Normal (Gaussian) Distribution over Cycle Time

Hello fellow Analysts,

Knowing the future is impossible (not yet), but you could easy guess what would be the outcome when you have a sequence of similar activities. Predictability is something you could gain by collecting meaningful (true) statistical data. With the power of the Analytics module, we at Kanbanize have, you could improve your planning and have better estimations just with a glimpse and few clicks over our Lead Time Chart.

Normal (Gaussian) Distribution and Standard Deviations.
If you present the cycle (lead) time of the cards that went through your board as a normal (Gaussian) distribution, you should have a bell-shaped curve, known also as the Bell curve. The average cycle time of all cards would be the mean of your curve (pointing at the top of the bell).

The 68-95-99 Statistical Distribution

Image source: http://www.muelaner.com/wp-content/uploads/2013/07/Standard_deviation_diagram.png

From a statistical point of view, 68% of all the values are within 1 standard deviation of the mean, 95% are 2 standard deviations from the mean and 99.7% are 3 standard deviations from the mean.
Note: For simplicity, I’m skipping details as formulas, calculations, constants values and the different cases of the normal distribution, skewed to the left or right, normalized and so on. If you are missing this information and want to to know more, just get back to me and I would love to follow up with you.

How to apply the Normal (Gaussian) Distribution into our daily work with Kanbanize.

Large 68% Range

The three horizontal lines on the Cycle Time Chart represent the mean cycle time of all shown tasks and 1 standard deviation from the mean. So, the cycle time average of all tasks is the Mean line. Then the Min and Max lines wrap 68% of that tasks that are within 1 standard deviation of the mean cycle time (the colored range area). From the mean value, we could easy find out what would be the expected time to finish new tasks. From the Max Deviation value, you could easy estimate what would be the worst, but still within the standard deviation time to finish your tasks,  so you could put a safety buffer in your estimation value.

The size of your 68% standard deviation area, depends on the density of your tasks around the mean value. If your cycle time values are scattered over the chart, your deviation range area will be wider. You can’t really predict what would be the cycle time of the next incoming tasks. If you, however, have a history of tasks with very similar cycle time values (high density), then your range will be thinner and you will have a very good predictability on how much time it will take for the new incoming tasks to get though.

Narrow Predictable Range

Note: It really depends on what time exactly you want to measure (lead, cycle, logged or some filtered one by different columns), but in the general case to have a meaningful statistical data, I would recommend to select only cards that are done (i.e. from your Done columns and/or from the Archives). This way the tasks in progress will not distort your ranges. The more cards (history) you have, the better your statistical data should be.

Hint: Once you have your range or cards displayed, don’t forget to use the zoom option of the chart to analyze a sub range that might be of interest to you.

That’s just another way to get closer to predictability. To end this blog post, I’m quoting something I’ve heard and I found it very meaningful especially for any business activity:
“You could only control things that you measure!”

ebook-blog-banner

So, measure, kanbanalyse and control!

This entry was posted in Lean on by .

About Bisser Ivanov

Keen on innovation, exploration or simply trying new things. Would that be a technology, new methodology or just cool gadgets. Got almost 2 decades of experience working as Software Engineer, Team Lead, QA/Processes Manager and Managing Director in mid-size and large scale Software Companies: Prosyst, SAP, Software AG.

9 thoughts on “Normal (Gaussian) Distribution over Cycle Time

  1. Pingback: Version 3.9 – What’s New? | Kanbanize Blog

  2. Jacopo Romei

    Hello!
    I’d like to know how to extract CSV data from my cycle time chart. It would be useful to watch the beautiful Poisson distribution of my own tasks 🙂

    I think normal distributions don’t help that much on estimation, since there’s no pre-known way to reduce variance thus std. deviation. Still though I need to draw my own skewed bell to show some truth to my fellow team mates.

    Thank you!

    Reply
  3. Dimitar Karaivanov

    Hi Jacopo,

    Good question! We are currently ramping up a new department to work on data management and analytics, which should satisfy the finest needs of all users! Requests like this will directly go into the backlog for the new guys, so stay tuned to see a lot of new stuff coming down the line.

    Feel free to share more ideas like this one, that would be appreciated.

    Cheers,
    Dimitar

    Reply
  4. Dimitar Bakardzhiev

    Lead time distributions are asymmetric. Lead time is neither Normally nor Poison distributed. If you check your data you’d find lead time can be modelled using Weibull two-parameter distribution where the different values for the shape parameter make the distribution resemble Exponential (k = 1), Rayleigh distribution (k = 2) and approximates the Lognormal distribution for several values of k(for most populations more thaн fifty samples are required to differentiate between the Weibull and Lognormal distributions).
    This is a good place to start exploring Lead time distributions http://connected-knowledge.com/2014/09/07/inside-lead-time-distribution/
    You can also check Troy Magennis’ work on the topic.

    Reply
    1. Bisser Ivanov Post author

      Hi Dimitar,

      Very good remark! That’s why we have another chart exactly converting high volume sample data into a histogram view. We also apply the Weibull distribution there showing exactly the shape and scale parameters. Take a look at this blog post: https://kanbanize.com/blog/version-4-1-whats-new, the Histogram View and (beta) Weibull Distribution paragraph.

      The Normal (Gaussian) distribution of cycle time is used only for some specific scenarios, like support/ticketing use-cases. There you have similar tasks/tickets with small cycle time variation because of the strict SLAs – 12h, 24h, 48h, etc. The Mean, Median and Mode mostly overlap around those times and having a deviation there is something you really need to pay attention to.

      Best regards,
      Bisser Ivanov

      Reply
    1. Monica Georgieff

      Hi Jody,

      apologies about the confusion. This is one of the older pieces on our blog and personally, I’m not quite sure where the image came from.
      How can we give you your due credit or would you prefer the image be taken down altogether?

      Thanks for letting us know,

      Monica

      Reply
    2. Dimitar Karaivanov

      Dear, Mr. Muelaner,

      Please excuse us for this unauthorized usage of your image. I will personally research this subject and will take measures not to allow it again.

      This is definitely not a standard practice for Kanbanize. Please accept our sincere apology.

      Best regards,
      Dimitar Karaivanov | CEO | Kanbanize

      Reply

Leave a Reply

Your email address will not be published. Required fields are marked *