In a combined meetup between the LimitedWiP Society Melbourne and the Leadership and Project Delivery group, this presentation on forecasting will help the maturity of conversations and catalyse change around predicting the likely outcome of project and product development knowledge work.
6. www.evogility.com.au @danploegdploeg@evogility.com.au
Where to Start
• Understand the start and end points of
the “system”
• Minimum – capture the exit time of work
items in the system.
• Capture start time and work item type.
• You don’t need loads of data to get
started. 15-20 samples will start to give
you an idea
9. www.evogility.com.au @danploegdploeg@evogility.com.au
During a project
Dimitar Bakardzhiev
https://www.infoq.com/articles/noestimates-monte-carlo
• Projects have a
“Z-curve”
• There are really
3 parts with
different
throughputs /
cycle times
• Calculate based
on work in each
of the three
parts
10. www.evogility.com.au @danploegdploeg@evogility.com.au
Dark Matter & Failure Load
Unexpected work items:
• Dark Matter – We often
end up discovering work
that we didn’t anticipate.
The more uncertain the
domain, the more dark
matter.
• Failure Load – Defects
are a natural part of the
work – ignore the effects
at your own peril!
15. www.evogility.com.au @danploegdploeg@evogility.com.au
I have multiple teams –
dependent work
% Time
95 14w
90 11w
85 10w
80 9w
75 9w
% Time
95 4w
90 3w
85 3w
80 2w
75 2w
% Time
95 8w
90 6w
85 5w
80 4w
75 4w
Team A
40-60 Stories
Team B
10-15 Stories
Team C
20-30 Stories
Queue
Queue
3w @90%
certainty
Highly variable
Samples between 2-24w
16. www.evogility.com.au @danploegdploeg@evogility.com.au
I have multiple teams –
dependent work
% Time
95 14w
90 11w
85 10w
80 9w
75 9w
% Time
95 4w
90 3w
85 3w
80 2w
75 2w
% Time
95 8w
90 6w
85 5w
80 4w
75 4w
Team A
40-60 Stories
Team B
10-15 Stories
Team C
20-30 Stories
Queue
Queue
3w @90%
certainty
Highly variable
Samples between 2-24w
Total of 21w
@85
percentile
w/ addition of
Q2 time
As we can predict roughly
when we expect this item
to appear, we can
negotiate sequencing here
to minimize queuing time.
Final forecast is to go with
25w @85 percentile
17. www.evogility.com.au @danploegdploeg@evogility.com.au
What else can go wrong?
• Historical evidence is not necessarily
representative of the future
• Rare, catastrophic events – eg earthquake
destroys office building
• Holiday periods – they have a longer
effect on throughput than what you might
think
• Flow Debt
18. www.evogility.com.au @danploegdploeg@evogility.com.au
What happens next
• Continual forecasting allows you to make
many micro-adjustments (eg modify
scope)
• Adjustments may have impacts on the
system performance
• Allows for conversation to be based on
likely outcomes rather than absolutes
• There is no such thing as absolute
certainty!
21. www.evogility.com.au @danploegdploeg@evogility.com.au
Further Information
Title Author URL
Essential Kanban Condensed David J Anderson and Andy
Carmichael
http://leankanban.com/guide
Focused Objectives Resources Troy Magennis http://bit.ly/SimResources
#NoEstimates Project Planning Using Monte
Carlo Simulation
Dimitar Bakardzhiev https://www.infoq.com/articles/noestimates
-monte-carlo
Probabilistic Project Sizing Using Randomized
Branch Sampling (RBS)
Dimitar Bakardzhiev https://www.infoq.com/articles/probabilistic
-project-sizing
Actionable Agile Metrics for Predictability – An
Introduction
Daniel Vacanti http://www.amazon.com/dp/B013ZQ5TUQ