Exploring Flow Metrics in Kanban Systems AN INTRODUCTION TO KEY FLOW METRICS THAT LOOK INSIDE PROCESSES AND REVEAL THEIR SECRETS ANDY CARMICHAEL @andycarmich [email protected]
Exploring Flow Metrics in Kanban Systems AN INTRODUCTION TO KEY FLOW METRICS THAT LOOK INSIDE PROCESSES AND REVEAL THEIR SECRETS
ANDY CARMICHAEL@andycarmich [email protected]
“Doing Agile” versus “Being agile”
Agile practices Pair programming Daily stand-ups Combining development and
operations processes (DEVOPS) Burn-ups, Burn-downs, CFDs Continuous Integration / Delivery Test-Driven Development Automated build and test “Sprints” (the cycle between
deliveries / plan changes) Story point estimating Multi-disciplinary teams Retrospectives
Agility (the quality possessed by those who are “agile”) Ability to change direction (or
deliver change) at speed Shorter time from idea to value Less waste from a change in the
plan Limited cascade of change from one
area to another Small changes are inexpensive Releases are frequent (and
inexpensive) Resistance to valuable change is low
Kanban Foundational Principles
of change management1. Start with what you do now
including current processes current roles and
responsibilities current job titles
2. Agree to pursue improvement through evolutionary change
3. Encourage acts of leadership at every level in your organisation - from individual contributor to senior management
of service delivery1. …2. …3. …
Watch David Anderson’s blog for more on this topic coming soon…
See Delivery as a “Flow System”
Pool of Ideas
Proposals Selected Development Acceptance Complete
Commitment Delivery
Lead Time
Work in Progress
Delivery Rate Items per time period
Work Item
Flow Systems follow Little’s LawIn 1961 Dr John Little (studying Queuing Theory) proved that, in a stationary system:
λ = L / W
λ is the average arrival rate L is the average number of items in the
queue, W is the average time in queue
Subject to similar assumptions we can apply this to delivery systems:
Throughput = WiP / TiP
The overline indicates the average (arithmetic mean) Throughput (Th) is the rate items depart the system
under consideration. If this is at the Delivery point (and there are no discards) we call this Delivery Rate
WiP is the number of items in the system TiP is “time in process” for an item from entering to
leaving the system (or part of the system) under consideration. We call this Lead Time for the time taken from the Commitment Point to Delivery Point.
Little’s Law is a precise relationship provided the system’s not trendingThat is, either:
The period being averaged is between 2 consecutive points where WiP=0or The system is “stationary”
In a “stationary” system… The age of WiP has not changed significantly over the period The amount of WiP has not changed significantly over the period Every item that arrives, eventually departs
In most Kanban delivery systems neither of these conditions will apply precisely over typical periods of control (e.g. 1-4 weeks)
Exercise – calculate DR, WiP and TiP from arrival and departure dates Then validate your working with Little’s Law
Av DR – WiP/TiP = 0 And plot Control Chart and
Cumulative Flow Diagram
Little’s Law is a fact rather than an aim…
Variability, batches and iterations are not the enemy Remember “value trumps flow trumps waste” But
Predictability is an aim (helped by smooth flow, limited variability, continuous flow)
Flow Debt is undesirable (delivering more quickly now… at the cost of slower times later)
Indicators: Net Flow (Troy Magennis, focusedobjective.com) Delivery Bias (xprocess.blogspot.com) “TiP Deficit” (Dan Vacanti, Actionable Agile Metrics) Age of WiP Indicator (xprocess.blogspot.com)
Buffer Usage (TOC, Dimiter Bakardzhiev) Net Flow (Troy Magennis, focusedobjective.com)
Flow MetricsThe basics…
Delivery Rate Work in Progress, WiP Time in Process, TiP
(Lead Time if between commitment and delivery)
Other metrics indicating “Flow Debt”…Net FlowDelivery BiasTiP DeficitAge of WiP Indicator
delivering more quickly now… at the cost of slower times later
Net Flow( DR – λ ) / TargetTh
Positive if more deliveries than new items
Negative if more arriving than being delivered
Simple / useful indicator Doesn’t look inside the system so not
a predictor of future TiP
Delivery Bias ( Th – WiP / TiP ) / TargetTh
Will be zero in a system which is “stationary” over the averaging period
Will be positive if Throughput is higher than “balanced” and/or WiP is increasing, and/or TiP is lower than balanced
“TiP Deficit”*( ExpectedTip - TiP ) / TargetTiP
Will be zero in a system which is “stationary” over the averaging period
Will be positive if Older WiP is being delivered ahead of newer WiP Age of WiP reducing
Will be negative if Newer WiP is being delivered ahead of older WiP
(expedite lane) Age of WiP increasing
* Dan Vacanti’s “Flow Debt” defined in Actionable Agile Metrics
Age of WiP Indicator( AgeOfWip – TiP/2 ) / TargetTiP
Possibly best predictor of future TiP increases
Age of WiP in a very regular system will be about half the average TiP
Normalised with “TargetTiP” so parameter can be used to compare different systems
Buffer Management Based on Takt Time
(a measure of Target Throughput) Delivery date includes buffer (time) As buffer is used up intervention
may be needed
Source: Dimitar Bakardzhievdimiterbak.blogspot.com
Steve Tendon and Wolfram Müller’s Tame the Flow
Probabilistic Forecasting
“”
The worse mistakes are not the result of wrong answers… but wrong questions PETER DRUCKER