1 Complementing Approaches in ERP Effort Estimation Practice: an Industrial Study Maya Daneva
Jun 09, 2015
1
Complementing Approaches in ERP Effort Estimation Practice: an Industrial Study
Maya Daneva
2
Table of Contents
1. Why ERP effort estimation is difficult?
2. The solution proposal
3. The case study
4. Validity threats
5. Future activities
3
• ERP projects are notorious for delays, budget overruns, cancellations
• Customization & reuse compromises
Example: 5000 parameters, 10 000 tables in SAP R/3
• Architecture is designed when most users are not known• Shortage of proper methodologies to evaluate functional
size, effort, productivity, schedule.• No historical data sets• Even when earlier project data exists, effort and duration
for similar ERP projects have been noted to vary widely.
Estimating ERP Projects: Some Challenges
4
The Solution
• Key idea: integrate the use of COCOMO II, Monte Carlo simulation and portfolio management
• The targeted effects are to systematically cope with:– Uncertainty of cost drivers
– Strong bias by vendors/consultants in effort estimation.
5
The Solution: A High-level View
Monte Carlo simulation
Probability distribution of
cost factor values
COCOMO IIProbability
distribution of effort/duration
portfolio management
method
Probability of success
under deadline/effort constraints
6
Estimate size
Formulate condition for
portfilo management
Adjust cost drivers to increase portfilo
success
Obtain probability distribution of effort & duration
Run 10000 trials using probabiliy
distribution of cost factor
values
Ascribe distribiution types to cost
drivers
Construct portfolios
Obtain ratio of increase of success probability
Step 1:
Step 2:
Step 3:Step 4:
Step 5:
Step 6:
Step 7: Step 8:
The Steps
7
The Case Study: Planning
1. Site: Telus Corporation2. Canada-wide roll-out of 8 ERP modules in 13
projects (1997-2003), 67 subprojects3. Size Measure: unadjusted Function Points (IFPUG) 4. Reuse levels: based on reuse percentage ratio5. No knowledge of uncertainty of cost drivers6. Used default levels proposed by other authors7. Monte Carlo simulation runs: 10 000
8
The Case Study: Execution
Effort: Frequency Chart
0
100
200
300
400
500
600
17,9 18,9 19,9 20,9 21,9 22,9
Example: Probability distribution of project effort in person/months
9
The Case Study: Execution
Example: Probability distribution of time in months
Time: Frequency Chart
0
100
200
300
400
500
600
3,8 4,8 5,8 6,8 7,8 8,8 9,8
10
The Results (I) • When adjusting cost drivers, we can increase the
probability of success under effort constraints and under time constraints.
– For each cost driver, two portfolios are constructed, with either VERY HIGH ratings or LOW ratings
– For 13 out of the 17 drivers, we observed that success could be maximized, when drivers are adjusted.
REUSE rating Probability of success
Under effort constraints Under time constraints
Very low 68.78% 76.52%
Very high 96.87% 98.88%
11
The Results (II): Are projects more successful when managed as a portfolio?
Uncertainty level of cost drivers
Probability of success Ratio of increase(a)/(b)Individual
projects(a)
Portfolio(b)
Low uncertainty 93.78% 98.81% 1.05
High uncertainty 84.31% 97.76% 1.16
Context: Under effort constraint
12
The Results (III): Are projects more successful when managed as a portfolio?
Uncertainty levelof cost drivers
Probability of success Ratio of increase(a)/(b)Individual
projects(a)
Portfolio(b)
Low uncertainty 15.76% 87.52% 5.55
High uncertainty 8.31% 75.91% 9.13
Context: Under time constraint
13
Validity Concerns
1. External validity: use of ASUG
2. Choice of techniques: why these three techniques and not others?
3. Replication plans
14
Conclusions
• Made a solution proposal with respect to two ‘targeted effects’.
• Early results looks promising
• Observations partly converge with experiences by other authors
15
Thank you !