THINK OF THE RISK SOURCES a project team must examine: company politics, government regulations, bad assumptions, difficult technologies, and so forth. The list is long. But perhaps the most insidious source of risk is the team and its own inability to get agreement from stakeholders on a true project cost or delivery date.
Let's illustrate with an example. A project manager is often told that the project must be completed by a specific date: “We need to have these specifications written and reviewed in three weeks.” In many environments, the project manager will have had no opportunity to participate in the generation or validation of that date. The team must work toward an artificial deadline.
They begin the assignment. Soon (if not immediately) the team recognizes the impossibility of a three-week delivery … The project manager has two options: (1) Inform the stakeholders that the work will be late. (2) Cut corners to meet the imposed deadline. In many project environments, the only choice is Option 2.
In order to meet the three-week delivery, the project manager decides to skip a critical design review. By skipping the review, the project manager has created the possibility for several downstream risks. These risks can affect the quality of the project and the quality of the product. In fact, unless other processes are in place to catch design flaws, the quality of the product will be compromised.
The decision to cancel the design review was made in order to meet an imposed schedule date. This decision could have been avoided had the project team and stakeholders agreed to a realistic date. A date based on a true estimate would have reduced the project's risk.
True project estimates begin with the cost and duration of each individual activity. Each activity estimate begins by establishing a range of likely values. However, before we generate activity and project estimates, let's answer the question, “Why a range of likely values?”
The Evils of a Single-Point Estimate. Traditionally, when estimating the cost or duration of a project activity, one number is generated—a single-point estimate. There are many problems with this approach. A single number does not take into account the variable nature of work. Let's say that one of your project activities is to review the published literature on some aspect of your product design. If your company library has the references, this review could take less than a day. If you must order the documents and wait until they arrive to review them, the activity could take up to a week.
Mark R. Durrenberger ([email protected]) is a co-founder of Oak Associates Inc., a project management consulting and training firm. He has 15 years of experience managing projects ranging from software implementation to high-technology new product development.
Single-Point Estimate: Mean of a Normal Distribution?
Exhibit 1. Is a single-point estimate the mean of a normal distribution? Given enough completed activities, do overs and unders cancel each other out over time?
Single-Point Estimate: The Resulting Distribution
Exhibit 2. The actual activity durations (or costs) probably fall at or above the single-point estimate.
When you sit down with your project management software to build your schedule, what duration do you use? One day? Five days?
Some believe it doesn't matter which duration you choose, because they assume each single-point estimate is the mean of a normal distribution and, given enough activity estimates, the overs and unders will cancel out over time (Exhibit 1).
Is this a good assumption? Let's try to answer the question, using our literature search activity. Do the overs and unders cancel out over time?
Assuming the single-point estimate duration for the literature search activity is two days, the project manager assigns it to a project team member. The team member visits the company research library, quickly reads the reference material and finishes in one day. Is the work turned in early? It depends on the project culture. If it's healthy, then the work is turned in early. More than likely the extra time is used up on further research or other work. From a duration perspective, the project activity is reported complete on schedule, not ahead of schedule.
Let's alter the situation a bit. This time our researcher arrives at the library and discovers that the documents are available only in print and it will take two days to get them on site. The net effect is that our researcher is not able to complete the work according to plan. At the end of two days, the project team member visits the project manager's office and asks for forgiveness. “Sorry, due to circumstances beyond my control,” he says, “I was unable to complete the work according to plan. Can I have more time?” The project manager, not having much choice, grants the extension.
Let's examine the results. In the first situation, the team member finished early but used up the remaining time by doing extra research or other work. In the second situation, the work could not be finished on time so it was delivered late, with the project manager's permission.
The answer to our question, “Do the overs and unders cancel out over time?” appears to be, “No.” In fact, the resulting distribution of actual durations probably looks something like Exhibit 2, where the actual task completion is on, or later than, the single-point estimate. Some have described it this way: The single-point estimate is the earliest a project manager can expect the activity to be reported complete.
True Estimates Are Ranges. To counteract that on-time-or-late behavior we need a change in communication dynamics that begins with a change in our estimation method. Each activity should have its own cost and duration range estimate. Developing a range estimate is a straightforward process. The best way is to use a range of historical values from similar activities. If historical data are not available, try a range-estimating thought experiment.
Range-Estimating Thought Experiment. This thought experiment can be done when generating activity cost or duration estimates. Let's demonstrate it using activity durations.
Start by drawing the triangle shown in Exhibit 3 on a sheet of paper, white board, or flip chart. Be sure to label the low, likely, and high points on the triangle. Next, select an activity from your project plan. Then ask yourself, “How long is it likely to take to complete this activity?” Write your estimate below likely.
Now ask yourself, “If things go very well, what is the shortest reasonable duration required to complete this task?” Write that estimate below low.
Finally, ask yourself, “If things go poorly, what is the longest reasonable duration required to complete this task?” Be creative when answering this question. Think of all the things that have gone wrong in the past or imagine doing the work twice! Write this estimate below high.
After generating the high duration estimate, reexamine the likely duration. Decide if you are still comfortable with it given the severity of the high estimate.
With your first few estimates, you should expect to revise the likely estimate upward. However, as you become more comfortable with the process, your low, likely, and high values will be more and more consistent with each other. Caution: If your triangles are consistently equilateral (the low and high values are equidistant from the likely), then your high estimate is probably not high enough; be creative when generating the high value.
Calculating the Activity Mean. Once you have generated range estimates for each activity, calculate the mean of the three numbers (mean equals (low + likely + high)/3). It is this mean value that will be used when generating your total project cost or duration. Note:This mean value, unlike a single-point estimate, is the point at which the highs and lows will cancel out over time.
Triangular Distribution
Exhibit 3. The triangular distribution provides a realistic model for generating activity costs or durations.
While the range estimates and mean values are key to creating the total project estimate, they can also be used when managing the day-to-day activities of the project.
Let's return to our literature search example. This time the project manager uses a range estimate (low equals 1; likely equals 2; high equals 5; and mean of 2.6 days) when assigning the activity to the researcher. The project manager also gives careful instructions to “Start on this right away, and stay in touch.”
If the researcher finishes in one day, it is clear that the work can be turned in early. If delivery of the research material causes delays, the researcher knows what to report to the project manager and that more time is unnecessary: “I can't finish by the mean duration, but it will be done as soon as possible and earlier than the high estimate.”
Reader Service Number 032
Sample Calculation of Project Duration
Exhibit 4. Using a spreadsheet, you can calculate project mean duration and standard deviation.
Probable Project Duration
Exhibit 5. Cumulative probability of the project finishing in 21 days or less is 50 percent, 24 days or less is 84 percent, and 18 days or less is 16 percent.
Activity range estimates and open communication encourage behavior that allows the overs and unders to cancel out. The activity estimate of 2.6 days is good because it summarizes and balances a range of possible values.
Calculate the True Project Estimate. The next step is to add up the individual activity costs and durations to generate the true project estimate. The calculations for this (easily done with a spreadsheet) follow the equations shown below. Sample project calculations are illustrated in Exhibit 4.
Activity mean = μ = (low + likely + high)/3
Activity variance = σ2= [(high-low)2 + (likely-low)(likely-high)]/18
Mean project cost = Σ activity mean cost
Mean project duration = Σ activity mean duration (using critical path items only) Project variance = Σ activity variance Project standard deviation = project variance
Using the results of the calculation in Exhibit 4, we can generate a project duration estimate bounded by the normal distribution (Exhibit 5).
From Exhibit 5, we can see that the mean project duration of 21 days has a 50 percent chance of being met (cumulative probability scale). The project duration one standard deviation below the mean (18 days) has a 16 percent chance of being met, and the project duration one standard deviation above the mean (24 days) has an 84 percent chance of being met.
Caution: When calculating the duration for a project with many parallel paths (such as fast-tracked projects), this technique will underpredict actual project duration due to path convergence. In those situations, a Monte Carlo analysis is recommended.
The Rational Push Back. Let's return to our initial example. The project manager was charged with delivering a reviewed product specification in three weeks (15 days). If we assume the data from Exhibits 4 and 5 represent the project, we can see why the project manager was frustrated. Based on the calculations, the cumulative probability of completing the project in 15 days is two standard deviations below the mean, or less than 3 percent!
Imagine going to your stakeholders with this information. It will change the conversation dramatically. Instead of feeling at the mercy of your stakeholders, you can use your true estimate to begin a constructive negotiation. “I understand that you would like the work completed in three weeks,” you'll be able to say. “However, my calculations suggest that we have a less than 3 percent chance of meeting that deadline.”
If your stakeholders are listening, they will want to know more (like how you arrived at a probability). You can explain to your stakeholders why they might wish to adjust their deadline or present shortcuts and the consequences of those shortcuts. In short, discuss risks and how to avoid them.
RANGE ESTIMATING IS A SIMPLE technique. With practice, it takes no more time than single-point estimating. The technique allows you to account for the uncertainty and variability inherent in project work. It also allows you to generate a risk-adjusted project estimate. By sharing the range-based estimate with your key stakeholders, you can begin the process of discussing project risks and rational risk taking. You and your stakeholders can make better decisions; decisions that avoid risk events instead of creating them.