Sunteți pe pagina 1din 40

Advanced Project

Schedule Risk Analysis


13th Annual International
Integrated Program Management

David T. Hulett, Ph.D.


Hulett & Associates, LLC
Project Management Consultants

Los Angeles, CA
(310) 476-7699
info@projectrisk.com

© 2001 Hulett & Associates, LLC

1
Agenda

• Uncertain activity durations


• Simple one-path schedule risk
• Risk at merge points: the “Merge Bias”
• Correlation among uncertain durations
• Probabilistic Branching
• Conditional Branching
• Resources
• Constraints

2
Activity Duration Risk

30d

Design Unit 1

• How long will this activity take? 30 days, right?


• Project personnel estimate the most likely
duration for each activity

3
Activity Duration Risk

• First, the schedule must be checked


– Calculates the right critical path
– Calculates the right end date when things change
– Finish-to-start relationships, no open ends…
• Then risks are identified using checklists and
brainstorming techniques
• Risk on activities is quantified using team meetings
and in-depth interviews
• A schedule risk analysis is a snapshot in time of
the risks that remain before further mitigation

4
Design Unit 1 Duration Uncertainty
Low=20d, Most Likely=30d, High=60d

Distribution for Design Unit 1


Triangle (20,30,60)

0.07
0.06
PROBABILITY

0.05
0.04
0.03
0.02
0.01
0.00
20

24

28

32

36

40

44

48

52

56
5
Risk Along a Path

Start Design Unit 1 Build Unit 1 Finish

6
Original Single-Path Schedule
ID Name Duratio Start Finish @RISK: Functions August SeptembOctober NovembDecemb January
1 Project 95 d 9/1 12/4
2 Start 0d 9/1 9/1 9/1
3 Design 30 d 9/1 9/30 Duration=RiskTRIANG(20,30,60) 9/1 9/30
4 Build 40 d 10/1 11/9 Duration=RiskTRIANG(30,40,65) 10/1 11/9
5 Test 25 d 11/10 12/4 Duration=RiskTRIANG(18,25,50) 11/10 12/4
6 Finish 0d 12/4 12/4 Finish=RiskOUTPUT() 12/4

• CPM schedule finishes on December 4. What is the


likelihood?
• Simulation Tools
@RISK for Project Professional from Palisade Corp. & RISK+ from C/S
Solutions, Inc. are MS Project Add-ins
Primavera P3 has Monte Carlo, Open Plan Professional simulates
7
Monte Carlo Simulation

• A simulation explores all combinations of durations


of uncertain (and certain) activities
• Durations are chosen at random from input
distributions
• The project is calculated (Press [F-9])
• Completion dates computed many times
• Distribution of completion dates
• Cumulative likelihood provides results

8
Completion Dates from
Simulation
Frequency Distribution
for Project Finish

0.10CPM Date Most Likely


Completion Date
0.08
PROBABILITY

0.06
0.04
0.02
0.00
11/15

11/24

12/3

12/12

12/21

12/30

1/8

1/17

1/26

2/4
Date
9
The Fallacy of Most Likely
Durations
• People sometimes say:
“Well, at least if we use the best estimates in our
schedule the CPM completion date is the most
likely date. Isn’t it?”

No, Never!

• In this case,
– CPM says December 4
– But the Most Likely completion date is
December 15
10
Cumulative Distribution --
December 10 is only 10% Likely
Cumulative Distribution
for Project Finish
Prob of Value <= X-axis

1.0
0.9
0.8
0.7 CPM
0.6 80% Likely
Value

0.5 Date
0.4 12/4 Schedule 1/3
0.3
0.2
0.1
0.0
1/8

2/4
11/15

11/24

12/3

12/12

12/21

12/30

1/17

1/26
Date
11
Results for Simple
Single-Path Schedule: CPM = 10%
S u m m a ry S ta tis tic s
fo r P ro je c t F in is h
M in im u m 1 1 /1 8
M a x im u m 2 /6
M ean 1 2 /2 2
S td D e via tio n 13
M ode 1 2 /1 5
5% 1 2 /1
10% 1 2 /5 CPM
20% 1 2 /1 1
30% 1 2 /1 4
40% 1 2 /1 8
50% 1 2 /2 1
60% 1 2 /2 5
70% 1 2 /2 9
80% 1 /3 80%
90% 1 /9 12
95% 1 /1 4
Risk at Path Merge Points
The “Merge Bias”

Design Unit 1 Build Unit 1

Start
Finish

Design Unit 2 Build Unit 2

13
Simple Two-Path Project
ID Name Duration Start Finish @RISK: Functions August Septemb October NovembeDecembeJanuary
1 Project 95 d 9/1 12/4
2 Start 0d 9/1 9/1 9/1
3 Component A 95 d 9/1 12/4
4 Design A 30 d 9/1 9/30 Duration=RiskTRIANG(20,30,60) 9/1 9/30
5 Build A 40 d 10/1 11/9 Duration=RiskTRIANG(30,40,65) 10/1 11/9
6 Test A 25 d 11/10 12/4 Duration=RiskTRIANG(18,25,50) 11/10 12/4
7 Component B 95 d 9/1 12/4
8 Design B 30 d 9/1 9/30 Duration=RiskTRIANG(20,30,60) 9/1 9/30
9 Build B 40 d 10/1 11/9 Duration=RiskTRIANG(30,40,65) 10/1 11/9
10 Test B 25 d 11/10 12/4 Duration=RiskTRIANG(18,25,50) 11/10 12/4
11 Finish 0d 12/4 12/4 Finish=RiskOUTPUT() 12/4

• CPM says this project also completes on


December 4
• But, Risk is greater than for the single-path
project!

14
Effect of the Merge Bias
D istribution for P roject F inish
O ne and Tw o P ath S chedules

1.0
0.9 CPM Date
P rob of value < = X -A xis

0.8
0.7
0.6
0.5
One-Path
V alue

0.4 Two-Path
0.3 Schedule Schedule
0.2
0.1
0.0
1/3

2/7
11/1
11/8

12/6

1/10
1/17
1/24
1/31
11/15
11/22
11/29

12/13
12/20
12/27

D ate
15
Comparison of Two Risky
Schedules: CPM < 5%
E vidence of the M erge B ias
T w o P a th s
O n e P a th M e rg e B ia s
M ean 1 2 /2 2 1 2 /2 9
M ode 1 2 /1 8 1 2 /3 1
S td D e via tio n 1 3 .1 1 1 .5
5% 1 2 /1 1 2 /1 1
CPM
CPM 10% 1 2 /5 1 2 /1 5
20% 1 2 /1 1 1 2 /1 9
30% 1 2 /1 5 1 2 /2 3
40% 1 2 /1 8 1 2 /2 6
50% 1 2 /2 2 1 2 /2 9
60% 1 2 /2 5 1 /1
70% 1 2 /2 9 1 /4
80% 1 /2 1 /8 80%
90% 1 /9 1 /1 3
16
95% 1 /1 4 1 /1 9
Defining the
Risk Critical Path / Activities
• With hundreds or thousands of activities, which
are most likely to delay the project?
– Depends on risk, project structure (float)
• Simulation program records whether an activity
was critical in each iteration

Percent of iterations each activity was critical


= its Criticality Index

17
Schedule with Risk Management
of Critical Unit B

ID Name Duration Start Finish @RISK: Functions August SeptembOctober Novemb DecembeJanuary
1 Project 95 d 9/1 12/4 Slack Path
2 Start 0d 9/1 9/1 Not Managed 9/1
3 Component A 93 d 9/1 12/2
4 Design A 28 d 9/1 9/28 Duration=RiskTRIANG(18,28,58) 9/1 9/28
5 Build A 40 d 9/29 11/7 Duration=RiskTRIANG(30,40,65) 9/29 11/7
6 Test A 25 d 11/8 12/2 Duration=RiskTRIANG(18,25,50) 11/8 12/2
7 Component B 95 d 9/1 12/4
8 Design B 30 d 9/1 9/30 Duration=RiskTRIANG(25,30,40) 9/1 9/30
9 Build B 40 d 10/1 11/9 Duration=RiskTRIANG(35,40,50) 10/1 11/9
10 Test B 25 d 11/10 12/4 Duration=RiskTRIANG(20,25,30) 11/10 12/4
11 Finish 0d 12/4 12/4 Finish=RiskOUTPUT() 12/4

Risk Managed
Critical Path

18
Criticality or % of
Iterations on Critical Path
Criticality Index
Percent
Task Critical
Component A 80%
Design A 80%
Build A 80%
Test A 80%
Component B 20%
Design B 20%
Build B 20%
Test B 20% 19
Correlation Between
Activity Durations
• Correlation when some risk factor (“driver”)
affects the durations of two activities together
• Difficult technology makes design and build take
longer
• Severe working conditions affect design and build
• Permit uncertainty affect design and build
S/W
Technology Development
Risk
S/W Testing

20
Correlation

• Correlation makes the durations “move” together


• If one activity takes longer than estimated the
other does too
• Both activities will take more (or less) time
together
• Correlation increases the risk of extreme results

21
Add Significant Correlation to
Single Path Schedule

C orrelation M atrix
D esign / B uild / Test /
D uration D uration D uration
D esign/D uration 1 0.8 0.6
B uild/D uration 0.8 1 0.9
Test/D uration 0.6 0.9 1

22
Correlations Increase the Spread
of the Results Distribution
Distribution for Correlated
and Not Correlated Durations

0.16 Not
Correlated
Relative Likelihood

0.12
Correlated
0.08

0.04

0.00
11/1

11/14

11/27

12/11

12/24

1/7

1/20

2/2

2/16

3/1
Date
23
Correlations Increase the Spread
of the Results Distribution
S-Curve for Correlated and Not
Correlated Durations
Prob Value <= Value on

1.0 Not
0.9 Correlated
0.8
0.7
Correlated
X-Axis

0.6
0.5
0.4
0.3
0.2
0.1
0.0
1/7

2/2

3/1
11/1

11/14

11/27

12/11

12/24

1/20

2/16
Date
24
Probabilistic Branching

• When the outcome of an activity is not certain


– Article is not certain to pass the test the first time
• The successor activity may be one or the other
– Pass the test? ==> Certify
– Fail the test? ==> End Test, Diagnose, FIXIT and retest
• Each one of these is a “branch” and has some
probability

25
Calculating Possibility of
Failure from 3 Events
Probability of Failure from 3 Events
Source of Probability Probability Merged
Failure Event of Failure of Success Probability
of Success
Facilities 5% 95% 64%
Equipment 10% 90% of Failure

Unit Under Test 25% 75% 36%

26
Computing the Impact of a
Failure from 3 Events

Im p act o f F ailu re fro m 3 E ven ts


S o u rce o f D iag n o se , B ran ch In p u ts
P ro b ab ility
F ailu re R e in stall R e m o v e , R e p air W e ig h te d b y
o f F ailu re
E vent & R e te st P ro b ab ility o f F ailu re
L o w M L H ig h L o w ML H ig h
F acilities 20% 25 10 20 40
E quipm ent 10% 25 12 17 35
U nit U nder
25% 25 8 24 90 34 47 95
T est

27
Probabilistic Branching

Unique ID Name Duration @RISK: Functions


1 Project 128 d
2 Start 0d
3 Design 30 d Duration=RiskTRIANG(20,30,60)
4 Build 40 d Duration=RiskTRIANG(30,40,65)
5 Test 25 d Duration=RiskTRIANG(18,25,50);RiskBRANCH(.36,.64,{t22},{t23})
22 FIXIT and Retest 30 d Duration=RiskTRIANG(34,46,87)
23 Certify Complete 3d
6 Finish 0d Finish=RiskOUTPUT()

Recommend Indicating Possibility


of Failure in the CPM Schedule –
Here at its Expected Value

28
Network Logic of Test Failure
Probabilistic Branch

29
Probabilistic Branching
Histogram
Distribution for Branch
of Probabilistic Failure

0.12 Success
0.1 Branch
PROBABILITY

Failure
0.08
Branch
0.06
0.04
0.02
0
1/8

2/9
11/5

11/21

12/7

12/23

1/24

2/25

3/13

3/29
Date
30
Cumulative Distribution of
Probabilistic Branch
Distribution for Finish of Probabilistic
Branching Network
Prob of Value <= X-axis

1.00
0.90
0.80
0.70
0.60 “Shoulder” at
Value

0.50
0.40 64% Success
0.30
0.20
0.10
0.00
11/5

12/7

1/8

1/24

2/9

2/25

3/13

3/29
11/21

12/23

Date
31
Conditional Branching

• Model decisions, e.g. alternative technology


decision
• Technology A
– Preferred by the customer
– A lot of schedule Risk
• Technology B
– Not preferred, but acceptable
– Less schedule risk than A

32
Model Technology Decision

ID Name Duration @RISK: Functions 3rd Quarte4th Quarte 1st Quarte


1 Start Milestone 0d 9/1
2 Technology A 125 d
3 Design Tech. A 50 d Duration=RiskTRIANG(40,50,100) 9/1 10/20
4 Make & Qual T 75 d Duration=RiskTRIANG(55,75,150) 10/21 1/3
5 Technology B 120 d
6 Design Tech. B 50 d Duration=RiskTRIANG(45,50,60) 9/1 10/20
7 Make & Qual T 70 d Duration=RiskTRIANG(60,70,90) 10/21 12/29
8 Finish Milestone 0d Finish=RiskOUTPUT() 1/3

33
Technology A Alone: No Plan
B? A = 100% Pr(Plan A) = 100%
Cumulative Distribution T echnology A : N o P lan B
Technology A: No Plan B Task C ritical Index
Mean 2/4
Mode 2/2
Technology A 100%
10% 1/5 D esign Tech. A 100%
20% 1/13 M ake & Q ual Tech A 100%
30% 1/20
40% 1/27
Technology B 0%
50% 2/2 D esign Tech. B 0%
60% 2/9 M ake & Q ual Tech B 0%
70% 2/17
80% 2/25
90% 3/10 34
If Technology A Design not
done by 10/25: Plan B

Unique I Name Duration @RISK: Functions


2 Start Milestone 0d
8 Technology A 125 d
3 Design Tech. A 50 d Duration=RiskTRIANG(40,50,100);RiskIF(t3[Finish]>10/25/01,Branch=t7)
5 Make & Qual Tec 75 d Duration=RiskTRIANG(55,75,150)
9 Technology B 120 d
4 Design Tech. B 50 d Duration=RiskTRIANG(45,50,60)
6 Make & Qual Tec 70 d Duration=RiskTRIANG(60,70,90)
7 Finish Milestone 0d Finish=RiskOUTPUT()

35
Switch to Plan B if Design for
Plan A is Not done by 10/25
C um ulative D istribution T echnology D ecision
S w itch to P lan B on 10/25 C onditionalB ranching:
M ean 1/10 S w itch to P lan B on 10/25
M ode 1/5 Task C ritical Index
10% 12/27 Technology A 30%
20% 12/31 D esign Tech. A 23%
M ake & Q ual Tech A 30%
30% 1/2
Technology B 70%
50% 1/7
D esign Tech. B 70%
60% 1/9 M ake & Q ual Tech B 70%
70% 1/12
80% 1/16 Pr(Plan A) = 30%
36
90% 1/27
Decision Rule Trade-Off

• Trade off
– Likelihood of completing on time
– Likelihood of using Preferred Technology A

Model the Technology / Schedule Trade-Off


Tech A: No Tech. B
Measure Plan B After 10/25
Optimistic (10%) 1/5 12/27
Mean Completion 2/4 1/10
Pessimistic (90%) 3/10 1/27
Probability of Using
Technology A 100% 30%
37
Resources and Constraints

• If there are scarce resources, they must not be in


conflict in the schedule
– In CPM, scheduling packages will “level” resources – this
means shifting activities out
– Since simulation is a number of CPM calculations, each
iteration must be leveled
• Schedulers often use constraints
– Eliminate the Constraints – let the project overrun on
the computer, not the real project
– Must Finish On, and Finish No Later Than will hide the
risk

38
Resource Leveling

E ffe ct o f R e so u rce L e v e lin g


P ro b V alu e < = X -A xis D a

1 .0
0 .9
0 .8
0 .7
0 .6
0 .5
0 .4
Not
0 .3 Leveled
0 .2 Resource
0 .1
0 .0
Leveled
1 1 /2 0

1 2 /1 0

1 2 /2 9
1 1 /1

1 /1 8

2 /2 6

3 /1 7

4 /2 5
2 /6

4 /6
D a te

39
Effect of Constraints

E ffe ct o f C o n strain in g S ch e d u le
to F in ish N o t L ate r T h an 12/11
P ro b o f V alu e < = X -ax

1 .0
0 .9
0 .8
0 .7
Constrained
V alu e

0 .6
0 .5
0 .4
0 .3
0 .2 Not
0 .1 Constrained
0 .0
1 1 /1 3

1 1 /2 5

1 2 /1 9

1 2 /3 1
1 1 /1

1 2 /7

1 /1 2

1 /2 4

2 /1 7
2 /5
D a te
40

S-ar putea să vă placă și