�ݺ�ߣ

PMI Virtual Library
© 2011 Bharat Gera

Project and Program Management
Analytics
By Bharat Gera, PMP

Uncertainty in processes cannot be neglected,
and a predictive approach is the rule of the day.
”
A
ny project executed in today’s marketplace, whether • Cost of US$3000 + US$100 with a 99% confidence
a gaming project launch in Las Vegas, Nevada, USA; interval
a healthcare project in Beijing, China; a government
policy project in the Philippines; or a space launch project Am I asking statisticians to be project managers? Not
at NASA, is based on strong analytical foundations. Today, really! Rather, I propose the basics of statistical analytics,
mankind has a steady influx of high technology, which can which can be learned by tenth-grade students and
be used to harness any information source or intelligence, incorporated in A Guide to the Project Management Body of
including our own mother nature, to gain deeper insights Knowledge (PMBOK® Guide) — Fourth edition (PMI, 2008).
to model and predict underlying process behaviors and Before I detail the enhancements to the project
outcomes with a sheer statistical paradigm. management principles (as per the PMBOK® Guide), I briefly
Uncertainty in processes cannot be neglected, and a introduce the analytical concepts required for this project
predictive approach is the rule of the day. The CEOs, CIOs, management report.
and CTOs (chief technology officers) of today no longer
accept uncertainty as the “ignore” factor when making A Brief Introduction to Analytical Concepts
decisions. As a project manager, I no longer report the Numerical measures of processes and systems have an inherent
estimated project schedule or cost with definitive numbers characteristic, which follows the probability distributions
and make a disclaimer of project risks. Traditional project determined by the underlying process type (e.g., a process
management envisages and supports the legacy paradigm. measure relying on independent accumulated data values).
Today, project managers need to report the project metrics in Essentially, this concept means that any numerical measure of
terms of “analytical certainty,” as Project X has: the system is likely to occur a certain number of times if the
• Schedule of 3PW (person weeks) + 0.5 PW with a 95% system process is repeated in the long term. The metrics of
confidence interval uncertainty is captured in a probability distribution.

Common Probability Distributions Observed outcomes of a number of uncertain quantities, different
in Project Management Scenarios (as per the or the same, the probability distribution of the outcome is
PMBOK® Guide) frequently a normal distribution.
In this section, I consider some common analytical
probability distributions frequently encountered in project The Poisson Distribution
settings. The distributions are “analytical” because they
are the products of analyses that mathematicians and
statisticians have performed on the nature of the underlying
processes that give rise to them. (They conducted these
analyses precisely because the processes are frequently
encountered.) The advantage of having these distributions
is that once we determine that they are appropriate for
a given situation, we can more accurately represent the
uncertainty we face (e.g., in building a project risk model)
and frequently gain important insight into a situation with
much less effort.

Uniform Distribution
We start with perhaps the simplest of distributions in terms of
the underlying process. If each value in a range has an equally
likely chance of being generated by a process, then a uniform
distribution accurately describes the likelihood of getting
individual outcomes. Figure 2: The Poisson distribution.

Normal Distribution (A Fundamental Aggregation The underlying process that gives rise to the Poisson
Distribution Observed in Nature): distribution (Figure 2) is a counting process with an indefinite
number of opportunities for an event to occur. Examples of
such counting processes include the number of refrigerators
sold in a department store in one week, the number of bonds
issued on 1 December 2004, the number of people arriving at
an ATM queue between 11:00 and 11:05 am on a given day,
or the number of change requests (as defined in the PMBOK®
Guide) that a project encounters per quarter. In general,
counting processes or discrete countable independent event-
generating systems are approximated to Poisson processes.
It is also the case that if the average number of
occurrences per interval is large, the normal distribution
gives us a good approximation of the Poisson probabilities.
Figure 1: Normal distribution. The deviation of a Poisson distribution is equal to the square
root of m, the distribution mean. The example below is a
The familiar bell-shaped curve of the normal distribution demonstration of how analytical results can be practically
(Figure 1) is undoubtedly the most recognizable of all applied in project executions.
probability distributions and also the most frequently For example, Project X is expected to hit 24 risks in the
misused. Although many processes in project management one-year schedule. Assuming a Poisson risk process, assess the
(as described in the PMBOK® Guide) and organizational level of risk preparedness required by the project team.
management/businesses give rise to normal distributions, it The above Poisson risk process can be approximated by
is by no means ubiquitous. The underlying process that gives a normal distribution of mean 24 and deviation SQRT(24),
rise to a normal distribution is an accumulation process. assuming the project team is geared to cover risk outages up
Whenever an outcome is really the sum (or average) of the to the upper limit of 99% risk distribution:

PMI Virtual Library | www.PMI.org | © 2011 Bharat Gera
2

Upper Risk Limit = NORMINV(0.99,24,SQRT[24]) ~35 and the task arrival is a Poisson process (rate λ). The time taken
to complete the task is exponentially distributed (service rate µ);
The team has to be geared toward covering 35 risks technically, this is an M/M/1 queue model.
during the year (average of ~10 day turn-around per risk), The total expected waiting time (queue service) is,
ensuring a 99% risk occurrence coverage in the project.
Also note that, only the mean value was used in the
calculation, and the deviation was derived assuming an
underlying Poisson process.
Determining the Appropriate Probability
Exponential Distribution (Decaying Systems) Distribution
Like the Poisson distribution, the underlying process that We must determine which analytical distribution to use
generates the exponential distribution (Figure 3) is concerned for a given problem. There are two major methods we can
with the occurrence of an event. We are concerned with the use: The first method compares the processes that generate
time until the next event occurs, and therefore, the resulting the data; the second compares the data distributions
probability distribution of the time until the next event or themselves. In the first approach, we must define the process
waiting time. The distribution is memory-less in the sense that is generating the experienced uncertainty/variation
that the event waiting time is not dependent on when the last and compare it with those known to generate particular
event occurred. analytical distributions. In the second approach, we generate
a frequency distribution from historical data and compare it
directly with known analytical probability distributions. The
second method is the philosophy behind “best fit” software
packages. If we find a significantly close match between our
distribution and a particular analytical distribution, we can
assume that the underlying process must be of the type that
produces that particular analytical distribution, even if we
do not have a precise definition of the process itself. It is
highly recommended that project and/or program managers
use the first method, given that the underlying processes are
fairly known or provided in most circumstances.
Analytical probability distributions can be very useful, if
and when they apply, and can be used only when the underlying
processes satisfy the conditions necessary for their derivation,
or when you have sufficient data you are comfortable with,
Figure 3: Exponential distribution. assuming the distribution fits. When you do not have sufficient
data, or cannot understand the underlying process sufficiently,
you may want to fall back on simple distributions like the
Queue Models — An Overview uniform distribution to represent uncertainty.

Application of Analytics to Project Management
Principles (as described in the PMBOK® Guide)
I used to skip management classes on analytical theory and,
here I am, writing a three-page article on analytical theory!
Well, I’m glad I just quantitatively used that statement of
pun… Let’s begin with real-time analytical application in
Figure 4: The basic M/M/1 queue model. project management.

We will just touch upon the basic queue model (Figure Risk Management
4), given the model’s relevance to analytically estimate project Risks in project execution typically arrive at a Poisson process
schedules. We assume we have a single task-processing system with the inter-risk time exponentially distributed.

3

For example: The oil & gas exploration industry has its implementation. The law required specific proportions
matured over the past century. Project data analyses over of bio-ethanol to be blended with gasoline. The prime
all initial project launch executions have revealed that these intent was to boost the local production of biofuel feedstock
start-up projects encounter an average of approximately six and reduce imports from low-cost production economies
risks within the first year. Ralance Industries, in conjunction (e.g., Brazil). The project management of the bio-ethanol
with the State Government of Mumbai, has jump-started supply chain network for local production was based on
an oil exploration project in Mumbai’s Southern Coast. highly optimized linear analytical models calculating the
The project manager of Ralance Industries is faced with the production requirements of bio-ethanol in various locations
unique challenge of lack of government cooperation in risk across the Philippines. The complete details are beyond the
management for the next 30 days. What should the project scope of this article; nonetheless, a basic appreciation of
manager recommend to the CEO of Ralance Industries in his optimization requirements and the importance of project
or her project status report? management costing to policymaking is the intent of listing
These are real-time situations that global project managers this case.
will need to deal with. The project decisions made today must
have strong analytical foundations. An analytical approach Case Study 2
with this background is to estimate the chances of Ralance The World Health Organization (WHO) has decided to
Industries being hit with a risk within the next 30 days. fast track its initiative to reduce the rapid spread of HIV in
Using the MS Excel® EXPONDIST formula to calculate Kenya. The pilot rollout project aims to reduce spread of
the additive/cumulative probability of the occurrence of a risk the disease and provide treatment to 60% of the infected
event in the next one month, where population. The expected average cost of medication for
those under 20 is US$10,000, US$25,000 for those in the
P(T <= 30) = EXPONDIST(1/12, 6,1) = 39% 20- to 35-year-old age group, and US$15,000 for those over
the age of 35. The National Health Institute of Nairobi
The distribution λ(rate) is assumed to be 6/year in the has reported that HIV cases are normally distributed with
above calculation. a mean age of 30 and a standard deviation of 7, and also
The project manager should clearly indicate that the estimates that there are six million people infected with
project is subject to an exposure of a 39% chance of being the disease.
halted due to lack of government support. In practical terms, The project plan and budget details have to be
Ralance Industries may go with a risk of less than 40% against submitted for the funding. How does the involved project
halting the project. office calculate the best execution plan and cost of the
project?
Cost Management The statistical information should be leveraged for
Project cost is a vital piece in today’s global project this case. As per the normal distribution (Figure 5),
management. Analytics in cost optimization has become approximately 68% of the population is covered in one
more vital because of the uncertainty of global economic standard deviation from the mean (either side).
equilibrium. Growth countries (e.g., India) have high The plot illustrates that approximately 68% of the
inflation rates and face the heat of inflationary forces, whereas infected population is between the age group of 30 +/−7
more mature markets (e.g., the United States) are being pulled (i.e., [23–37]). This suggests that the project plan should
into short-term imbalances of credit investments without target the age group of 23 to 37 for the initial project
underlying dollar generation. rollout. This will ensure that the WHO guideline of 60% is
Project management principles (as per the PMBOK® met and the highly dense HIV distribution is targeted.
Guide) rightly advocate the principle of lowering project The project costing has to be split for the age groups [23–
execution costs. Analytics brings in enormous power to this 35] and [35+]. The percentage of population covered in this
decision-making factor. age group can again be calculated from the distribution graph.

Case Study 1 Population covered under [23–35] =
The Philippine government passed the Biofuel Act in 2006, NORMDIST (35,30,7,1) - NORMDIST (23,30,7,1)
and the agriculture department was in charge of overseeing = 60.4%

4

HIV Infected Age Distribution
0.06

0.05

0.04

0.03 HIV Infected Age
Distribution
0.02

0.01

0
0 20 40 60

Figure 5: HIV-infected population age distribution.
Figure 5: HIV-infected population age distribution.
The plot illustrates that approximately 68% of the infected population is between the age
Population covered under [35,∞ ] (i.e., [23–37]). This suggests that the project plan should target the age driven
group of 30 +/−7 = The project service level agreement should be
group of 23 to 37 for the initial project rollout. This will ensure that the WHO ensure customer
1 – NORMDIST (35,30,7,1) from the calculated response times to guideline of
= 23.6% 60% is met and the highly dense HIV distribution is targeted.
commitments, in addition to avoid levying any penalty
charges to the organization.
Cost = [25K * The project costing has to be* split for the age groups [23–35] and [35+]. The percentage of
60.4% * 6 M] + [15K * 23.6% 6 M] =
population covered in this age group can again be calculated from the distribution graph.
$US11.2 billion Conclusion
I have made a sincere effort to demonstrate the modeling of
Schedule Managementcovered under [23–35] =
Population and Resource Allocation project management principles to include statistical business
NORMDIST (35,30,7,1) - NORMDIST enable projects to execute optimally in all subject
One of the challenges faced in maintenance projects is analytics to (23,30,7,1)
the uncertainty associated with = 60.4%
defect inflow from end areas. By no means do I expect project managers to be
consumers. Scheduling and budgeting for these processes statisticians; but a culture of leveraging statistical know-how
are also prone to high inaccuracies in estimating the defect to processes can be a key factor these days. I hope the project
closure rate. Population covered under [35, ] = management principles driven by Project Management
1 – NORMDIST (35,30,7,1) Institute will be enhanced over time to add relevant analytics
It is worthwhile modeling such projects as an M/M/1
= 23.6%
queue to analytically estimate the characteristics of the system, in key project calculations and evaluation procedures.
such as the average system response delay to incoming requests
Cost = [25K * 60.4% * 6 M] + [15K * 23.6% * 6 M] = Author
or defects. In general, decaying systems and processes fall under About the$US11.2 billion
the umbrella of Poisson and Exponential Distributions. Bharat Gera is a line manager at IBM Corporation. He
For example, consider a level 3 software maintenance has been involved in the project management of enterprise
project that has historically encountered ten defects from software and hardware projects over the past decade and
Schedule Management and Resource half. He has worked primarily in storage, database,
the market end customer per month. The arrival rate can a Allocation
be approximated to be a Poisson process with a rate data warehouse, and analytics domains. He is a Project
One of the challenges faced inproject has a projects is the uncertainty associated withexecutive
maintenance Management Professional (PMP) and has an defect
λ = 10/month. The team working on the level 3 ®
collective average throughput of 24 defects per Scheduling and management degree with a specialization in prone toanalytics
inflow from end consumers. month, which budgeting for these processes are also business
is equivalent to a service rate of µ =in estimating the defect closure the Indian Institute of Management, Bangalore. He can
high inaccuracies 24/month. from rate.
Modeling the system as an M/M/1 queue, the average be reached at bharat.gera@alumni.iimb.ernet.in
It is worthwhile modeling such projects as an M/M/1 queue to analytically estimate the
response time of the system is 1/(μ − λ) = 1/(24 − 10) =
0.0714 monthcharacteristics of the system, such as the average system response delay to incoming requests
~ 2 days. Any defect being logged will take an Disclaimer: The views expressed here are strictly the personal
or defects. In general, decaying systems and processes fall under the umbrella of Poisson andglobal
average of two days for closure. thoughts of the author as deemed applicable to today’s
Exponential Distributions. project execution.

9
5

�ݺ�ߣ

Analytic Driven Management

More Related Content

Analytic Driven Management