1 Getting Blood From a Stone: Reducing Credit Card Delinquency Rates Roger Hoerl General Electric R&D
1
Getting Blood From a Stone: Reducing Credit Card
Delinquency Rates
Roger HoerlGeneral Electric R&D
2
Outline
• Why the big deal over the bottom line?• How Six Sigma addresses the bottom line• A Six Sigma project in collections• Why we are missing this opportunity, and
what can be done about it• Summary
3
Why the Big Deal Over the Bottom Line?• Despite what Deming preached, the vast majority of
business leaders believe that they are in business to make money (societal and legal expectation)
• Even Deming pointed out that a company needed to make money to fulfill its mission
• Many improvement initiatives have been viewed as “feel good” cultural movements that are disconnected from, or even oppose, making money
• To be taken seriously in business circles today we have to clearly and convincingly demonstrate that our recommended approach directly improves the bottom line (“Put up or shut up”)
4
How Six Sigma Addresses the Bottom Line
In good companies:
• Projects are selected based on estimated financial benefits (among other things)
• Financial results are required for the BB or MBB to become certified
• Both estimated and actual financial benefits are determined by the finance department
• Benefits must actually show up on the bottom line to be counted (no “soft” savings)
• The financial numbers are audited by independent parties
5
Collections Project
• Majority of work done by GE Card Services (CS) Black Belt team
• DOE led by GE Corporate Audit Staff (CAS)
• Following pages from CAS project summary (confidential details deleted)
• Key challenge – communicating statistical concepts to financial leadership
6
CS - IntroductionUnderstanding The Need ...
0%
10%
20%
30%
40%
<$25M
$25-50M
$50-100M
$100M<
19921995
Household Income
Income level
“Personal Bankruptcy is at an all-time high” - Washington Post, 12/08/96
Consumer Credit
“The expansion of credit has led to all but the richestAmericans spending a greater share of their household income on consumer credit” - Washington Post, 12/08/96
Source: Federal Reserve
GE Capital Card Services is the world’s largest providerof Private Label Credit Cards (PLCCs) with over $15B inreceivables and over 70 million cardholders.
CS Vision:
“To be the most efficient, consistent, innovative, and highestvalue provider of retail financial services and consumer financialservices in the world”
Segment Key AccountsCS operates in every major retail segment:
»Department Macy’s, Wards, Disney»Home Improvement Home Depot, Lowe’s»Commercial Payless, Sams»Oil & Gas Exxon, Imperial Oil»Big Ticket Levitz, The Wiz
In Order To Understand The Issue
7
Note: Some sensitive data have been deleted Problem StatementStep 1Understanding The Process ...
With consumer delinquency on the rise, CS needs to adjust it’s business processes to increase the effectiveness and timeliness of collections. The general business process is as follows:
New CardApplicationNew CardApplication
SalesProcessing
SalesProcessing BillingBilling Payment
ProcessingPayment
ProcessingCust Svc
/ CollectionsCust Svc
/ CollectionsCard
ProductionCard
ProductionMarketingMarketing
CS has the ability tocontrol this
area
DialerCollections
ManualCollections
SkipTracing
Writeoffs
$ XX MM in annual writeoffs$ XXX MM in writeoffs YTDDialer, Manual, & Call Back Center Collections Skip Tracing
Customer will not payBankruptcyPotential legal
No contact informationInvalid contact information
Impact:Reason:
Call Back Center
Skip tracing is the process of locating customer contact information when the customer has become delinquent in making creditcard payments.
Problem StatementScope: Focus on core private label cardsCTQ: Timeliness & Effectiveness of CollectionsDefect: Non-callable number on skip accountOpportunity: Each Skip AccountCustomer: CS clients purchasing credit card services Cost Savings: Estimated $X MM annually
Since skip tracing costs are mostly paid by the client and collection writeoffs are shared, an effort to minimize tracing loss is required. Skip tracing associates need to prioritize which skip tracing tools are the most cost efficient and most effective.
To Effect Timeliness of Collected Dollars
8
Note: Some sensitive data have been deleted CTQ TreeDrilling Down On The Process ...
Need to identify root causes that can be controlled to yield improvement. Using a CTQ tree, one can start with what is important to the client and drill down to the sub-processes that needed to be changed\adjusted. The shaded boxes indicate the branches ofthe tree that are traversed for this problem.
In Order To Set Initial Scope
New CardApplicationNew CardApplication
SalesProcessing
SalesProcessing BillingBilling Payment
ProcessingPayment
ProcessingCust Svc
/ CollectionsCust Svc
/ CollectionsCard
ProductionCard
ProductionMarketingMarketingBusinessProcess
• Delinquency at highest level• 2nd highest collection CTQ• XX% of clients dissatisfiedTimeliness &
Effectiveness of Collections
Dialer Manual Call Back Center Skip Tracing
• Teams in place• Training Programs• Hired Consultants • Low Controllability
• Highest Opportunity• Loss Prevention• Controllable
Volume : XX MMAccts/month
Volume : XX MMAccts/month
CTQ
Level 1Process
Volume : XX MMAccts/month
Volume : XX MMAccts/month
Skip Tracing Time Meetings Dialer Time BreaksLevel 2Work Allocation
XX% XX% XX% XX%Time Allocation:
Skill Level Skip Reason Tool Efficiency TechnologyLevel 3Time Allocation Account Type
9
Note: Some sensitive data have been deleted CTQ Tree - Root CauseDrilling Down Further...
Having identified the root cause of the CTQ,one now needs some initial data to bettercharacterize the problem.
Timeliness & Effectiveness of
Collections
In Order Isolate Root Causes - Initial Focus- Secondary Focus
Dialer Manual Call Back Center Skip Tracing
Skip Tracing Time Meetings Dialer Time Breaks
XX% XX% XX% XX%
Skill Level Tool Efficiency Technology
• Teams in place• Training Programs• Hired Consultants • Low Controllability
Account Type
CTQ
X1 X3
X2
X5 X7 X8
X6X4
No Phone Wrong Phone
Wrong Area Code
Disconnect Dialer Training
App Process
Number Changed
Restricted Number
20.3%9.1 act/hr$0.01 /act
11.6%7.4 act/hr$0.02 /act
14.2%18.3 act/hr$0.67 /act
35.5%5.3 act/hr$1.00 /act
12.8%8.7 act/hr$0.01 /act
20.8%13.5 act/hr$0.01 /act
6.8%18.8 act/hr$4.50 /act
TestingEffectiveness:Cycle Time:
Cost:
Skip Reason
Time Allocation:
Data
Volume : XX MMAccts/month
Volume : XX MMAccts/month
Volume : XX MMAccts/month
Volume : XX MMAccts/month
Level 1Process
Level 2Work Allocation
Level 3Time Allocation
Note: Data suspect dueto non-random sampling!
10
ApproachHow To Proceed ...
This project is focused around optimizing the results of using skip tools to locate information on skip accounts. The solutionwill identify the best skip tool combination based on tool effectiveness, cost, and cycle time. The business problem can be transformed to a statistical problem of optimization. The statistical tool to perform optimizations is the Design of Experiments(DOE). Once a statistical solution is achieved, the results can be translated to a business solution.
Steps to DOE:
Define
Measure
Analyze
Improve
• Plan the experiment1. Define the business problem2. State the experimental objectives3. Define the input and output variables
• Gather data to characterize the process4. Determine quantity and type of data needed5. Collect data6. Verify data validity
• Setting up the DOE 7. Map data to DOE matrix8. Include all output variables9. Utilize statistical package for analysis, Minitab
• Design based on specifications10. Observe Main Effects plots11. Observe Interaction plots12. Optimize the results13. Transform statistical solution to solve business problem
Note: Used DOE to obtain unbiased data, and data on combinations of tools.
To Solve The Business Problem
11
Data CollectionsGathering Information On Tool Combinations
What Can Be Learned?
For each skip account:Data gathered with the following conditions:
• 100 randomly selected skip accounts• 2 due accounts used (30 days past due)• Accounts were not sent to pre-screen
• Every tool was tried on each skip account• Possible and verified numbers were recorded• Root causes were recorded• Characteristics of each account were noted
Analyze Interaction Between ToolsMost Effective Combinations
Understand Effect of Uncontrollable Factors Measure Drivers of Skip Accounts
The following tally sheet will be used to gather more data on how effective certain tool combinations are. The three business groups: XXX.
*Account Number X1 X2 X3 X4 X5 X6 X7
Ex. CG129472883495 V P V P P
"P" indicates potential phone number, "V" means actually verified asbeing correct.
Gathering Data To Optimize Tool Ranking Using DOE
12
Gauge Reliability & ReproducibilitySampling Measured Data ...
Why?
• Prove statistical validity of data collected• Ensure recommendation based on data
has merit• Eliminate the “gut” feel in the improvement
process
How?
• Take 30 random samples from measured data for each business group
• Have a different person evaluate the results for each business group
• Compare measured samples to actual occurrence
• Difference should be < 5%
Client 1 Client 2 Client 3
• 30 samples taken• 100% accuracy for Reason
account sent to skip• 99% accuracy for Tool
evaluation 208/210 correct
• 30 samples taken• 100% accuracy for Reason
account sent to skip• 98% accuracy for Tool
evaluation 177/180 correct
• 30 samples taken• 100% accuracy for Reason
account sent to skip• 100% accuracy for Tool
evaluation 150/150 correct
Poor Excellent Poor Excellent Poor Excellent
… To Validate Accuracy And Reliability
13
EffectivenessPerforming The DOE ...
V = Verified Number CodesTRY ALL TOOLS P = Possible Number Yes or No Below
Account Number X1 X2 X3 X4 X5 X6 X7 SSN PO Return Unlisted Reason73160057623 P P NFD NO#71130021463 V P P NO#70500192607 P NFD NO#71930029567 V V CRNT NO#70510242558 P P P WN71130042360 V V V V DT
Mapping the the data from the tally sheet to the DOE matrix, in order to calculate combination effectiveness.Home Depot Example
DOE Mapping Algorithm100 skip accounts
• Count number of accounts found by tool combination• Equate to effectiveness, since 100 accounts used• Continue until effectiveness for all combinations are calculated
%X1 X2 X3 X4 X5 X6 Effectiveness-1 -1 -1 -1 -1 -1 01 -1 -1 -1 -1 -1 38-1 1 -1 -1 -1 -1 111 1 -1 -1 -1 -1 42-1 -1 1 -1 -1 -1 361 -1 1 -1 -1 -1 55-1 1 1 -1 -1 -1 39
X1 found verified numbers for 38 of the 100 accounts, therefore resulting in a 38% effectiveness.X2 found verified numbers for 11 of the 100 accounts, therefore resulting in a 11% effectiveness.X1 and X2 found verified numbers for 42 of the 100 accounts, therefore resulting in a 42% effectiveness rate.
1
2
3
1
2
3
Examples
1 = Tool Used-1 = Tool Not Used
Calculating DOE Output For Analysis
14
Minitab Data EntryBuilding The DOE ...
Client 1 ExampleMinitab is used to perform the actual DOE analysis. The data is entered into a Minitab worksheet and then sorted foreffectiveness.
• Sorted by effectiveness, sinceeffectiveness is higher priority tothe CS.
• This data on it’s own is not enoughto fully complete the analysis. What is missing:
• How do the skip tools contribute to the overall effectiveness, cost, and cycle time?………………. Main Effects Plot
• How do the tools interact for the desired combinations?……Interaction Plots
Minitab generates these tool combinations
Setting Up A Minitab Analysis
15
Main Effects PlotsPerforming The DOE ...
Client 1 ExampleMain effects plots identify which tools contribute the most to the output. The steeper the slope of the plot for the particular tool the more it effects the output being measured. Therefore the best tools would have steep slopes with regards to effectiveness and relatively flat slopes with regards to cost and cycle time.
Understanding The Major Drivers
Examples
1
X6X5X4X3X2X1
1-1 1-1 1-1 1-1 1-1 1-1
55
50
45
40
35
Effe
ctiv
enes
Main Effects Plot - Data Means for Effectivenes
1
X1 X2 X3 X4 X5 X6
-1 1 -1 1 -1 1 -1 1 -1 1 -1 1
500
600
700
800
900
Cos
t
Main Effects Plot - Data Means for Cost
X1 X2 X3 X4 X5 X6
-1 1 -1 1 -1 1 -1 1 -1 1 -1 1
30
45
60
75
90
Cyc
le T
ime
Main Effects Plot - Data Means for Cycle Time1 2
2
3 3
3
From the Minitab Menu Bar:
Stat\DOE
\Main Effects Plot
1. X1 has the best effectiveness, but also the highest cost.2. X6 is the cheapest tool, has a medium effectiveness,
but is the slowest.3. X3 is a highly effective tool at a low cost anda low cycle time.
16
Interaction PlotsPerforming The DOE ...
Client 1 Example
Interaction plots reveal the relationships between tools relative to the CTQ. The two lines for each pair of tools represent a main effects type plot for the tool listed at the bottom when the tool to the left was excluded (labeled -1) and separately when it was included (labeled 1). The focus is on looking for lack of parallelism between the lines, which indicates interaction (i. e. tool x is effective if tool y was not used, but not very effective once tool y was used.) Tools with parallel lines behave independently of one another.
From the Minitab Menu Bar:
Stat\DOE
\Interactions Plot
X3 is very efficient when X1 is excluded (-1 line is steep), but somewhat less effective when X1 is included ( 1 line is not as steep).This implies significant overlap between X1 and X3.X3 and X5 behave independently ( -1 and 1 lines are parallel), so X5 effectiveness is not impacted by X3.
Examples
1
2
1
2
-1 1 -1 1 -1 1 -1 1 -1 1
304050
304050
304050
304050
304050X1
X2
X3
X4
X5
X6
-1
1
-1
1
-1
1
-1
1
-1
1
Interaction Plot - Data Means for Effic
1
2
Understanding The Tool Relationships
17
Tool Combination SelectionSelecting The Tool Combination ...
Client 1 Example
Considering all tool combinations resulting in greater than 58% effectiveness, a plot of the potential effectiveness resultsare plotted for cost and cycle time. The plot will allow for selection of the most suitable tool combination.
30
60%
58%58%
58%
60%60%
60%61%
61%
61%
61% 62%
I II
III IV
IdealQuadrant
QuadrantI
II
III
IV
Combinations in this quadrant would be ideal. This quadrant consists of both lowest cost and shortest cycle time.
Combinations in this quadrant would be second best,since low cost is still maintained.
Combinations in this quadrant would be third best,since low cycle time is still maintained.
Combinations in this quadrant would be a last resort,as both cost and cycle time are high.
This is the best tool combination considering effectiveness, cost,and cycle time. It yields 60% effectiveness while maintainingsignificantly low costs and an acceptable cycle time.
Resulting Tool Combination (order to be determined):
•X1•X2•X3•X6 This combination results in the best solution for
cost and cycle time, but yields only 58% effectiveness
A tool combination from DOE resulting in > 57% effectiveness
Effectiveness vs Cost & Cycle Time
900$
850
800Cost
(per 100accounts)
750
700
650
600
550
50020 40 50 60 70 80
Cycle Time(hours)
With The Best Overall Value
18
Tool RankingRanking The Tools ...
Client 1 Example
Tool ranking will be performed to achieve a 60% effectiveness using the tool combination X1, X2, X3, and X6.Based on the data analyzed thus far, the first tool to be used is the one with the lowest cost, X6. Since X6 hasa 21% effectiveness rate, this implies that 21 accounts would be found using X6 at a cost of $112 and a cycletime of 48 hours.
Number of Verified Total Verified Total TotalTool Accounts Accounts Accounts Cycle Cycle Cost CostX6 100 21 21 48 48 $112.00 $112.00
Next the most effective tool, of those remaining, at the lowest cost is selected, X3. Adding X3 to the tool suitewould promise to find an additional 24 accounts of the remaining 79 (100 minus 21). The cycle time and cost for the 79accounts is 6 hours and $139, respectively
X3 79 24 45 6 54 $139.00 $251.00
Number of Verified Total Verified Total TotalTool Accounts Accounts Accounts Cycle Cycle Cost CostX6 100 21 21 48 48 $112.00 $112.00
Based On Business Drivers
19
Tool RankingRanking The Tools ...
Client 1 Example
Next the most effective tool, of the remaining, at the lowest cost is selected, X2. Adding X2 to the suite yields anadditional 3 found accounts of the remaining 55 (100 - 21 - 24). The cycle time and cost for the 55 accounts is 3.2 hoursand $111.00, respectively
X2 55 3 48 3.2 57.2 $111.00 $362.00X3 79 24 45 6 54 $139.00 $251.00
Number of Verified Total Verified Total TotalTool Accounts Accounts Accounts Cycle Cycle Cost Cost
X6 100 21 21 48 48 $112.00 $112.00
Completing the matrix using X1 last for process and business reasons, the completed matrix is as follows:
Number of Verified Total Verified Total Total
X1 52 12 60 9.9 67.1 $280.00 $642.00X2 55 3 48 3.2 57.2 $111.00 $362.00X3 79 24 45 6 54 $139.00 $251.00
Tool Accounts Accounts Accounts Cycle Cycle Cost CostX6 100 21 21 48 48 $112.00 $112.00
39 unfound skip accounts to be sent to other sources
For Best Performance
20
ResultsRecommended Tool Ranking :
Client 1
X6 X3 X2 X1 Outsource1 2 3 4 5
Cycle Time (cumulative hours)
1.00
2.00
3.00
4.00
5.00
6.00
7.00
Cost
Efficiency
Cost perAccount
(cumulative cost)
$2.51
$3.62
$6.42
48 54 57.2 67.1
$1.12
70%
60%
50%
Efficiency40% (cumulative efficiency)
30%
20%
10%
0 0%X6 X3 X2 X1
Client 1 Tool Order
Optimizing Effectiveness, Cost & Cycle Time
21
Bottom Line Results• Estimated savings across CS - $2MM annually• About half in increased cash, half in reduced
write-offs• Senior management did not believe the estimated
savings, and asked for a 6 month trial• After 6 months, about $1.45MM in actual savings
had been documented - $2.9MM annually• Secondary benefit: analysis determined optimal
number of collectors – resulted in additional hires
22
Why Are We Missing This Opportunity?• We tend to have a non-financial culture
– A focus on making money seems “sleazy”– We sometimes value knowledge of the tools more than the
results gained from the tools– Many of us don’t understand finance or accounting
• What can we do about it?– Accept the fact that we have to make money to be successful in
business (its not money that’s “evil”, it’s the love of money!)– Adopt an approach analogous to Six Sigma – the purpose of the
improvement effort is to improve the bottom line– Take the time to justify projects financially– Take the time to verify the financial impact before declaring
success– Learn basic accounting and finance before you ask them to
learn statistics
23
Summary• The bottom line of this is, well, the bottom line!
– “Profit is not the explanation, cause, or rationale of business behavior and business decisions, but the test of their validity” (Peter Drucker)
– It is not immoral to make money in business!• Our long term success is dependent on clearly
taking improvement efforts to the bottom line• Part of the reason for the popularity of Six Sigma
is its focus on financial justification and validation of projects
• The benefits of the collections project were not believed by management until they actually hit the bottom line