The PERT system was developed for the Polaris Missile Program in the 1950s. At that time there was a lot of pressure on the United States Navy to complete the Polaris Missile Program. The Cold War was raging, and the United States needed a deterrent that would discourage the threat of nuclear war with Russia. A mobile missile that could be carried aboard a submarine and launched from beneath the surface of the sea would be a formidable weapon.
The problem for the U.S. Navy was that there were two separate projects to be done. One was to develop a submarine that could launch these missiles. The second project was to develop a missile that could be launched from a submarine. The durations of the project plan activities had a great deal of uncertainty in them. The navy needed a method to predict the project schedule with better reliability than was possible in the past. PERT was developed to assist in analyzing projects where there was uncertainty in the duration of the tasks. The normal probability distribution relates the event of something happening to the probability that it will occur. It turns out that by experiment, the normal distribution describes many phenomena that actually occur. The duration as well as the estimated cost of project activities comes close to matching a normal distribution. In reality, another distribution, called the beta distribution, fits these phenomena better, but the normal curve is close enough for practical purposes. If it were possible for this project to be done thousands and thousands of times, sometimes the time to do the activity would be 35 days, other times it would be 33 days, and still other times it would be 37 days. If we were to plot all of these experiments we would find that 35 days occurred most often, 34 days occurred a little less often, 30 days even less, and so on. Experimentally, we could develop a special probability distribution for this particular activity. The curve would then describe the probability that any particular duration would occur when we really decided to do the project and that task.
In the experiment, if 35 days occurred 134 times and the experiment was performed 1,000 times, we could say that there is a 13.4 percent chance that the actual doing of the project would take 35 days. All 1,000 of the activity times were between 20 and 50 days. It is impractical to do this activity a thousand times just to find out how long it will take when we schedule it. If we are willing to agree that many phenomena, such as schedule durations and cost, will fit the normal probability distribution, then we can avoid doing the experiment and instead do the mathematics. To do this we need only have a simple way to approximate the mean and standard deviation of the phenomena.
The problem for the U.S. Navy was that there were two separate projects to be done. One was to develop a submarine that could launch these missiles. The second project was to develop a missile that could be launched from a submarine. The durations of the project plan activities had a great deal of uncertainty in them. The navy needed a method to predict the project schedule with better reliability than was possible in the past. PERT was developed to assist in analyzing projects where there was uncertainty in the duration of the tasks. The normal probability distribution relates the event of something happening to the probability that it will occur. It turns out that by experiment, the normal distribution describes many phenomena that actually occur. The duration as well as the estimated cost of project activities comes close to matching a normal distribution. In reality, another distribution, called the beta distribution, fits these phenomena better, but the normal curve is close enough for practical purposes. If it were possible for this project to be done thousands and thousands of times, sometimes the time to do the activity would be 35 days, other times it would be 33 days, and still other times it would be 37 days. If we were to plot all of these experiments we would find that 35 days occurred most often, 34 days occurred a little less often, 30 days even less, and so on. Experimentally, we could develop a special probability distribution for this particular activity. The curve would then describe the probability that any particular duration would occur when we really decided to do the project and that task.
In the experiment, if 35 days occurred 134 times and the experiment was performed 1,000 times, we could say that there is a 13.4 percent chance that the actual doing of the project would take 35 days. All 1,000 of the activity times were between 20 and 50 days. It is impractical to do this activity a thousand times just to find out how long it will take when we schedule it. If we are willing to agree that many phenomena, such as schedule durations and cost, will fit the normal probability distribution, then we can avoid doing the experiment and instead do the mathematics. To do this we need only have a simple way to approximate the mean and standard deviation of the phenomena.