return getKanban()

A lighthearted look at the lessons learned from writing a getKanban simulator. Presented at London Lean Kanban Days 2019.

Transcript

1. www.devopsgroup.com | Phone: 0800 368 7378 | e-mail: team@devopsgroup.com | 2018 © DevOpsGroup DOGPublic return getKanban() What cheating at a board game taught me about Kanban
2. I CHEAT AT BOARD GAMES FOR £££
4. 4 @DevOpsGroup © DevOpsGroup DOGPublic Most common question after playing getKanban?
5. 5 @DevOpsGroup © DevOpsGroup DOGPublic The number one question is… What is the highest score you’ve ever seen?
6. Workers
7. © DevOpsGroup DOGPublic Dealing with the Luck Factor
8. 8 @DevOpsGroup © DevOpsGroup DOGPublic The best getKanban strategy is just to roll a six every time !
9. Loading dice is not applicable to the real world
10. Assume an average of 3½ for every roll
11. 3½ is not the average you’re looking for
12. 12 @DevOpsGroup © DevOpsGroup DOGPublic Sam Savage, The Flaw of Averages The Flaw of Averages Plans based on average assumptions are wrong, on average!
13. I need a simple simulation simulator
14. getKanban uses about 85 dice rolls
15. getKanban has nearly as many dice roll outcomes as ATOMS in our GALAXY
16. 16 @DevOpsGroup © DevOpsGroup DOGPublic n=10 0 1 2 3 4400 4800 5200 5600 6000 6400 6800 7200 7600 8000 8400 8800 9200 9600 10000 10400 10800 11200 11600 12000 12400 12800 13200 13600 14000 14400 14800 15200 15600 16000 16400 16800 17200 17600 18000 18400 18800 19200 19600 20000 20400 20800 21200 Frequency Total Revenue ($)
17. 17 @DevOpsGroup © DevOpsGroup DOGPublic n=100 0 1 2 3 4 5 6 7 4400 4800 5200 5600 6000 6400 6800 7200 7600 8000 8400 8800 9200 9600 10000 10400 10800 11200 11600 12000 12400 12800 13200 13600 14000 14400 14800 15200 15600 16000 16400 16800 17200 17600 18000 18400 18800 19200 19600 20000 20400 20800 21200 Frequency Total Revenue ($)
18. 18 @DevOpsGroup © DevOpsGroup DOGPublic n=1,000 0 1 2 3 4 4400 4800 5200 5600 6000 6400 6800 7200 7600 8000 8400 8800 9200 9600 10000 10400 10800 11200 11600 12000 12400 12800 13200 13600 14000 14400 14800 15200 15600 16000 16400 16800 17200 17600 18000 18400 18800 19200 19600 20000 20400 20800 21200 Frequency (0s) Total Revenue ($)
19. 19 @DevOpsGroup © DevOpsGroup DOGPublic n=10,000 0 1 2 3 4400 4800 5200 5600 6000 6400 6800 7200 7600 8000 8400 8800 9200 9600 10000 10400 10800 11200 11600 12000 12400 12800 13200 13600 14000 14400 14800 15200 15600 16000 16400 16800 17200 17600 18000 18400 18800 19200 19600 20000 20400 20800 21200 Frequency (00s) Total Revenue ($)
20. 20 @DevOpsGroup © DevOpsGroup DOGPublic n=100,000 0 1 2 3 4400 4800 5200 5600 6000 6400 6800 7200 7600 8000 8400 8800 9200 9600 10000 10400 10800 11200 11600 12000 12400 12800 13200 13600 14000 14400 14800 15200 15600 16000 16400 16800 17200 17600 18000 18400 18800 19200 19600 20000 20400 20800 21200 Frequency (000s) Total Revenue ($)
21. 21 @DevOpsGroup © DevOpsGroup DOGPublic n=1,000,000 0 1 2 3 4400 4800 5200 5600 6000 6400 6800 7200 7600 8000 8400 8800 9200 9600 10000 10400 10800 11200 11600 12000 12400 12800 13200 13600 14000 14400 14800 15200 15600 16000 16400 16800 17200 17600 18000 18400 18800 19200 19600 20000 20400 20800 21200 Frequency (0000s) Total Revenue ($)
22. 22 @DevOpsGroup © DevOpsGroup DOGPublic Sample Size Accuracy Speed 10 ! " 100 # $ 1,000 % & 10,000 ' ( 100,000 ) * 1,000,000 + ,
23. 23 @DevOpsGroup © DevOpsGroup DOGPublic Sample Size Accuracy Speed 10 ! " 100 # $ 1,000 % & 10,000 ' ( 100,000 ) * 1,000,000 + ,
24. 24 @DevOpsGroup © DevOpsGroup DOGPublic To Do Doing Done Luck
25. © DevOpsGroup DOGPublic Game Strategy
26. Work Item Types
27. 27 @DevOpsGroup © DevOpsGroup DOGPublic Start fixed date tickets about 3 days before they’re due and no sooner.
28. 28 @DevOpsGroup © DevOpsGroup DOGPublic Start intangible tickets as soon as possible
29. 29 @DevOpsGroup © DevOpsGroup DOGPublic Automate deployments, then testing, then the database features
30. Classes of Service
31. 32 @DevOpsGroup © DevOpsGroup DOGPublic Avoid expediting where unless outcomes are (highly) probable and set WIP as low as possible
32. WIP Limits
33. 35 @DevOpsGroup © DevOpsGroup DOGPublic Use upstream limits to control downstream WIP
34. 36 @DevOpsGroup © DevOpsGroup DOGPublic Context-free policies yield unpredictable outcomes in the face of uncertainty
35. Worker Allocation
36. 39 @DevOpsGroup © DevOpsGroup DOGPublic “We are dice, not resources”
37. 40 @DevOpsGroup © DevOpsGroup DOGPublic Allocating workers to work items was by far the hardest policy to model
38. 41 @DevOpsGroup © DevOpsGroup DOGPublic Workers should stick to their own speciality but be prepared to generalise (T- or π-shaped)
39. 42 @DevOpsGroup © DevOpsGroup DOGPublic Stop starting, start finishing
40. 43 @DevOpsGroup © DevOpsGroup DOGPublic Create slack to allow dynamic reallocation to higher-value opportunities
41. 44 @DevOpsGroup © DevOpsGroup DOGPublic Keep your options open: don’t over-commit
42. Prioritisation
43. 46 @DevOpsGroup © DevOpsGroup DOGPublic Prioritise by age of ticket for items already in progress
44. 47 @DevOpsGroup © DevOpsGroup DOGPublic Exception: if you’re not doing automated deployment, there’s more wiggle room
45. 48 @DevOpsGroup © DevOpsGroup DOGPublic Use a financial model when choosing what to select
46. What about the Cost of Delay?
47. 50 @DevOpsGroup © DevOpsGroup DOGPublic
48. 51 @DevOpsGroup © DevOpsGroup DOGPublic
49. 52 @DevOpsGroup © DevOpsGroup DOGPublic Using CD3 and WSJF only improved average outcome by $100
50. Scenarios
51. 54 @DevOpsGroup © DevOpsGroup DOGPublic BOO
52. Head to Head
53. 59 @DevOpsGroup © DevOpsGroup DOGPublic Head to Head Naïve • Prioritise Fixed Date, then Value • No WIP Change • Limit Dice to Cards to 3 • Glen’s Feature: Expedite E2 • Ted Training Champion (So Far) • Prioritise using WSJF • Drop Selected WIP on D11 to 2 • Limit Dice to Cards to 2 • Glen’s Feature: Ignore E2 • No Training for Ted !
54. 60 @DevOpsGroup © DevOpsGroup DOGPublic Naïve vs Champion 0 50 100 150 200 250 300 350 400 4700 5100 5500 5900 6300 6700 7100 7500 7900 8300 8700 9100 9500 9900 10300 10700 11100 11500 11900 12300 12700 13100 13500 13900 14300 14700 15100 15500 15900 16300 16700 17100 17500 17900 18300 18700 19100 19500 19900 20300 Frequency Revenue ($) Champion Naïve Mean $10,900 Mean $16,100
55. © DevOpsGroup DOGPublic Lessons Learned
56. 62 @DevOpsGroup © DevOpsGroup DOGPublic Start with What You Do Now and Evolve Designing policies up front is hard work and usually doesn’t work very well: start simple and improve through evolutionary change.
57. 63 @DevOpsGroup © DevOpsGroup DOGPublic Those Nearest the Problem Should Manage Flow Attempting to manage flow statically in the face of complexity yields poor results: allow those nearest the work to manage flow dynamically
58. 64 @DevOpsGroup © DevOpsGroup DOGPublic Visualisation Makes Life Easier Reading program logs is a poor substitute for good visualisation: displaying state made things much easier to find improvements.
59. 65 @DevOpsGroup © DevOpsGroup DOGPublic Best Results Come From Self-Organising Teams It would be a lot easier if the dice could self-organise ! !
60. © DevOpsGroup DOGPublic One Last Thing
61. 67 @DevOpsGroup © DevOpsGroup DOGPublic What Happens When You Load the Dice? •Total revenue of $28,135 ! •You run out of tickets on Day 21 " •Sometimes you get a lead time of zero days #
62. 68 @DevOpsGroup © DevOpsGroup DOGPublic Contributions Welcome! #LLKD19 https://github.com/seize-the-dave/get-kanban-simulation (Java) @seize_the_dave
63. www.devopsgroup.com @DevOpsGroup team@devopsgroup.com 0800 368 7378 Questions? © DevOpsGroup DOGPublic