How long should a project buffer be?

How long should a project buffer be?

Tom Kemp 1214084339|%e %b %Y, %H:%M %Z|agohover

The classic way of calculating the length of the project buffer for a critical chain is to throw away half the assumed saftey quoted in each task and then put the remaining half at the end of the chain as the project buffer. Has anyone done any research into how much safety really is included in the quoted time for each task in particular industries? Anyway, how do we know that half the length of the trimmed critical chain is appropriate as a project buffer?

The mathematics of combining the skewed probablity distributions of each task in the critical chain are not that simple. The resulting distribution curve for the entire critical chain has an extremely long tail and a very low "body". If we look at the median of this curve then it is way to the right of the sum of the medians of the individual critical chain tasks. So just adding up the (trimmed) estimated task times does not give us anywhere near the correct answer for the duration of the project which gives us a 50-50 chance of finishing it on time, ie the project buffer should start much later than we are told. The fact that projects run using critical chain project management finish on time therefore implies that there is still too much safety in those projects.

On the other hand, if we look at the duration which gives us a 90% probability of finishing the project on time then we find it is way to the left of the the sum of the 90% times for each task. So the overall project is relatively much less risky than any individual task.

The result is that the level of safely required for the entire critical chain (ie the project buffer) is much less (proportionately) than is required for any individual project, which justifies Goldratt's claim that we can throw away much of the assumed saftey in each task. How long should the project buffer be? Unless we know the real distribution curves for the tasks involved then we cannot know and this is why we need evidence from particular industries in order to build some heuristics.

last edited on 1214156990|%e %b %Y, %H:%M %Z|agohover by Tom Kemp + show more

Re: How long should a project buffer be?

cching 1214156383|%e %b %Y, %H:%M %Z|agohover

How long? This long.

Clarke (testing)

Re: How long should a project buffer be?

Productivity Guy 1214342492|%e %b %Y, %H:%M %Z|agohover

Three small pieces:

1. Although my own experience supports Eli's rule of thumb, it's really a rough approximation, dependent on the experience of the estimator with the task involved, coupled with the tolerance of the customer (which is the only real brake on infinitely extending task / project durations). Once people start to get their heads around the CC concepts, their starting estimates are no longer reliably uninformed (i.e. the complex adaptive estimating system has changed - and one would expect that estimate to shorter than when they estimated without the CC insight.)
2. It may be that the the whole measurement thing needs consideration, because there isn't really a minimum time for any particular task - there are, perhaps only two really tangible numbers:
   1. The estimated time - accurate or inaccurate, ambitious or conservative (this is a forecast of how long it should take in the cognitive system)
   2. The actual time taken - in a specific instance or across multiple instances (this can be very precise, but doesn't indicate, repliably, how long it could actually take, because of the impact of the cognitive system on the actual one).
3. One would assume that the proportion of safety allocated should take the level of unpredictability within the environment into consideration, particularly if there is a small number of tasks with wildly assymetrical time probabilities?

Re: How long should a project buffer be?

Tom Kemp 1214766547|%e %b %Y, %H:%M %Z|agohover

I actually have very little practical experience in 'project' environments. My posting was a response to the numbers quoted in the book, 'Critical Chain', and the online, 'Insights'. It was just a warning that there is an inherent assumption that the sum of the ideal median times assumed for tasks could be safely added up to give the median time for the critical chain as a whole. In fact, as the diagram shows (but I haven't bothered to prove it mathematically), this is definitely not the case. I agree with every point you make and that we are dealing with reality. I just hoped that someone may have done some empirical studies in a particular industry (such as software development) to show the true skew of the times it takes to execute common tasks.

The safety required for a critical chain should be the additional time needed from the median time to finish up to, say, a time where we are 90% sure of finishing. Are we just lucky that we can add about half the sum of the medians of the tasks to the length of the critical chain to get about the right amount of safety or, as I'm starting to think, we could have a lot less safety and still hit a due date? I'm just questioning the rule of thumb and wonder how the intuition that came up with it justifies that rule.

last edited on 1214766584|%e %b %Y, %H:%M %Z|agohover by Tom Kemp + show more

Re: How long should a project buffer be?

dagny 1214388797|%e %b %Y, %H:%M %Z|agohover

Using CC tasks to size the PB should be just the starting point for buffer size. Prudent risk analysis, including checking for likelihood of task iterations could cause a considerable change in buffer size.