Project Management

Use PERT technique for more accurate estimates


Regardless of the technique you use, the tendency in project estimation is to provide one number for each estimate. In other words, if you have 100 activities on your schedule, each activity would have one estimate associated with it. This is generally viewed as the "most likely" estimate.In many cases you can be more accurate by applying a simple PERT (Program Evaluation and Review Technique) model. PERT is an estimating technique that uses a weighted average of three numbers (see below) to come up with a final estimate.

  • The most pessimistic (P) case when everything goes wrong
  • The most optimistic (O) case where everything goes right
  • The most likely (M) case given normal problems and opportunities

The resulting PERT estimate is calculated as (O + 4M + P)/6. This is called a "weighted average" since the most likely estimate is weighted four times as much as the other two values. You'll notice that the final PERT estimate is moved slightly toward either the optimistic or pessimistic value - depending on which one is furthest from the most likely. Generally this ends up moving the final estimate toward the worst case, since the worst case value tends to be further out from the most likely that the optimistic number.

For example, let's say you estimate a piece of work to most likely take 10 hours. The best case (everything goes right) is six hours. The worst case (everything goes wrong) is 26 hours. The PERT estimate is (6 + 4(10) + 26)/6. The answer is 72/6, or 12 hours. Notice that the number was pulled a little toward the far extreme of the pessimistic estimate, but not by much, since the result is still weighted heavily toward the most likely value.

You can use the PERT estimates two ways. You can provide these three estimates for all activities in your schedule or you can only use the PERT formula for those activities that are of high risk. These are the ones where you're not really sure of the estimate so there's a wide variation between the optimistic and pessimistic values.

Speaking of variation - if you subtract your pessimistic value from the optimistic value and divide the result by six, you would have the standard deviation, which is a measure of the volatility of the estimate. In our example above, the standard deviation would be 3.34 ((26 - 6) / 6). The larger this standard deviation is, the less confidence you have in your estimate, since it would mean you have a large range between the optimistic and pessimistic estimates. If the standard deviation was small, it would mean you were pretty confident in your estimate, since the optimistic and pessimistic estimates would be close.

Remember the PERT formula and use it to make estimates when you have a high level of uncertainly.

16 comments
joebailey77
joebailey77

how is 72 arrived - 6+4M +26/6 ???? dont get the same answer...

animeshmehra
animeshmehra

If you have Software Project Estimation experience as a IT Project Manager or Business Analyst or IT Architect then you can earn money at this site. They are looking for experienced people in software project estimation techniques who can associate with us online. Check out. http://www.econcinnity.com

hareendran_e
hareendran_e

It is very useful to clear my understanding

PonderousMan
PonderousMan

This article was very helpful - I am just about to put together an estimate for a small project, but one with a lot of potential variance, since we are dealing with 14 customers at once.

sharathv99
sharathv99

well written article to get a basic idea of PERT. But I didn't quite understand why was the value 6 used as the divisor to divide the sum of O, P and M

BrainSlugs83
BrainSlugs83

@joebailey77 it's not 72, it's 12 (reread it, 72 is before you divide by 6).  It's just basic arithmetic, I show my work below: 

(6 + (4 * 10) + 26) / 6   ==   (6 + 40 + 26) / 6   ==   72 / 6   ==   12.

galleman
galleman

There is an extensive body of literature describing the serious flaws in the use of PERT, or at least in the "belief" in the PERT numbers. http://herdingcats.typepad.com/my_weblog/2007/06/pert-analysis-r.html is a short summary of this topic. While the PERT formula may stimulate discussion about the variances in the completion dates or durations of a project, it should not be used without serious consideration of its 20% to 50% unfavorable bias - optimistic bias - on anything but the most trivial activity networks. DID 81650 - the Federal guideline for the construction of the Integrated Master Schedule - requires a schedule risk analysis ?2.4.1.23 using statistical methods. The standard in today?s project management practice is the Monte Carlo Simulation that removes the variances and errors of PERT.

BrainSlugs83
BrainSlugs83

@sharathv99 Think of it like you're averaging six numbers, 1 of those numbers is the optimistic, 1 is the pessimistic, and the other 4 are the "most likely" repeated four times (because you multiplied the most likely by 4, you're now averaging six numbers -- that's why it's called a "weighted" average -- one of the numbers has 4x the amount of "weight" as the other two).

jslarochelle
jslarochelle

..In practice it is very tempting for the PM to ignore the P. What we have done in some project was to use the method on an individual basis: each team member evaluates his tasks using the OMP and then only the result of the weigthed average is given to the PM. Also, each member should tailor the weigth according to his own tendency to underestimate or overestimate. In practice I don't think the weigth given in the article are a good starting point. A better starting point for me (in software) would be: (O + 3M + 2P)/6 Team members should keep tab of how well they are doing and keep adjusting the weights to get more accurate estimates. JS

daniellgtr
daniellgtr

most estimating methods, I know this is one we use quite a bit with delphi estimation.

ihameed786
ihameed786

Why divide by 6? simply because we are really finding the average of 6 values (most-likely x 4) 1) Optimistic 2) Pessimistic 3) Most Likely 4) Most Likely 5) Most Likely 6) Most Likely

dtitler
dtitler

That's the method to figure the value for standard deviation for the original values given. Six samples or evaluation values with one extreme subtracted from the other extreme divided by the total number of observation yields the chi value (value of one standard deviation).

donstrayer
donstrayer

There are certainly more statistically sound estimating techniques but PERT is relatively simple and tends to be better than "best guess", unless: In a past position, I reviewed numerous projects where the PM obviously gave little thought to O and P and used PERT (a.k.a. 3-point estimating) only because it was required. Most of their estimates were even distributions such as 4,6,8. Feature-rich PM tools allow you to record not only OMP but supporting data for each estimate, and then generate different views of the schedule. I used one that even calculated a shift toward P based on sigma (standard deviation)since a wide range of uncertainty tends to make P more likely. But if you aren't going to take the time to consider what could reasonably go wrong (or right) PERT won't help and you aren't prepared to use more sophisticated techniques either.

Mark Johnson
Mark Johnson

of course in IT O&P are tainted! But turn it around - use the request for O&P to help technicians think about what might go right and what might go wrong. Then help them to explore how to expedite the good and mitigate the bad. And get a better estimate into the bargain!

donstrayer
donstrayer

Can't argue that O&P estimates aren't tainted. My comment was based on what I've seen reviewing hundreds of projects in large enterprises. Most engineering projects rigorously estimate as a matter of course. Internal business projects often rely on educated guesses based on experience. It's hard enough to get them to document the basis for their estimates, quantitative or qualitative, much less to develop three different estimates and the basis for each as PERT demands. Chock it up to different schools of thought. We expect business managers to be strategic, big picture people. PERT can help you come up with a more likely best guess. My point is that if you aren't going to put the effort into estimating O&P it won't help and may even mislead.

galleman
galleman

Don, There is a wealth of literature showing the O&P estimates are tainted in most cases. These "cardinal" values are most often uncalibrated. One way out of this in our experience in aerospace and enterprise IT is to us an "ordinal" value scale (ABCDEF) and assign ranges to those values. These too are uncalibrated and need to be fixed before much confidence can be applied. The primary source of this approach is _Effective Risk Management: Some Keys to Success_, 2nd Edition, Edmund Conrow, AIAA Press, 2003. Ed's Appendix H describes "Some Characteristics and Limitations of Ordinal Scales in Risk Analysis." While neither affordable or accessible (tough reading) Ed's book is mandatory for anyone working the programmatic or technical risk area.