[ACCU2013] Measure and Manage Flow in Practice

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
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
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
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/