View
1.066
Download
0
Category
Tags:
Preview:
DESCRIPTION
Successful presentation to the I.T. Director of Giant Plc., and to the co-team from Barclays Banks Plc.
Citation preview
1
Giant Plc.Barclays Plc.
Applying Six Sigma to Financial Software Development
Ali Raza KhanGiant Plc.
1 New Oxford Street, London ♦ United Kingdom(+44) 18443247700 ♦ alirazakhan83@gmail.com
© 2009 Giant Plc.
2
Giant Plc.Barclays Plc.
Motivation
Financial Software Financial Software DevelopmentDevelopment
• Large Opportunity for Improvement
• Approximately 25% of software projects are canceled
• Average project exceeds – Costs by 90% – Schedule by 120%
• Risk of project failure increases with size
© 2009 Giant Plc.
Six SigmaSix Sigma
• Well-defined improvement approach
• Impressive track record of achievements
• Adaptable
3
Giant Plc.Barclays Plc.
But Software Development is Not a Typical Application
DesignRequirements Testing IntegrationCoding Release
• Process Oriented, but– Inputs often ill-defined
– Outputs often difficult to fully evaluate
– Performance highly influenced by human factors (e.g., knowledge, skills, experience, etc.)
• Significant natural variation
© 2009 Giant Plc.
4
Giant Plc.Barclays Plc.
Key Factors in Software Project Failures
Risk Factor % of “MIS” Projects
Requirements Failures
Creeping Requirements 80%
Expectation Failures
Excessive Schedule Pressure 65%
Execution Failures
Low Quality 60%
Cost Overruns 55%
Inadequate Configuration Management
50%
5
Giant PlcBarclays Plc
Applying Six Sigma to Software Development
Design CodeIntegr-ation
Test
DMAIC
FuzzyFront-End
DFSS
BetaRelease
© 2009 Giant Plc.
6
Giant PlcBarclays Plc
Fuzzy Front End
Six Sigma DFSS
© 2009 Giant Plc
7
Giant PlcBarclays Plc
Balance the VOC and the VOB
Voice of the Customer
Voice of Business
8
Giant PlcBarclays Plc
VOC – Voice of the Customer
• Understand Internal and External Customers and Target Environment
© 2009 Giant Plc.
www.estm.biz 9
Building a Customer MatrixS
eg
men
ts
Types of Customers
Lead User
Demanding
Lost Lead
Had But Lost
U.S.
Europe
Asia
© 2009 Giant Plc.
10
Giant PlcBarclays Plc
VOC – Voice of the Customer
• Understand Internal and External Customers and Target Environment
• Identify, Characterize and Verify Critical to Quality (CTQ) Requirements– Interviews, focus groups, use cases, etc. – Preference surveys and Kano analysis
© 2009 Giant Plc.
11
Giant PlcBarclays Plc
Kano Analysis
• Dissatisfiers (or basic requirements)– “Must be” requirements– These features must be present to meet minimal
expectations of customers• Satisfiers (or variable requirements)
– The better or worse you perform on these requirements, the higher or lower will be your rating from customers
• Delighters (or latent requirements)– These are features, factors, or capabilities that go
beyond what customers expect, or that target needs customers can’t express themselves
© 2009 Giant Plc.
www.estm.biz 12
The Kano ModelH
ow
th
e C
ust
om
er F
eels
Neutral
Satisfier
Must-Be
Delighter
Level of Functionality Delivered
for a particular requirement
Low to None
High
Delighted
Very Dissatisfied
Satisfied, but I expect
it this way
I can live with it
© 2009 Giant Plc
13
Giant PlcBarclays Plc
VOC Output: Prioritized CTQsRequirement Use-case Kano Priority
Manage database interfaces Verifying data
content integrityS 4
Manage Network I/O
Moving client-server data
M 3
Optimizing data transfer
D 3.5
Provide real-time user access
Minimizing system response time
M 5
© 2009 Giant Plc.
14
Giant PlcBarclays Plc
VOC – Voice of the Customer
• Understand Internal and External Customers and Target Environment
• Identify, Characterize and Verify Critical to Quality (CTQ) Requirements– Interviews, focus groups, use cases, etc. – Preference surveys and Kano analysis
• Establish measures for CTQ requirements
© 2009 Giant Plc.
15
Giant PlcBarclays Plc
VOC Output: Fully Characterized CTQsRequirements Use-Case Kano Priority Measure
Manage Database Interfaces
Minimum Average Strong
Verifying data content integrity
S 4 ≤ 1 record/1,000 ≤ 1 record/10,000 ≤ 1 record/100,000
Manage Network I/O
Moving client-server data
M 3 100 records/min. 500 records/min. 800 records/min.
Optimizing data transfer
D 3.5 Hooks for user supplied
compression
Top 5 compression
schemes supplied
Top 10+ compression
schemes supplied and fully
integrated
© 2009 Giant Plc.
16
Giant PlcBarclays Plc
VOB - Voice of Business
• Analyze Design Options– Estimate customer satisfaction– Level of effort– Capability to deliver– Balance VOC and VOB
© 2009 Giant Plc.
17
Giant PlcBarclays Plc
Analyze Design Options
Requirement Use-Case Kano Priority Base Full Base
Effort
Full Effort
Manage database interfaces
Verifying data content integrity
S 4 1 3 1000 1500
Manage Network I/O
Moving client-server data
M 3 1 3 5500 7500
Optimizing data transfer
D 3.5 1 3 12000 18000
Customer Sat. Score
Effort Score
Design Options Level of Effort
© 2009 Giant Plc.
18
Giant PlcBarclays Plc
Analyze Design Options
Requirement Use-Case Kano Priority Base Full BaseEffort
Full Effort
Manage database interfaces
Verifying data content integrity
S 4 1 3 1000 1500
Manage Network I/O
Moving client-server data
M 3 1 3 5500 7500
Optimizing data transfer
D 3.5 1 3 12000 18000
Customer Sat. Score
Effort Score
= F(Kano, Priority, Feature Level)
© 2009 Giant Plc.
19
Giant PlcBarclays Plc
Analyze Design Options
Requirement Use-Case Kano Priority Base Full BaseEffort
Full Effort
Manage database interfaces
Verifying data content integrity
S 4 1 3 1000 1500
Manage Network I/O
Moving client-server data
M 3 1 3 5500 7500
Optimizing data transfer
D 3.5 1 3 12000 18000
Customer Sat. Score
Effort Score
= ∑ Effort Estimates
© 2009 Giant Plc.
20
Giant PlcBarclays Plc
Concept Selection
4 3
1 2
CustomerSatisfaction
Eff
ort
© 2009 Giant Plc.
21
Giant PlcBarclays Plc
Computing Productivity
Historically, for each project we should know Size, Effort, and Duration
)(
)(/
/3431
*
(SLOC)
yearsDurationB
StaffYearsEffort
SizePP
© 2009 Giant Plc.
22
Giant PlcBarclays Plc
Schedule Compression
3)()(
torelates
yearsDurationStaffYearsEffort
MBI
MBI Buildup Rate Equation Output
1 Slow 7.3
2 Mod. Slow 14.7
3 Moderate 26.9
4 Rapid 55
5 Very Rapid 89
Manpower Buildup Index, MBI
© 2009 Giant Plc.
www.estm.biz 23
Rayleigh Curve
Rayleigh Summary
0.0
2.0
4.0
6.0
8.0
10.0
12.01 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Time Interval
Eff
ort
0.0
20.0
40.0
60.0
80.0
100.0
120.0
140.0
160.0
Def
ects Effort
Defects
© 2009 Giant Plc.
24
Giant PlcBarclays Plc
MBI = 1 (Slow) Concept 1
Duration (months) 15.2
Effort (staff months) 77.3
Released Defects 14.1
Effort Cost $966,250
Duration Adjustment
Defect Repair Cost $239,700
Net Value $19,482,050
MBI = 3 (Moderate) Concept 1
Duration (months) 13
Effort (staff months) 119.4
Released Defects 267
Effort Cost $1,492,500
Duration Adjustment $1,400,000
Defect Repair Cost $453,900
Net Value $20,141,600
Balancing VOC and VOB
Business Value $20,000,000
Feature Value $688,000
© 2009 Giant Plc.
25
Giant PlcBarclays Plc
Balancing VOC and VOB
MBI = 5 (Very Rapid) Concept 1
Duration (months) 11.7
Effort (staff months) 220.7
Released Defects 508
Effort Cost $2,758,750
Duration Adjustment $1,400,000
Defect Repair Cost $663,600
Net Value $18,665,650
MBI = 3 (Moderate) Concept 1
Duration (months) 13
Effort (staff months) 119.4
Released Defects 267
Effort Cost $1,492,500
Duration Adjustment $1,400,000
Defect Repair Cost $453,900
Net Value $20,141,600
Business Value $20,000,000
Feature Value $688,000
© 2009 Giant Plc.
26
Giant PlcBarclays Plc
VOB - Voice of Business
• Analyze Design Options– Estimate customer satisfaction– Level of effort– Capability to deliver– Balance VOC and VOB
• Select Concept and Approach– Flesh out concept
• QFD• FEMA
– Verify and refine approach• Defect analysis• Schedule simulation
© 2009 Giant Plc.
www.estm.biz 27
Capability to Deliver on TimeProbabilistic Scheduling
Frequency Chart
Certainty is 94.90% from -Infinity to 289.17 days
.000
.007
.014
.021
.028
0
7
14
21
28
250.00 262.50 275.00 287.50 300.00
1,000 Trials 4 Outliers
Forecast: F52
How much confidence should we have in the schedule?… At a 95% confidence level
• latest mid March, 2003 (+ 43 days)• earliest mid January, 2003 (- 15 days)
Upper Spec Limit
(USL)
© 2009 Giant Plc.
28
Giant PlcBarclays Plc
Process Improvement
MeasureDefine Analyze Improve Control
Standard Six Sigma DMAIC Process
© 2009 Giant Plc.
29
Giant PlcBarclays Plc
Application to Software
DesignRequirements Testing IntegrationCoding Release
Requirements ReleaseDevelopmentX
Prerequisites:Processes must be well defined
© 2009 Giant Plc.
30
Giant PlcBarclays Plc
DMAIC Example
• Problem Statement– Post release maintenance has increased by 30%
since the end of last fiscal year and is now limiting new product development.
• Goal Statement– Reduce post release maintenance by 40% by the
end of Q4’2003.
© 2009 Giant Plc.
31
Giant PlcBarclays Plc
Measure – Data Collection
• Total Problems Fixed Prior to Release Per Project– Pre-Release Defects: defects found and fixed during
development and testing
• Total Post Release Problems Per Project– Released Defects: defects reported by customers
• Types of Post Release Problems– All projects
– Per project
© 2009 Giant Plc
32
Giant PlcBarclays Plc
Analysis
To
tal P
re-R
ele
ase
De
fec
ts
Project Size (LOC)
x x
x xx
xx
x
x
xx
x
Pre-Release Defects = f(Size)
© 2009 Giant Plc.
33
Giant PlcBarclays Plc
Analysis
Esc
aped
Def
ects
Pre-Release Defects
x
x
x
x
x
xx
xx
x
x
x
• Escaped defects proportional to pre-release defects – No significant variation in Defect Containment Effectiveness
• DCE = Pre-Release Defects/(Pre-Release Defects + Escaped Defects)
© 2009 Giant Plc.
34
Giant PlcBarclays Plc
Analysis
40%29% 26%
5%
Code
DesignReq’ts
Test
• Most Escaped Defects are Code Related
© 2009 Giant Plc.
35
Giant PlcBarclays Plc
Improve
• Improve the Effectiveness of Code Inspections– Factors
• Size of unit (LOC)
• Preparation time (LOC/hour)
• Inspection time (LOC/hour)
• Number of reviewers
– Measure• Number of identified defects
© 2009 Giant Plc.
36
Giant PlcBarclays Plc
Improve
• Improve the Effectiveness of Code Inspections– Conduct DOE
• Determine most effective combination of factors
– Verify DOE results• Pilot test using real project
© 2009 Giant Plc.
37
Giant PlcBarclays Plc
Control
• Establish Performance Standard for Code Inspections– Defects/KLOC
• Monitor Performance– Take action when unacceptable performance
observed
© 2009 Giant Plc.
38
Giant PlcBarclays Plc
Questions
© 2009 Giant Plc.
Recommended