Professional Documents
Culture Documents
Improve Cure Rate in Credit Card Collections
Improve Cure Rate in Credit Card Collections
Client Speak
‘We are looking at expanding our customer base and
Credit Ops needs to be geared up
Head, Collections
Client Speak
‘Customers looking for Delinquency managers rather than
collectors’
-2-
VOC to CTQ Drill Down
ard
VOC
vi ng fo r hi gh er m ar ket share of Credit C
We are stri
market
Customers require delinquency managers rather than collectors
Credit Ops un
its need to re
objectives align themse
lves with current organ
izational
Customer
Gross Credit Loss
SPECIFIC
Hardship/Special
Roll Back Rate Static Rate Cure Rate
Queue rate
-3-
Business House of Debt Collections- How ‘Cure Rate’ fits in Big Picture
Customer Debt
Output
Delinquent Inventory
Credit Losses
Operational Metric
-4-
Credit Card Collections Lifecycle
• Credit Card Collection revolves around Bucket structure. A bucket refers to a 30 day period from the time a
customer misses payment due date. There are total 6 buckets i.e. 180 day cycle
• Post each 30 day period based on the payment received, account either moves to next bucket (Roll Forward),
stays in the same bucket (Static) or moves back a bucket (Roll Back)
• If a customer clears the entire amount due on account, it is termed as a Cured account
• Schematic diagram below represents the lifecycle of account in Collections
CORE
Account Allocation Processing
Assess nature of System
business
Align collection strategies Initiate Write Off
based on ageing
Review notes and
Prioritize accounts
ageing
Allocate inventory
Initiate write off as per
policy
Process Information
Operational Definition of Metric
Domain Banking
Cure Rate
FTE Size 45
=
Amount of Inventory Resolved (0-30 days)/Amount of Total Opn 6:30AM – 4:30PM (IST) /
Inventory (0-30 days) Window 12:00 PM – 9:00 PM (AEST)
Type Collections
Product Credit Cards
MSA
Client Large Global Bank
Cure Rate M1 Continuous Inventory resolved and Monthly System Reports 31 Months
Total Inventory (Jan’12 to Jul’14)
8
Normality Check & Baselining
M ean
S tD ev
0.36862
0.04019
Normality: P value is greater than
V ariance 0.00162
S kew ness 0.683055
0.05
Kurtosis 0.833728
N 31 Shape: Normal
M inimum 0.30416
1st Q uartile 0.33939 Measure of central tendency : As
M edian 0.36479
0.34 0.35 0.36 0.37 0.38 0.39 Stability factor = 0.33/0.40 = 0.82
Inference:
• The P - value is 0.66, greater than 0.05, that indicates that the data considered for the project is normal &
the Process is following Normal distribution
-1-
Stability & Capability Analysis
I Char t of Cure
High value accounts got
0.50 1
Individual Value
0.40
_ The process in under control
X=0.3686
0.35 with exception of one instance
which has an assignable
0.30
cause as mentioned
LCL=0.2664
1 4 7 10 13 16 19 22 25 28 31
Observation
LSL
P rocess D ata W ithin
LS L 0.45 O v erall
Target *
USL * P otential (Within) C apability
S ample M ean 0.368617 Cp *
C PL -0.80
Capability
S ample N 31
S tD ev (Within) 0.0340558 C PU *
S tD ev (O v erall) 0.040189 C pk -0.80
O v erall C apability
Inference:
Pp *
PPL -0.68 As per Capability Analysis it is
PPU *
P pk -0.68 evident that current capability of
C pm * process is below customer
expectation
Process
Systems
Technology
21
Contactable customers)
Absence of analysis on Best Analytics team to NA
a agents (Aug’14)
Account level
time to call pick up study
a analysis
r
Quick Wins/Solution Identified Ownership
Deployed Partially Deployed/In Progress Not Deployed
S.N Quick Win Ownership Status Controls
o
1. Initiate Account level ownership xxxx Deployed Reporting structure included account level inputs with
agent name
2 • TLs to devote more time on coaching xxxx Deployed Automated tracker introduced to log activities and
agents to give acceptance on feedback sessions
• Introduction of subjective/situation based process tests
3 Ask Client to introduce a separate queue to call back on xxxx Not Request shared with client. Capacity not agreed upon
left messages Deployed
4 Ask client to introduce voicemail for Inbound calls xxxx Not Request shared with client. Systems team assigned task
Deployed
5 Ask customer to allow calling from unmasked numbers xxxx Deployed System change, no controls required
6 Dialer stamping on accounts which have not been dialed xxxx Partially System issue. Taken up with client IT team. Samples of
thereby reducing opportunity to collect rectified instances where attempts were counted without dialing
customer shared
7 Make customized MIS available xxxx Deployed Daily, Weekly and Monthly MIS changed
8 Build automated tools to track real time promises and xxxx Deployed Automated tool built to track real time performance. TLs
accounts touched to audit the correct usage
9 Ask client to allow pacing control on dialer xxxx Not Pacing cannot be given to team
Deployed
10 Shared data with IT team. More bandwidth to be allowed IT Team Deployed System change. No process control required
for process
11 Analytic study initiated to ascertain Best Time to call xxxx In Study in progress
customers to achieve Contacts Progress
Study of Relationship (Y vs X)
Scatterplot of Cure vs PTP Rate Scatterplot of Cure vs Contacts
0.50 0.50
0.45
r= 0.237 r 0.45
r= -0.036 r
Cure
Cure
0.40 0.40
0.35 0.35
0.30 0.30
0.90 0.95 1.00 1.05 1.10 1.15 3000 4000 5000 6000 7000
PTP Rate Contacts
r
Cure Rate
0.45
r= 0.088 0.45 r= -0.074 r
Cure
Cure
0.40 0.40
0.35 0.35
0.30 0.30
600 700 800 900 1000 1100 1200 1300 1400 1500 200 300 400 500 600 700 800
Talk Hrs
No Correlations Wrap Hrs
0.45
r= -0.184 r 0.45
r= -0.081 r
Cure
0.40
Cure
0.40
0.35
0.35
0.30
0.30
0.50 0.52 0.54 0.56 0.58 0.60 0.62 0.64
0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 Kept Rate
Contact Rate
Inference:
• No Correlation was observed between Key Input metrics and Cure Rate which was opposite to expectations
• Most surprising was almost zero correlation between Kept Promise Rate and Cure Rate
Study of Relationship between Multiple factors (X vs X)
Post observing No Correlation between key Input Metric and Project Y, an Input metric behavioral check was initiated to see if
they correlated in expected manner
Matrix Plot of Contacts, Talk Hrs, ... vs Contacts, Talk Hrs, ...
1
800 1200 1600 3000 4500 6000 0.04 0.08 0.12
Promises Kept
4
a
Desirable
7000 &
Contacts
5000 Expected
3000 1600
Talk Hrs
1200
3
800
5
Promises Taken
Wrap Hrs
600
400
200 2
6000
PTP
4500
3000
2
Contacts/Hr
3.2
2.4 Contacts
1.6
Not
PTP Kept Contact Rate
0.12
Desirable
3
0.08 r
4000
4 0.04
Wrap Hours
3000 1
2000
5 Talk Hours
3000 5000 7000 200 400 600 1.6 2.4 3.2 2000 3000 4000
Contacts Talk Hrs Wrap Hrs PTP Contacts/Hr Contact Rate PTP Kept
Kept Kept
Talk Time
Promise Talk Time Promises
s
Input Metric
Study
Promise
Promise Contacts
Taken
Contacts Taken
All Input metric talking to each other and
pieces of puzzle fit together
Cured Accounts
r
When Number of accounts Cured is
introduced the puzzle doesn’t get solved
Contacts Promises Taken
Solving the Puzzle- Exploring Static Inventory
Post Brainstorming team zeroed in on Static Inventory – Accounts where Minimum dues are collected which was not a key
number monitored until now. It was decided to collect data and draw Correlation with Kept Promises
0.5
Static Rate
0.4
Kept promises majorly cater to
collecting Minimum Balance due
0.3
Inference:
• Clear Correlation exists between PTPs Kept and Static Inventory
• This indicates that agents are focusing efforts on taking single payments
• This is due to KPIs and Incentive Model which rewards them on taking PTPs (Promise to Pay) and not on curing/resolving the
account
Statistical Validation of Kept Promises vs Static Inventory
Post viewing the correlation, a causation test was conducted to understand how Static Inventory varies with Kept Promises
0.6 Source DF SS MS F P
Kept Promises Range 3 0.22999 0.07666 36.84 0.000
Error 24 0.04995 0.00208
Total 27 0.27994
0.5
S = 0.04562 R-Sq = 82.16% R-Sq(adj) = 79.93%
Static Rate
0.4
Individual 95% CIs For Mean Based on
Pooled StDev
Level N Mean StDev --+---------+---------+---------
0.3 +-------
1500-2000 10 0.30800 0.02974 (--*--)
2000-3000 7 0.40286 0.04957 (--*---)
3000-4000 7 0.48429 0.02992 (--*---)
0.2
4000-4500 4 0.55750 0.08539 (----*---)
1500-2000 2000-3000 3000-4000 4000-4500
--+---------+---------+---------+-------
Kept Promises Range
0.30 0.40 0.50 0.60
Inference:
• ANOVA analysis clearly shows how Static Inventory moves with increase in Kept Promises
• Most of the promises taken from customers are for minimum balance payment i.e. making account Static
Puzzling Situation- Correlations Dilemma Solved
Kept
Talk Time Promises Kept
Talk Time Promises
Input Metric
Study
Promise
Promise Contacts Taken
Contacts Taken
a
Inventory Study
Static
The puzzle got solved moment
Static inventory was introduced
in the equation
Promises Taken
Contacts
Non Productive Hours Analysis
The work in any calling process is split into 2 major heads which are Call time and After Call work (ACW). The times were analyzed to ascertain where
majority of the time was being spent
39%
28%
21%
12%
Inference:
• 49% time spent on Wrap and Preview activities
• Reduction in the Non prod hours will enable agents to spend more time speaking to customers (Value Added)
thereby resulting in higher resolutions
Calls/Hr versus Cure Rate
Study of impact of Calls/hour on Cure Rate to determine if efficiency metric impact the Project metric
Inference:
Since P Value is > 0.05, Mean Cure Rate is not significantly different based on Calls/hour segmentation
Agent & Weekly Outbound Calls
Study of how number of calls varies by agents
0.325
Level N Mean StDev
0.300 40 and above 9 0.33222 0.04055
Between 20-30 18 0.33000 0.04472
0.275 Between 30-40 7 0.30857 0.03078
Less than 20 11 0.31727 0.05120
0.250
Individual 95% CIs For Mean Based on Pooled
StDev
40 and above Between 20-30 Between 30-40 Less than 20 Level +---------+---------+---------+---------
Process Score 40 and above (-----------*-----------)
Between 20-30 (-------*-------)
Between 30-40 (------------*-------------)
Less than 20 (----------*----------)
+---------+---------+---------+---------
0.275 0.300 0.325 0.350
Inference:
• Since P value is >0.05 we can conclude that the process knowledge for agents does not impact Cure performance.
CureC Rate V/s Customer Demographics
Study if Cure Rate gets impacted by type of employment of a customer Study if Cure Rate gets impacted by Status of employment of a customer
Chi-Square Test: Cured, Not Cured vs Type of Chi-Square Test: Cured 1, Not Cured1 vs
Employment (Salaried/Self Employed) Employment status (Employed/Unemployed)
Expected counts are printed below observed counts Expected counts are printed below observed counts
Chi-Square contributions are printed below expected Chi-Square contributions are printed below expected
counts counts
Inference: Inference:
Since P value is >0.05 we can conclude that the Type • Since P value is <0.05 we can conclude that the Employment
of employment (Salaried/Self Employed) does not impact status (Employed/Unemployed) does impact
cure performance Cure performance
• Segregate inventory by employment status
• Assign better negotiators to such inventory
Summary of Analysis (1 of 2)
Analysis Tool Used Inference/Observation Recommendation
Brainstorming to Ishikawa Diagram Probable X’s found NA
find probable
reasons for Low PROCESS
People Customer
Measurement
0.40
0.35
0.30
0.90 0.95 1.00 1.05 1.10 1.15
PTP Rate
Study correlation Matrix Plot • Strong correlation observed All trends as expected.
between X’s to between Talk time and Contacts Expected correlations meant all
understand how Matrix Plot of Contacts, Talk Hrs, ... vs Contacts, Talk Hrs, ...
and further between Contacts and input metrics were in control and
one input metric 7000
800 1200 1600 3000 4500 6000 0.04 0.08 0.1 2
PTPs moving as expected
Contacts
interacts with other • Stable trend of Kept rate and PTPs It further increased the issue of
5000
3000 1600
Talk Hrs
1200
800
600
400
Cure at all
200
6000
PTP
4500
3000
PTP Kept Contact RateContacts/Hr
3.2
2.4
1.6
0.12
0.08
0.04
4000
3000
2000
3000 5000 7000 200 4 00 600 1.6 2.4 3.2 2000 3000 4000
Contac ts Talk Hrs Wrap Hrs PTP Contacts/Hr Contact Rate PTP Kept
Summary of Analysis (2 of 2)
Analysis Tool Used Inference/Observation Recommendation
Study split of time spent by Bar Chart Non Value Added time Introduce good Wrap and Preview practices
agents (Wrap+Preview) constitutes 49% of and track the times to increase Talk hours
total time Talk hours is a driving factor for PTPs and
hence Kept
Study if Cure Rate gets Box Plot, ANOVA Cannot be concluded that Not Applicable
impacted by number of calls Calls/hour impacts Cure Rate
made per hour
Study variation in Attempts Box Plot, ANOVA Attempts by agents differ Share best practices between top and average
made by Agents performers
Study variation in Contacts Box Plot, ANOVA Number of Contacts differ by agents Best Practice Sharing/ Mentoring
made by Agents
Study variation PTPs by Box Plot, ANOVA Number of PTPs taken differs by Best Practice Sharing/Mentoring
agents agents
Study variation in Kept Rate Box Plot, ANOVA Number of Kept Promises differs Best Practice Sharing/Mentoring
by agents significantly by agents
Study Variation in Cure Rate Box Plot, ANOVA Cure rate does not differ by Age of Not Applicable
by Age agents
Study dependence of Type Chi Square Test Cure Rate is independent of Type of Not Applicable
of Employment on Cure Employment
Rate
Study impact of Chi Square Test Cure Rate is dependent on Segregate inventory by employment status
Employment status Employment status of customer Assign better negotiators to such inventory
(Emp/Not Emp) on Cure
Key Solutions Implemented
S.
No Root Cause Addressed Recommendation Description
The dialer team logs into an automated Dialer system and the
time they spend on the system is divided into productive and
High amount of time spend in Reduction in Non Productive Non-productive. Non-productive time includes the time spent
3
Wrap and Preview activities Hours on Wrapping the calls and the time spent on previewing the
accounts before making a call. This non-productive time is very
high and needs to be reduced.
Best Practice Sharing between Analysis showed that agent performance differs significantly.
Associates
4 Variation between agents on Best Practice sharing was initiated and TLs were asked to
PTPs, Kept Rate etc. Skill Based allocation of mentor the agents in order to reduce agent to agent variation
in performance
accounts
Segregate inventory by
Cure Rate impacted by employment status of the Inventory to be segregated by employment status of the
5 employment status of customer. 100% segregation customer. If a customer is unemployed currently assign better
negotiators to the accounts who can explain the impact of
customers not possible so agent to tag delay and offer plans on the same.
accounts
Key Solutions Implemented- Adjusted Balances Saved
Solution: Adjusted Balances Saved Methodology for performance measurement whereby team targets were shifted from Number of
promises taken and Kept to ABS points
Partial Payment
collected
Key Solutions Implemented- Reducing Non Productive hours
Initiatives to reduce Non-Prod hours
Purpose
To create an Rationale Solution
incentives model
which drives The incentives A New
competition within a structure had incentives
team and gives Payout structure was
maximum incentives
to the top depending on developed,
performers an individual’s using stack
Need for an performance. ranking
incentives model
which not only The payout methodology.
promotes an was made Performance
individual’s basis the were
performance but also
compares it with the band wise segregated
other associates in kitty. into 4 Bands
the team.
Top
20%
50%
Kitty
Level2- Next 25%
30% Kitty
0.50
_
0.45 X=0.4519
Individual Value
0.40
LCL=0.3723
0.35
0.30
Cure% increased from
36.04% to 45.19% by
1 5 9 13 17 21 25 29 33 37 41 Dec’14-Jan’15. Delivered the
Observation highest ever cure in Feb
2015 at 51.73%
Two-Sample T-Test and CI: Cure Rate1_1, Cure Rate_2 The improvement translated
in annual savings of
Two-sample T for Cure Rate1_1 vs Cure Rate_2 Australian $4.68MM in Roll
Losses
N Mean StDev SE Mean
Cure Rate1_1 31 0.3666 0.0355 0.0064
Cure Rate_2 11 0.4519 0.0217 0.0065
Roll Forward% reduced from
17.43% in Sep 2013 to 14.17%
Difference = mu (Cure Rate1_1) - mu (Cure Rate_2) in Dec 2014. Delivered the
Estimate for difference: -0.08527 lowest ever roll forward rate in
95% CI for difference: (-0.10393, -0.06661) Nov 2014 at 13.84%
T-Test of difference = 0 (vs not =): T-Value = -9.35 P-Value = 0.000 DF = 29
- 36 -
Benefit- Business and End Customer Benefit
45%
Cure Rate
35%
$160MM
Business
Benefits
Business
Benefits
$ 80MM $ 4.68MM
Delinquent Inventory Reduction
Credit Loss Prevention by
by 50%
$4.68MM
Out of Debt!!
End
Implementation • All KPIs shifted to ABS from Promise Rate based metrics
2 ABS
of ABS • Score cards updated xxxx
• MIS team creates Incentives based on ABS methodology
Incentive • MIS team to ensure Incentives are allocated using the new
Allocation of split
3 allocation based xxxx
Incentives
on ABS scores • No agent wise Promise Rate available to Incentive
preparation team
Best Practice
sharing to reduce Ensure Best • Online Best practice repository created
4 agent Practices are • Best Practice questions included in Process Tests xxxx
performance shared • Quality team/TLs to do spot checks
variation
Accountant
Media Professional Storekeeper