iSixSigma

Analysis of a dynamic system

Six Sigma – iSixSigma Forums Old Forums General Analysis of a dynamic system

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #51037

    Sorour
    Participant

    Hi all- an analysis query for you…
    Background:
    I have recently implemented a prioritisation model for all IT change work in my company. Items are given a score (which is a function of determined value and costs), and then ranked 1 to n in asecending order based on this score. The theory is that items will then be addressed in this order. This, however, presumes that the required development skills are available. For example, items ranked 1 & 2 may need specialist skills, only possessed by one person. Therefore, items 1, 3, 4… will be being actioned, while 2 awaits the recousce currently working on item 1 to finish.
    To help identify areas of skills shortage and its impact, I also plan to implement a breakdown of costs per skill when work is first requested (i.e., a total cost of 10k could be seen to consist of 6k of SQL, and 4k of Excel). The amount of work in the queue could then be compared to the amount of available resource for each skill.
    Question(s):
    My initial thoughts on how to analyse centered around areas like:
    a) What percentile of work in the queue is being actioned?
    b) Based on skills profile of the queued work, how many more FTE (possessing which skils) would be required to be able to action an additional X% of the queued work items?
    However, I am at a loss as how to do this as the system is dynamic (i.e. there are items being created and completed everyday, meaning that the profile of prioritisation scores and calculated ranks are constantly changing.
    In short:
    1) Does the above analysis approach (a&b) strike you as the best way forward?
    2) How would this be achieved with the dynamic nature of the data?
    3) Should I just quit my job and open a bar in Mexico?
    Many thanks.
    Paul

    0
    #176275

    Vallee
    Participant

    Paul,
    When rating your IT issues, what’s worse: a system with numerous defects but crashes infrequently; or a system with limited defects but crashes often? The reason I asked this is to see how it may change your rating perspective. If you base your risk rating on types of errors that are critical to crashes instead of the frequency of certain defects, this would allow you to manage the dynamics of what is reported. You can also look for downtime and error during repair for certain types of crashes.
    Look at the right in the blue section under channels and you will see a software/ IT forum link.
    HF Chris Vallee
     

    0
    #176276

    Sorour
    Participant

    Hi Chris- thanks for the reply.
    The question I have, however, is not based around the quality of the ratings system which has been developed (we acknowledge that this may need refinement and I haev built in several manual touch points to accommodate for a review of calculated ratings).
    It is more around the appropriate statistical methods which should be used to best mine the data being produced by the constantly changing nature of the ranking data and how the skills required to complete some or all of these work items relate to the skills profile within existing headcount.
    Kind regards,
    Paul

    0
    #176277

    Szentannai
    Member

    Hi Paul,
    I would use a simulation aproach. At a first glance it seems to me that you might be able to reformulate the prblem in terms that can be handled by Crystal Ball.
    If this is too difficult, maybe you can google for some queueing simulation system. I would define a few scenarios (normal load, high load, catastrophe) and run them past the simulation.
    You might want to check the ATAM methodology at the SEI webpage for guidelines on defining scenarios – they are really good IMHO.
    hope this helps
    Sandor

    0
    #176294

    JustaGeek
    Participant

    The methodology we are trying to implement is one closer to a typical manufacturing situation.   Where the skill sets of the people are analogous to machine capabilities.  This then becomes a somewhat traditional Kanban situation.
    each development priority is broken down into “segments” or “features” that becomes a piece of “inventory” managed. 
    The statistics then center around throughput of the “features” (aka leadtime), the number of “features” being completed (aka inventory turns), % of overall project complete(we developed a point system to measure relative value of a “feature” which then allows the development team to quickly concentrate on the value added “features”), etc.
    We haven’t done this (at least yet, perhaps we will) but each skill could be looked at on a “utilization” basis as well and standards set for what a “good” utilization rate is for a given skill set

    0
    #176299

    Vallee
    Participant

    Paul,Using Wheeler’s thoughts, your quality rating system (data collected) should be developed on what you expect the data to be used for. Therefore, your are in a holding pattern until your understand the data better. While your type ranking pattern may vary from day to day, there should be some commonality. One data mining technique deals with looking for common data fields used in reports to see who creates them, who edits them, and which one’s are not used once created to weed out the chunk.You also need to observe what the employees work on daily and identify the core skills, core tasks, and how they are trained. This will help you decide if you need to cross train, have a floating specialists, or segregate types of errors.I know I have not answered your question fully but I believe you need to understand your data, your process, and what you truly need to collect.HF Chris Vallee

    0
    #176306

    Severino
    Participant

    Paul –
    If you are attempting to manage live load, you will need to define a period of time within which you will assume conditions are fixed.  This window should be based on how quickly you can add/remove/reschedule resources in reaction to your data.  You then move to the next review period and manage the exceptions along with the new demands.
    You should also be trending your historical performance.  Look at the on-time-to request of your closed tickets and the reasons you were late.  Don’t assume that is soley resource driven as your own prioritization may be causing queue times for certain issues to extend beyond acceptable limits.

    0
Viewing 7 posts - 1 through 7 (of 7 total)

The forum ‘General’ is closed to new topics and replies.