iSixSigma

Developing Better Capital Estimates

Six Sigma – iSixSigma Forums General Forums Implementation Developing Better Capital Estimates

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #55699

    Veejay Shan
    Participant

    Hi

    I am coaching a project whose objective is to decrease the variation of capital estimates from one new program to another. Currently some of the new product programs have estimates that match actual spending within 5% and some programs have variation up to 50%. I understand there are many variables that may effect this from program to program. I would like to know if there is any example projects that i can use as reference for the above mentioned problem?

    If we dont have accurate estimation it effects the budget down the life of program as we need to scramble for extra money if the estimate is well below the actual.Again these estimates are done well early in the program – like 3-4 years before actual production.

    0
    #201239

    MBBinWI
    Participant

    @veejayshan – I can’t share an example, but can suggest some actions that you might want to take.

    – Have you benchmarked the various projects and methods that were used to develop the capital estimates?

    – What variability is there for the same individual/group in creating the capital estimates? Are there some individuals/groups that are consistently more accurate and conversely less accurate than the mean?

    – Are there characteristics of the projects that can be identified that have higher/lower variability? Perhaps those that have significant construction costs are always more variable, or those that use union labor are more variable, or …

    – One tool that you might want to investigate is Monte Carlo simulation. This method can help you to apply variability to the inputs and see the impact on the outputs. A good tool will provide you insight into the variables that are most sensitive to variation so that you can focus on those to get better data. I once used this method to identify a variable external to those directly controlled by the development team that was going to have a huge impact on the outputs. Justified spending quite a bit of money on understanding this external variable so as to ensure the project benefits were more accurately understood.

    Hope this helps.

    0
    #201243

    Strayer
    Participant

    One piece of advice I can offer from experience is to standardize estimating techniques, including training for project managers/estimators. When you have such wide variation it’s likely that different people are estimating in different ways, and some may just be guessing. When I was a CMM/CMMI lead appraiser I wanted to see that estimates had some basis from past, similar projects/programs and that some statistical method was employed, such as 3-point PERT. If you’re looking at making this improvement as a six sigma project be aware that it will take months or years and it won’t be easy.

    0
    #201248

    Mike Carnell
    Participant

    @veejayshan Part of your issue on the estimates is in your question. They are done 3-4 years in advance. Of course conditions change but more importantly after 3-4 years the estimates were in all likelihood done by someone who is long gone and it happened so long ago nobody owns the numbers. Nobody owns them normally because nobody is holding anyone accountable for them.


    @MBBinWI
    had some good input. Analyze what has gone on in the past and see if there are types of projects, project teams etc that are the ones that are missing. That is the difficulty of mixed quality. If they were all good there would be no question from you. If they were all bad there would be no question because the system would be broken. Mixed results, just like a factory line, when it does good an bad you have to break it down and analyze it.

    The other issue is when “financial” people get involved. If they are truly finance people they will be different from accountants. Financial people are actually very well trained in stats and comprehend variability and distributions. Accountants understand a point. They don’t comprehend (generally) measurement system variability, They understand an average but not a variance. Use accountants in a finance role and the results are generally not that good.

    The other thing you might want to consider is if you understand your own data. If I have a prediction and an actually and I pair them and plot them it should ideally be a 45 degree line – identical pairs. That is pretty unlikely. This analysis will also produce prediction limits and confidence limits. Are your models within the limits? I am guessing you probably only know about the model when it is low. If it is high then it doesn’t cause anyone any extra work so nobody cares. Accuracy and precision should reflect low and high unless you want to train people to sandbag your models.


    @MBBinWI
    is also dead on with the Monte Carlo analysis. One of the things we do is when we lack data to identify the distribution for the simulation we use a uniform distribution since we have no real idea what the outcome will be.

    If you are serious about getting this right then you need to measure it and make someone accountable for the metric. If it is going to be a 3-4 year break then there has to be some review periodically. Make the models/forecasts part of the review so even if someone has left the new people either sign up to it being their number or they update it.

    Just my opinion.

    0
Viewing 4 posts - 1 through 4 (of 4 total)

You must be logged in to reply to this topic.