Zsolt Fabok
April 12, 2013
1.5k

[ACCU2013] Measure and Manage Flow in Practice

April 12, 2013

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
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
/* detour */
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

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

19. 95%
time was spent
on waiting

20. 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

21. #3 Still too many open
work items

22. How many
times the item
has been
rejected

23. 0
4
8
11
15
31-32 33-34 35-36 37-38
Number of rejected work items
count
week

24. 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

25. #4 Being predictable

26. Sales: “I want to know when the
new features can hit the market!”
Management: “I want to know
how much it will cost me!”

27. All the work items we had so far
(~20 work items)
~
~
~
~ ~
~
~
~
~
~
~
~
v
v
v
v
v
v
v
v

28. Categorizing them into three groups
~
~
~
~
~
~
~
~
~
~
~
~
v
v
v
v
v
v
v
v
S
M
L

29. 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

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
SLA
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
Expired

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

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

34. #5 Forced improvement

35. #1 We decided that we would
force ourselves to keep the SLA
#2 Nothing changed.
Still the same ratio

36. Evolution of the
team’s workﬂow

37. Final thoughts on
measurement

38. “If you can not measure it,
you can not improve it.”
Lord Kelvin
image: http://en.wikipedia.org/wiki/File:Lord_Kelvin_photograph.jpg

39. “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

40. image http://gjlh1249.wordpress.com/2011/03/20/waiting-for-the-bus/
The bus stop effect

41. The Hawthorne Experiments
(Elton Mayo, 1297) can
explain the bus stop effect:
Observation may inﬂuence
the measurements.

42. “You can easily ﬁnd data that
perfectly suits your argument.”
Zsolt Fabok
(the handsome bloke who is standing in front of you)

43. Here is a good example

44. A sad data:
“In 2011 drunk driving caused 15% of
road fatalities in the UK”
image: http://www.carrentals.co.uk/blog

45. 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

46. 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!

47. The key ideas

48. 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.

49. Thank you very much for your attention!
@ZsoltFabok
http://zsoltfabok.com/