Академический Документы
Профессиональный Документы
Культура Документы
Businesses
February 2015
Agenda
Introduction
Analytics: Future of Business Consulting
Business Cases: Examples from PwC US Advisory
PwC
February 2015
2
PwC
February 2015
3
Big Data
Cloud Computing
Mobile
Data Visualization
UX Design Skills
PwC
February 2015
4
February 2015
5
Finding how many people are going to attrite from an organization in next 1
year: regression modeling
PwC
February 2015
6
Professors
PwC Diamond
Researchers
Fractal Analytics
IEOR
Google Analytics
Opera Solutions
EXL Inductis
PwC
February 2015
7
Business Cases
PwC
February 2015
8
Outcomes:
Hypotheses inventory
Area
Area
Current and Past Managers
Area
Current and Past
Managers
Current
andand
PastPast
Managers
Current
Managers
Current
and Past
Managers
Current
and
Past
Managers
Currentand
and
PastManagers
Managers
Current
Past
Current and Past Managers
Demographics
Demographics
Demographics
Demographics
Demographics
Education/
Credentials
Demographics
Education/
Credentials
Education/ Credentials
Education/ Credentials
Education/ Credentials
Education/ Credentials
Engagement Levels
Engagement Levels
Engagement
Levels
Engagement
Levels
Engagement
Levels
Engagement Levels
Hypotheses
Hypotheses
Current/past manager/coach Hypotheses
attrition will lead to attrition
Current/past manager/coach attrition will lead to attrition
Time
under the same
manager may
affect will
attrition
Current/past
attrition
lead to attrition
Time
under themanager/coach
same manager may
affect attrition
Number
of
employees
churned
under
the
manager
will affect
Time under
the samechurned
manager
may
affect
attrition
Number
of employees
under
the
manager
will affect
attrition
Number
of
employees
churned
under
the
manager
will affect
attrition
Tenure
will have a U curve for attrition
attrition
Tenure
will have a U curve for attrition
International
employees
have
higher
attrition
Tenure will have
a U curve
attrition
International
employees
havefor
higher
attrition
PartInternational
time education/MBA/credentials
in last
cycle will lead to
employees have higher
attrition
Part time education/MBA/credentials
in
last cycle will lead to
attrition
Part time education/MBA/credentials in last cycle will lead to
attrition
Specific
academic
background
/institutes
will
have higher
attrition
Specific
academic background /institutes will have higher
attrition
Specific academic background /institutes will have higher
attrition
Lowattrition
engagement score lead to attrition
Low engagement score lead to attrition
Employee
engaged inscore
an industry
having higher growth
Low engagement
lead to vertical
attrition
Employee
engaged in an industry
vertical having higher growth
willEmployee
have higher
attrition
propensity
jobhaving
availability)
engaged
in an
industry(more
vertical
higher growth
will have higher
attrition
propensity
(more job
availability)
will have higher attrition propensity (more job availability)
Filters
Attrition is higher for some particular vertical/ business unit
Filters
Attrition is higher for some particular vertical/ business unit
Filters
Some
officesiswill
havefor
higher
Filters
Attrition
someattrition
particular
Filters
Some
offices higher
will have
higher
attrition vertical/ business unit
Filters
Smaller
higher
attrition
Filters
Someoffices
officeshave
have higher
attrition
Filters
Smaller
officeswill
have
higher attrition
Hours
worked and vacation time High
one-offoffices
vacations
to past may lead to churn
Filters
Smaller
havecompared
higher attrition
Hours worked and vacation time High one-off vacations compared to past may lead to churn
Hours worked and vacation time High one-off vacations compared to past may lead to churn
Hours worked and vacation time Overall number of working hours would have higher attrition
Hours worked and vacation time Overall number of working hours would have higher attrition
Hours worked and vacation time Overall number of working hours would have higher attrition
Hours worked and vacation time Change in working hours compared to past year would have
Hours worked and vacation time Change in working hours compared to past year would have
Hours worked and vacation time
Change
in working hours compared to past year would have
impact
on attrition
impact on attrition
impact
on attrition
Hours worked and vacation time Change
in working
hours compared to peers would have impact
Hours worked and vacation time Change in working hours compared to peers would have impact
Hours worked and vacation time
Change in working hours compared to peers would have impact
on attrition
on attrition
on attrition
Hours worked and vacation time Change
in health condition will lead to attribution
Hours worked and vacation time Change in health condition will lead to attribution
Hours worked and vacation time Change in health condition will lead to attribution
Hours worked and vacation time Sabbatical will lead to higher attrition
Hours worked and vacation time Sabbatical will lead to higher attrition
Hours worked and vacation time Sabbatical will lead to higher attrition
Hours worked and vacation time Leave utilization will affect attrition
Hours worked and vacation time Leave utilization will affect attrition
Hours worked and vacation time Leave utilization will affect attrition
PwC
Comments
Client to scramble original IDs
Yes/no
Yes/no
Yes/no
Yes/no
Yes/no
Yes/no
To be
customized as per requirements
To be customized as per requirements
Yes/no
Outcomes:
Raw Analytic
Dataset
48
Company Industry
List
49 of variables
Role from Datawarehouse/ HRIS
49
Role
50
Salary
50
Salary
51
Level
Category
DataLevel
Elements
51
52
Function
1 ID52
EmplID
Function
53 Post Client status Left for competition?
2 ID53 Post Client statusScrambled
Left for ID
competition?
Continued
to work as a contractor?
354Demographics
Job
Title
54
Continued to work as a contractor?
Left for job in Federal sector?
45555
Sex
Left for job in Federal sector?
Filters
Role
55656
Race
Filters
Role
Sub Role
65757
Country
origin
SubofRole
Line of Service
75858
Age/Date
Birth
Line ofofService
Sub Level
of Service
85959
Full/Part
Time
Sub Level
of Service
Job Code
960
Function
60
Job Code
Level Descr
1061
Function
61
Level
Levelfacing
Descrrole
1162
Client
62
Level Descr
Role
1263Hiring/Termination Hire
Date
63
Role
Role Descr
1364information
Status
(active/departed)
64
Role Descr
Geo MarketDate
1465
Termination
65
Geo Market
Region reason code
1566
Termination
66
Region
Country name
Interviewer
1667Interview
67
Country
HIPO/ Pivotal
Identified level
as HIPO (yes/no)
1768Performance
Interviewer
68 HIPO/ Pivotal
Identified as HIPO (yes/no)
Identified rating
as Pivotal employee (yes/no)
1869 identification
Interviewer
69 identification
Identified as Pivotal employee (yes/no)
Identified
1970
Starting
payas HIPO previous years (yes/no)
70
Identified as HIPO previous years (yes/no)
Identified
as Pivotal employee previous years (yes/no)
2071
Sign-on
bonus
71
Identified as Pivotal employee previous years (yes/no)
Total (yes/no)
hours worked current year
2172 Hours worked and Rehire
72 Hours worked and Total hours worked current year
Total(yes/no)
hours worked in previous years
2273 vacation time
Intern
73 vacation time
Total hours worked in previous years
Clientsource
hours worked current year
2374
Hiring
74
Client hours worked current year
Client hours
worked in previous years
2475Previous company Previous
Company
75
Client hours worked in previous years
Biz development
worked current year
2576
Direct
from college hours
(yes/no)
76
Biz development hours worked current year
Biz development
hours(yes/no)
worked in previous years
2677
Direct
from grad school
77
Biz
development
hours
worked in previous years
2778
Worked
as hours
a contract
employee
Vacation
current
year for Client (yes/no)
Vacation
hours current
287978
Years
of experience
in prior year
company
Vacation
hours in previous
years
Vacation
in previous
years
298079
Total
years Leave
of hours
experience
prior to
joining
Parental
Hours current
year
Parental
Leavesalary
Hours current year
308180
Previous
company
Parental
Leave Hours
in previous years
81
Parental
Leave
Hours in previous years
3182
Previous
company
Sick Leave
hourslocation
current year
Sickcompany
Leave hours
current year
328382
Previous
Sick leave
hoursIndustry
in previous years
Sick leave hours
in previous years
338483
Previous
Currentcompany
Vacation role
Balance
Current
Vacation Balance
348584
Previous
company
Vacation
balancefunction
in previous years
85
Vacation balance in previous years
3586Referer
Referer
Hours over capacity (Month by Month) current year
Hours over of
capacity
(Month by Month) current year
368786
Current
referrer
Hours Status
over capacity
(Month by Month) previous years
Hours
capacity
(Month
by Month)
previous years
378887
If referer
noover
longer
with
firm,
reason
for departure
Increase/decrease
from
previous
years
Increase/decrease
from date
previous
years
388988
longer
with firm,
of departure
Commute distance If referer
Currentnowork
zipcode
89 Commute
Current work zipcode
3990Hiring
Managerdistance
Hiring
Pastmanager
work zipcodes
Past workunder
zipcodes
409190
Years
Homeworked
zipcode hiring manager
Home
zipcode
419291
Current
Status
of hiring manager
Old home
zipcodes
Old home zipcodes
429392
If hiring
with firm, reason for departure
Miles manager
commuteno
(if longer
available)
Miles commute
(if available)
439493
If hiring
longer
with firm, date of departure
Miles manager
commuteno
past
(if available)
94
Miles commute past (if available)
Comments
Client to scramble original IDs
Yes/no
Yes/no
Yes/no
Yes/no
Yes/no
Yes/no
To be customized as per requirements
To be customized as per requirements
Learning
Management
System
Yes/no
Recruitment
System
HR
Information
SystemPeopleSoft
ERP Finance
& Operations
Macroeconomic/I
ndustry data
Employee
Survey
Data
PwC
Core Analytic
Dataset
Computed
Predictors
10
2. Variable Treatments
3. Modeling dataset
Flooring
Core Analytic
Dataset
Capping
Missing
value
imputation
Modeling dataset
KS value
Maximum KS of
41.55% indicate that
the model is able to
discriminate
between attrites and
non-attrites
100%
90%
80%
80%
70%
70%
60%
60%
50%
50%
40%
40%
30%
30%
20%
20%
10%
22.2%
20.0%
20.0%
Attrition rate
100%
90%
25.0%
% Cumulative Percentage
% Cumulative Attrite
15.0%
10.0%
5.6%
5.2%
5.0%
4.4%
4.8%
0.0%
10%
0%
0%
0%
20%
40%
60%
80%
100%
% Cumulative Non Attrite
%Culmulative Non-Attrite
0%
10%
20%
30%
40%
% Cumulative Attrite
50%
60%
70%
9. Final model
Predictor Category
8. Model fine-tuning
Attrition Predictors
Demographics
Tier
Awards
Training
Multivariate visualization
Variable interpretation
Multicollinearity checks
7. Iterative multivariate
analysis
Iteration 1
Param eter
DF
Estim ate Standard Error Wald Chi-Square Pr > ChiSq
Intercept
1
-2.1 7 85
0.3284
44.01 66 <.0001
A ttrtn_under_mgr_flg
1
0.2265
0.0831
7 .4239
0.0064
srv y _Q7 _y 0_flg
1
0.927 4
0.1 867
24.6622 <.0001
Iternation 2
prtm_flg
1
1 .4231
0.801 5
3.1 528
0.07 58
Param eter
DF
Estim ate Error
Chi-Squ are
Pr > ChiSq
dpndnts_cnt_y 0
1
-0.1 338
0.0482
7 .7 1 29
0.0055
Intercept
1
-2.1 1 02
0.3329
40.1 904 <.0001
age
1
-0.01 99
0.00687
8.3956
0.0038
A ttrtn_under_mgr_flg
1
0.2223
0.0831
7 .1 591
0.007 5
awrd_cat1 _flg_y 0
1
1 .7 1 27
0.1 486
1 32.909 <.0001
srv y _Q7 _y 0_flg
1
0.9601
0.1 888
25.8642 <.0001
bnft_chng_cnt_y 0
1
0.97 84
0.3528
7 .6902
0.0056
prtm_flg
1
1 .4261
0.8067
3.1 251
0.07 7 1
lv l_mod_y 0
1
-0.1 34
0.07 06
3.6033
0.057 7
dpndnts_cnt_y 0
1
-0.1 281
0.0483
7 .0263
0.008
age
1
-0.01 85
0.00695
7 .1 1 1 8
0.007 7
Percent Concordant
7 0 Somers' D
0.41 9
awrd_cat1 _flg_y 0
1
1 .6325
0.1 638
99.2895 <.0001
Percent Discordant
28.1 Gamma
0.428
ltst_tnr
1
-0.00251
0.0021 9
1 .3088
0.2526
Percent Tied
1 .9 Tau-a
0.055
bnft_chng_cnt_y 0
1
0.967 8
0.3527
7 .532
0.0061
Pairs
1 1 57 285 c
0.7 1
lv l_mod_y 0
1
-0.1 324
0.07 05
3.5283
0.0603
Percent Concordant
Percent Discordant
Percent Tied
Pairs
PwC
7 0.4
27 .8
1 .8
1 1 57 285
Somers' D
Gamma
Tau-a
c
6. Variables clustering
X12 X7
X14 X1 X6
X9 X3 X2
X5
X16 X4 X8 X13
X11 X10 X15
Cluster 3
Cluster 1
Set of all
variables
Cluster 2
0.426
0.434
0.055
0.7 1 3
11
Outcomes:
Where
PwC
12
Business Cases
PwC
February 2015
13
PwC
Traditional PC Users
Family Fortunes
Heavy buyers of PC/
Laptop/ Software
Online channel
preferred
Power Couples
More preference
towards PC/Laptop
High preference for
tablets
18 mi
Incremental benefit :
100 customers / mi
Heavy Gamers
Family Sprawl
Large family
size(4+)
High PC preference
as well
School Daze
High propensity for
game systems
Medium Online
preference
High
Slide 15
10000
9000
Absolute distance
threshold is around 13
miles after which the
increment is marginal
Sales vs distance
8000
# Customers
7000
6000
5000
Gaming
Tablets
Software
PC/Laptops/Tablets
4000
Sales
Poly. (Sales)
3000
2000
1000
0
10
20
30
40
50
Data used : sanitized sales data (mocked up from a different project) by zipcode
PwC
Slide 16
Competition Analysis
10000
9000
Without competition
8000
With competition
# Customers
7000
6000
5000
4000
3000
2000
1000
0
0
10
20
30
Distance from store
40
50
Sales
60
Data used : sanitized sales data (mocked up from a different project) by zip code
PwC
Slide 17
18 mi
High
client Store
PwC
Slide 18
Sales vs Distance
Online sales+ Retail PoS
10000
# Customers
8000
6000
Sales
Competition Presence
Poly. (Sales)
4000
0
0
10
20
30
40
50
60
Distance
Data used : sanitized sales data (mocked up from a different project) by zip code
PwC
Slide 19
Business Cases
PwC
February 2015
20
Size : Penetration
Low
Product A*
High
Product B*
Product C*
November 2014
21
Demand
Size : Population
Low
Low
High
Medium
High
November 2014
22
PwC
November 2014
23
Customer Profile
The following segment Economizers dominates the Dallas area. This segment consists of the poorest financial
groups. Consists of racially mixed singles and single-parent families, watching wresting and listening to gospel
radio
PwC
24
PwC
February 2015
25
PwC
February 2015
26
PwC
February 2015
27
February 2015
28
Questions...
This publication has been prepared for general guidance on matters of interest only, and does
not constitute professional advice. You should not act upon the information contained in this
publication without obtaining specific professional advice. No representation or warranty
(express or implied) is given as to the accuracy or completeness of the information contained
in this publication, and, to the extent permitted by law, PricewaterhouseCoopers LLP, its
members, employees and agents do not accept or assume any liability, responsibility or duty of
care for any consequences of you or anyone else acting, or refraining to act, in reliance on the
information contained in this publication or for any decision based on it.
2015 PricewaterhouseCoopers LLP. All rights reserved. In this document, PwC refers to
PricewaterhouseCoopers LLP which is a member firm of PricewaterhouseCoopers
International Limited, each member firm of which is a separate legal entity.