Upgrade to Pro — share decks privately, control downloads, hide ads and more …

[ACCU2013] Measure and Manage Flow in Practice

[ACCU2013] Measure and Manage Flow in Practice

6189cdb415d0b7cdbfac8ba2054b2fc1?s=128

Zsolt Fabok

April 12, 2013
Tweet

Transcript

  1. Measure and Manage Flow in Practice @ZsoltFabok http://zsoltfabok.com/ by Zsolt

    Fabok 2013-04-12 Broke the WIP limit TWICE Still on the team #accu2013 http://accu.org/index.php/conferences/accu_conference_2013/
  2. 5 Stories from the life of a team by using

    real application and data The collected data is the courtesy of Digital Natives
  3. #1 Too many open items

  4. WIP point of observation Visualize the situation with Cumulative Flow

    Diagram solved
  5. The Cumulative Flow Diagram Done Started Queued lead time cycle

    time WIP backlog time number of work items It offers more than just the WIP...
  6. ~ ~ ~ ~ ~ ~ ~ ~ ~ ~

    ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ Q D 3 4 1 2 The simplest way of collecting data:
  7. It is not easy to understand and see throughput from

    the CFD A quick detour:
  8. work item WIP lead time /* detour */ time

  9. lead time work item throughput 2 WIP /* detour */

  10. work item 2 2 /* detour */

  11. Let’s add more people to the project so that “things

    speed up”! /* detour */
  12. work item 2 2 3 coordination + communication cost /*

    detour */
  13. Actually, “things slow down”, so it was not a good

    idea (solve the right problem instead - systems thinking). End of the detour.
  14. #2 It takes too much time to deliver

  15. lead time

  16. lead time CFD was not much help here... ...because we

    didn’t know much about the nature of the lead time
  17. Distribution of lead times days count 0 3 5 8

    10 13 15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 22 28 33 56 average median* *Calculation of medians is a popular technique in summary statistics and summarizing statistical data, since it is simple to understand and easy to calculate, while also giving a measure that is more robust in the presence of outlier values
  18. time spent on implementation (hours) time spent waiting (hours) %

    1 63 98 7 57 90 2 62 97 2 62 97 3 61 96 Some examples of work items with 8-day lead time
  19. None
  20. 95% of the lead time was spent on waiting

  21. Distribution of lead times days count 0 3 5 8

    10 13 15 1 4 7 10 13 16 33 Before average 0 3 5 8 10 13 15 1 3 5 7 9 11 13 22 count days After median
  22. #3 Still too many open work items

  23. How many times the item has been rejected

  24. 0 4 8 11 15 31-32 33-34 35-36 37-38 Number

    of rejected work items count week
  25. 0 4 8 11 15 31-32 33-34 35-36 37-38 39-40

    42-43 44-45 Number of rejected work items count week
  26. #4 Being predictable

  27. Sales: “I want to know when the new features can

    hit the market!” Management: “I want to know how much it will cost me!”
  28. All the work items we had so far (~20 work

    items) ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ v v v v v v v v
  29. Categorizing them into three groups ~ ~ ~ ~ ~

    ~ ~ ~ ~ ~ ~ ~ v v v v v v v v S M L
  30. The lead time distribution ~ ~ ~ ~ ~ ~

    ~ ~ v v v v M 0 1 1 2 3 3 4 1 2 3 4 5 6 7 8 9 12 13 16 days count
  31. The lead time distribution ~ ~ ~ ~ ~ ~

    ~ ~ v v v v M 0 1 1 2 3 3 4 1 2 3 4 5 6 7 8 9 12 13 16 SLA days count
  32. The lead time distribution ~ ~ ~ ~ ~ ~

    ~ ~ v v v v M 0 1 1 2 3 3 4 1 2 3 4 5 6 7 8 9 12 13 16 SLA days count Expired
  33. The spent time distribution ~ ~ ~ ~ ~ ~

    ~ ~ v v v v M 0 1 2 3 4 5 6 6 7 8 9 10 hours count
  34. The spent time distribution ~ ~ ~ ~ ~ ~

    ~ ~ v v v v M 0 1 2 3 4 5 6 6 7 8 9 10 SLA hours count
  35. None
  36. #5 Forced improvement

  37. #1 We decided that we would force ourselves to keep

    the SLA #2 Nothing changed. Still the same ratio
  38. Evolution of the team’s workflow

  39. None
  40. None
  41. None
  42. Final thoughts on measurement

  43. “If you can not measure it, you can not improve

    it.” Lord Kelvin image: http://en.wikipedia.org/wiki/File:Lord_Kelvin_photograph.jpg
  44. “If you start measuring something you start optimizing it, and

    I know it's the wrong thing to optimize.” Paul Graham source: http://paulgraham.com/swan.html
  45. image http://gjlh1249.wordpress.com/2011/03/20/waiting-for-the-bus/ The bus stop effect

  46. The Hawthorne Experiments (Elton Mayo, 1297) can explain the bus

    stop effect: Observation may influence the measurements.
  47. “You can easily find data that perfectly suits your argument.”

    Zsolt Fabok (the handsome bloke who is standing in front of you)
  48. Here is a good example

  49. A sad data: “In 2011 drunk driving caused 15% of

    road fatalities in the UK” data: http://news.hastingsdirect.com/drink-driving-caused-15-percent-of-road-deaths-in-2011/ image: http://www.carrentals.co.uk/blog
  50. If 15% of the road fatalities is caused by drunk

    drivers, then 85% is caused by sober drivers. Therefore we can [mistakenly] conclude from data that drunk driving is safer. image: http://www.volker-doormann.org/cavesokr.htm
  51. So let’s advise people to have at least one beer

    before driving... image: http://www.consumeraffairs.com/news/2012/study-shows-big-drop-in-teen-drinking-and-driving-since-1991.html Don’t do that! This is a bad advice based on an even worse conclusion!
  52. The key ideas

  53. 1. We develop software not models (value) 2. Demand first,

    supply second 3. Observe the system (lead time, throughput) 4. Start measuring, look back if necessary 5. Manage 6. Mind that data expires 7. Goto step 3.
  54. Thank you very much for your attention! @ZsoltFabok http://zsoltfabok.com/