of a customer both within the lifestage
that the customer is in and across all lifestages
We broke down the
1 portfolio into 5
different lifestages Lifestage
based on age
L1 L2 L3 L4 L5
High (Premier)
2
For the Near Term Wealth
models, we defined target
wealth customers for each of
Wealth
the different lifestages 4
For the Lifetime
Wealth Scores, we
established the ideal
The Near Term Wealth Scores target wealth profile
provide a measure of how
close a customer resembles
the target wealth profile in their
The Lifetime Wealth Scores
life stage
provide a measure of how close
a customer resembles the ideal
target wealth profile across all
3 lifestages
Low (Mass) 5
27
28. Lifestages
Model Variables Below 25 25-35 35-45 45-60 Above 60
Average over 6 months - ATM Transactions × × ×
Average over last 6 months - Number of TD Transactions × × × × ×
Average over last 6 months - Outgoing EFT Trans Amt × × × × ×
Average over last 6 months - Total CA Balance × ×
Education × ×
Professional Group × × × ×
Residential Status × × × ×
Revolver Segment × × × × ×
Total # of products × × ×
Transaction Band ×
• TD Transactions, EFT Transactions, and Revolver Segment are the three variables
that are significant in all the Lifestage segments
• Education and Transaction Band are the least significant across all Lifestage segments
28
29. The scorecard was used along with other criteria in creating multi dimensional segmentation. Usage
and activation strategies were based on this segmentation.
Life stage Probability band
0-0.4 0.4-0.7 0.7+ Missing Overall
% of Customers 7.30% 0.56% 0.24% 0.01% 8.11%
% of profit -4.09% -0.38% -2.06% -0.03% -6.57%
<=0
Avg. Balance A1 2,032.2 A215,035.3 A3 45,737.3 A4 35,271.4 21,564.7
Strategies
Avg. spend 2,611.3 16,158.0 49,492.7 35,513.1 15,066.6 developed to
% of Customers 16.63% 01.65% 0.43% 0.05% 18.77%
move
% of profit 1.14% 0.10% 0.03% 0.00% 1.26% customers to
0-40
B1 1,068.1 B2 2,648.2 B3 B4
Avg. Balance 12,162.9 1,997.9 3,307.0 higher value
Avg. spend 2,264.2 4,171.8 14,375.1 1,997.9 4,257.6
bands
Value band
% of Customers 53.01% 2.92% 1.96% 0.04% 57.93%
40-500
% of profit
C1 33.52% C2 1.57% C3
1.33% 0.01%
C4 11,105.8
36.43%
Avg. Balance 2,186.4 5,516.1 19,850.9 6,045.6
Avg. spend 4,801.6 10,890.0 25,464.4 11,105.8 8,707.7
% of Customers 11.51% 0.85% 1.79% 0.00% 14.15%
% of profit D1 47.48% D2 4.60% 16.76% 0.03% 68.87%
500+
D3 D4
Avg. Balance 9,314.5 23,797.9 106,071.5 163,365.1 55,780.2
Avg. spend 13,337.3 33,074.8 118,127.3 163,365.1 58,874.2
% of Customers 0.79% 0.16% 0.07% 0.00% 1.03%
Missing
% of profit 0.00% 0.00% 0.00% 0.00% 0.00%
Avg. Balance 18,932.5 19,108.2 52,394.9 2,143.5 36,503.8
Avg. spend 18,603.6 20,792.6 53,633.4 2,143.5 37,380.0
% of Customers 6.16% 4.49% 0.11% 100.00%
Overall
89.24%
% of profit 5.89% 16.05% 0.01% 100.00%
78.05%
Avg. Balance 12,068.9 69,323.5 36,385.8 24,744.9
3,909.6
Avg. spend 17,959.4 77,900.2 36,440.9 25,125.2
6,291.9
29
30. Objectives Results
• The Bank was using Behavior Segments/Scores to drive • 95% of the value comes from two segments that have
the portfolio management strategy. This strategy focused 24% of the accounts
on incremental lifts in response rates, with no insight/control • 9% of the accounts destroy 27.5% of the value
on the profitability of the customers
• Two thirds of the accounts are neutral in value
• Develop algorithms & easy-to-use interfaces that calculate
account level profitability based on forecasted revenue/cost
• The Least Value Contributor is Transactor and not
drivers and use these outputs in conjunction with behavior High Loss group
segments/scores to drive profitable portfolio growth
Approach Business Impact
• Forecasting assumptions used on pre-defined
segments • Detailed analyses of CLV drivers can help in designing
• Forecast revenue drivers instead of actual P&L line of campaigns to maximize value
items • Leverage historical value data and CLV index for
• Vintage based forecasting approach balance building activities
• Seasonality of revenue drivers built into the forecasting • Leverage historical value data and CLV index for the
methodology evaluation of credit line strategies and for determining
• Event based cost allocation methodology new opportunities
• Attrition and delinquency handled using probabilistic
rates
30
31. Vintage based forecasting engine is the cornerstone of this architecture. This methodology provides the
granularity that is required to achieve accuracy and consistency.
Functional
Forms from
Historical account Normalized
level data – Portfolio Vintage Curves
KPIs and detailed & Forecasting
revenue lines assumptions for
pre-defined
Segments
Forecasting
Engine CLV
Segmentation
Framework Forecast Output
for Revenue
Drivers
Revenue Drivers –
Key Portfolio
parameters which Calculation
drive portfolio P & L Algorithm
for creation of account
level P&L
(SAS code)
31
32. Average Spends Average Balance Average Revenue Limit Camp Accounts Average CLV
CLV DRIVERS 7955 15732 509 27%
489 YTL
High Value
CLV RESISTORS Average Cost of Funds Average loss amount Average cost
244 0.02 -20 In depth understanding of what is
driving different levels of CLV on
two products
Average Spends Average Balance Average Revenue Limit Camp Accounts
CLV DRIVERS 10,409 13323 12 10%
Loss Makers
Average cost of Funds Average loss amount Average cost Average CLV
CLV RESISTORS -235 YTL
209 228 -17
32
33. Less than 20% of Accounts contribute more than 90% of total profitability
Profit Contribution by Decile (%)
Top 20% contributes
100%
97% of total profit
80% Bottom 40%
destroys 25% of
60%
total profit
40%
20%
0%
1 2 3 4 5 6 7 8 9 10
-20%
-40%
33
34. Segmentation is built based on value and spend groups
Value SPEND + VALUE
Segmentation
GROUPS
Low Spend(0 High
to <=3000) Spend(>3000)
Loss Makers Loss Makers
Marginal Marginal
Low Low
Medium Medium
High High
34
36. Objectives Results
• To identify those customers that are likely to close their • For each behavior segment, we have identified possible high level
cards. marketing strategies that address the key customer opportunities
• Perform behaviour segmentation based on their
likeliness to attrite.
• For each behavior segment, Identify high level
marketing strategies that address the key customer Summary Gains Table
concerns and issues. Number of Cumulative Marginal
Non KS Prob
Deciles Non Non Closed Closed
Closed Closed Closed Non Closed Rate (Closed)
Closed Closed % Rate
%
0.0% 0.0% 0.0%
Approach 0 575 4444 575 4444 20.7% 9.4%
11.3
%
11.5% 88.5% 11.5%
16.6
1 416 4603 991 9047 35.7% 19.1% 8.3% 91.7% 8.3%
%
All 19.6
2 357 4662 1348 13709 48.5% 28.9% 7.1% 92.9% 7.1%
Cards %
20.2
3 293 4726 1641 18435 59.1% 38.9% 5.8% 94.2% 5.8%
%
20.3
All Active All Inactive 4 280 4739 1921 23174 69.2% 48.9% 5.6% 94.4% 5.6%
Activity Breakout %
Cards Cards 17.8
5 212 4807 2133 27981 76.8% 59.0% 4.2% 95.8% 4.2%
%
15.2
6 210 4809 2343 32790 84.4% 69.2% 4.2% 95.8% 4.2%
%
Shop & All other 11.0
Product Breakout Bonus 7 168 4851 2511 37641 90.4% 79.4% 3.3% 96.7% 3.3%
Miles products %
8 144 4875 2655 42516 95.6% 89.7% 5.9% 2.9% 97.1% 2.9%
9 122 4897 2777 47413 100.0% 100.0% 0.0% 2.4% 97.6% 2.4%
1 2 3 4
Models These figures show the cumulative percentage of cards.
Here 36% of the attrition has been captured within the first
two deciles.
36
37. The process we follow considers an exhaustive list of independent variables to make sure that
predictive power of the model is maximized.
1 2 3 4 5
Data Variable Model Scorecard
Validation
Validation Selection Building Development
We started with over We reduced the We then ran stepwise We then validated the We then developed
300 variables in the variable set down to regression to models based on the an appropriate
modelling universe. about 60 based on determine the final statistical results. scorecard.
the bivariate analysis variables in the
We conducted
and the overall model.
bivariate and
information value.
univariate analysis for
the categorical and Then we conducted
continuous variables correlation analysis
to make sure the and eliminated any
trends were correct. variables that were
highly correlated.
37
38. • Modelling has been done
Open Cards Closed Cards at a card level
55% sample 85% sample
Data Used
Customer Credit Burearu
Card Usage EFT Data Revenue Data
Product Holdings Data
Current Month 11 Months 11 Months 11 Months When Pulled
Customer Transactional Authorisation
Call Center Data
Demographics Data Data
Current Month 11 Months 11 Months 11 Months
Customer
Complaint Data
Current Month
38
39. Parameter Description of variables Bonus Shop & Miles Other
Total volume of non instalment purchases in last 3
AMT_SPEND_3M_6mnths
statement periods x x
woe_age_band The age of the card x x x
The value of transactions done in home improvement
woe_AMT_HOUSEWARE_3_6mnths
category lst 3 months x
woe_AMT_SPEND_3M_Ratio Change in spend over the past six months. x
Ratio of spend in Bonus network / total spend (as volume of
woe_AMT_WEB_RATIO_6_3mnths
transactions) x x
woe_ASSETS_TOTAL_6mnths Average YTL value of all assets in bank last calendar month x
woe_ASSETS_TOTAL_CURR_3mnths Total current YTL value of all assets in bank x
woe_ASSETS_TOTAL_CURR_ratio Total assets change in past six months. x
woe_BHVR_SCORE_CURR_6mnths Last calculated behaviour score (scores calculated monthly) x
Whether the customer has been or is enrolled in a spending
woe_BNS_PROM_FLAG
commitment for Bonus card x
woe_CURR_CUST_LIMIT_A_ratio Current available customer limit x
woe_CURR_DEBT_6mnths Customer's Current outstanding balance total (of all cards) x
woe_CURR_DEBT_ratio Current debt change in past six months. x
Total new transactions in last statement / Maximum total of
woe_LAST_PUR2MAX_PUR
new transactions in last 6 statements x x
woe_limit_band Limit of product x x
woe_mob_band The month on book group that the card is in x x
A flag that indicates if the owner has multiple cards with
woe_multi_card_flg
Garanti x
Whether customer is payroll customer and receives salaries
woe_PAYROLL_FLAG
in current account x
woe_segmentation The business segment that the card is in. x x
39
40. A heat map was created based on the scorecard. Clear actionable groups were identified and
appropriate strategies were designed.
• % of Closed Cards Number of Cards
VIP VG1 VG2 HS Others TOTAL VIP VG1 VG2 HS Others TOTAL
High High
38 4,931 3,976 73 11,058 20,076
Attrition 3% 8% 8% 14% 8% 8% Attrition
0% 10% 8% 0% 22% 40%
Risk Risk
Med Med
163 5,413 3,844 423 5,215 15,058
Attrition 8% 4% 4% 6% 5% 5% Attrition
0% 11% 8% 1% 10% 30%
Risk Risk
Low Low
339 5,887 4,406 2,104 2,321 15,057
Attrition 3% 3% 3% 3% 3% 3% Attrition
1% 12% 9% 4% 5% 30%
Risk Risk
540 16,231 12,226 2,600 18,594 50,191
TOTAL 4% 5% 5% 4% 7% 6% TOTAL
1% 32% 24% 5% 37% 100%
• 10% of the cards • 11% of the cards
• Take more urgent proactive measures to • Take moderate proactive measures to
ensure that these customers are happy strengthen the relationship
• 26% of the cards • 17% of the cards
• Stay focused on BAU activities and • Take moderate proactive measures to
promoting the benefits of the product reinforce use of the product
40