Agile DC Lead Time

Lead Time:What We Know About ItAnd How It Can Help Forecast Your

Projects

Alexei Zheglov

@az1#agiledc

Goodhart’s Law

Kanban System Lead Time

DeliveredIdeas AnalysisInputQueue

Ready to

Deliver∞325

Development Test

3

Lead Time

The FirstCommitment

Point

AB

C

Discarded

D

Ask Not

DeliveredIdeas AnalysisInputQueue

Ready to

Deliver∞325

Development Test

3

Lead Time

AB

C

Discarded

D

Not “how long will it take?”

Do Ask

DeliveredIdeas AnalysisInputQueue

Ready to

Deliver∞325

Development Test

3

Lead Time

AB

C

Discarded

D

When should we start?

When do we need it?

Decide

DeliveredIdeas AnalysisInputQueue

Ready to

Deliver∞325

Development Test

3

Lead Time

AB

C

Discarded

DOne event

precedes (leads) another one

by this much

One eventprecedes (leads) another

oneby this much

Why?

DeliveredIdeas AnalysisInputQueue

Ready to

Deliver∞325

Development Test

3

Lead Time

The FirstCommitment

PointAB

C

Discarded

D

Includes the time the work item

spent as an option

Depends on the transaction

costs (external to the system)

Measures the true delivery

capability

Customer Lead Time

DeliveredIdeas Activity 1InputQueue

Output Buffer

∞???

Activity 2 Activity 3

?

Customer Lead Time

AB

Kanban system(s) lead time

+time spent in the

unlimited buffer(s)

C

Discarded

D

(Local) Cycle Time

DeliveredIdeas Activity 1InputQueue

Output Buffer

∞???

Activity 2 Activity 3

?

AB

C

Discarded

D

Cycle time is always local

Always qualify where it is from

and to

Often depends mainly on the size of the local

effort

Discussion 1: Gaming Metrics

Readyto Test

Flow Efficiency

F

Q E

C A

J

GD

GYBG

DE NP

P1

AB

Customer Lead Time

Wait Wait WorkWork

IdeasReadyto Dev

5IP

Development Testing

Done3 35

UATReady toDeliver

∞ ∞

Work WaitWork

Official training material, used with permission

Readyto Test

Flow Efficiency

F

Q E

C A

J

GD

GYBG

DE NP

P1

AB

Customer Lead Time

Wait Wait WorkWork

IdeasReadyto Dev

5IP

Development Testing

Done3 35

UATReady toDeliver

∞ ∞

Work WaitWork

Official training material, used with permission

Work is waiting

Work is still waiting!Multitasking creates

hidden queues!

Readyto Test

Flow Efficiency

F

Q E

C A

J

GD

GYBG

DE NP

P1

AB

Customer Lead Time

Wait Wait WorkWork

IdeasReadyto Dev

5IP

Development Testing

Done3 35

UATReady toDeliver

∞ ∞

Work WaitWork

Official training material, used with permission

%100time elapsed

time touchefficiencyflow

Readyto Test

Measuring Flow Efficiency

F

Q E

C A

J

GD

GYBG

DE NP

P1

AB

Customer Lead Time

Wait Wait WorkWork

IdeasReadyto Dev

5IP

Development Testing

Done3 35

UATReady toDeliver

∞ ∞

Work WaitWork

Official training material, used with permission

Timesheets arenot

necessary!

Rough approximations (±5%) are often

sufficient

In Aggregate

Sampling

Readyto Test

Measuring Flow Efficiency

F

Q E

C A

J

GD

GYBG

DE NP

P1

AB

Customer Lead Time

Wait Wait WorkWork

IdeasReadyto Dev

5IP

Development Testing

Done3 35

UATReady toDeliver

∞ ∞

Work WaitWork

The results are often between 1%

and 5%*

*-Zsolt Fabok, Lean Agile Scotland 2012, LKFR12; Hakan Forss, LKFR13

The result is not limited to the number!

What did you decide to do?

If the Flow Efficiency Is 5%...

If... Before After Improvement

Hire 10x engineers 100 95.5 +4.7%

The task is three times bigger 100 110 -9.1%

The task is three times smaller 100 96.7 +3.4%

Reduce delays by half 100 52.5 +90%

Consequences of Low Flow Efficiency

Goodhart’s Law’s Corollary

Start Measuring?

Discussion 2: Measuring Lead Time

Deterministic approachto a probabilistic process?

probabilistic

!!!

0-4 5-9 10-14 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 95-99 100-104

0

2

4

6

8

10

12

14

16

18

20

Example

0-4 5-9 10-14 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 95-99 100-104

0

2

4

6

8

10

12

14

16

18

20

Example

Best-fit distribution:Weibull with

shape parameter k=1.62

Heterogeneous Demand

DeliveredIdeas AnalysisInputQueue

Ready to

Deliver∞325

Development Test

3

AB

C

Discarded

D

E

G

F

H

Demand placed upon our system is differentiatedby type of work and risk

Drill down by project type

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

0-4 5-9 10-14

15-19

20-24

25-29

30-34

35-39

40-44

45-49

50-54

55-59

60-64

65-69

70-74

75-79

80-84

85-89

95-99

100-104

0

2

4

6

8

10

12

14

16

18

20

Mixed data from different types of

projects

4 types, 4 different distributions

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

5-910-14

15-1920-24

25-2930-34

35-3940-44

45-4950-54

55-5960-64

65-6975-79

80-8485-89

100-1040

2

4

6

8

10

12

14

16

18

0-4 5-9 10-14

15-19

20-24

25-29

40-44

55-59

60-64

65-69

70-74

75-79

95-99

0

1

2

3

4

5

6

...

...

Delivery Expectations

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

Shape Average In 98%

1.62

1.23

1.65

3.22

In 85% of cases

30 d

35 d

40 d

56 d

<51

<63

<68

<78

<83

<112*

<110*

<99

Delivery Expectations

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

0-410-14

20-2430-34

40-4450-54

60-6470-74

80-8495-99

02468

101214161820

Shape Average In 98%

1.62

1.23

1.65

3.22

In 85% of cases

30 d

35 d

40 d

56 d

<51

<63

<68

<78

<83

<112*

<110*

<99

The averages are insufficient

to specify delivery capabilities!

The average says nothing about variability!

Needed:the average and a high percentile (usually 80-

99%)

Another Example

0-2.5 2.5-5 5-7.5 7.5-10 10-12.5 12.5-15 15-17.5 25-27.50

2

4

6

8

10

12

Development

0-3 3-6 6-9 9-12 12-15 15-180

2

4

6

8

10

12

14

Support

Shape: 1.16 Shape: 0.71

Weibull DistributionsOccur Frequently

Operations, support (k<1)

New product development (k>1)

Weibull DistributionsOccur Frequently

Operations, support (k<1)

New product development (k>1)

The unique signature of your

process

The unique signature of your

process

Bias

Feedback

How to “Read” a Distribution

Scale

Control

Expectations

Forecast

Mode: how we rememberthe “typical” delivered work

item.Trouble: it’s a very low

percentile.18-28% common.

Median: 50% more, 50% less.

Perfect for creatingvery short feedback loops

Average: we need it for Little’s Law

LeadTime

WIPteDeliveryRa

Little’s Law:handle with care

The 63% percentile isthe best indicator of

scale

High percentiles (80th-99th):critical to defining

service-level expectations

High percentiles (80th-99th):critical to defining

service-level expectations

Statistical process control:Sprint duration in iterative

methods,SLAs in Operations, etc.

Forecasting Cards

While I Was Preparing This Presentation, Somebody Sent Me This...

Discussion 3:Probabilistic or Deterministic?

TestReady

S

RQ

P

ON

F

A Few Words About Projects…

H

E

C

I

G

D

M

DevReady

5Ongoing

Development Testing

Done3 35

UATReleaseReady

∞ ∞

ProjectScope

Official training material, used with permission

Delivery Rate

Lead Time

WIP=

Applying Little’s Law

From observed capability

Treat as a fixed variable

Targetto

achieve plan

Calculated based on known lead time

capability & required delivery

rate

Determines staffing level

Official training material, used with permission

Delivery Rate

Lead Time

WIP=

Applying Little’s Law

From observed capability

Treat as a fixed variable

Targetto

achieve plan

Calculated based on known lead time

capability & required delivery

rate

Determines staffing level

Complicating factors here:

Dark matter“Z-curve effect”

Scope creep

Complicating factors here:Variety of work item types and

risks

Delivery Rate

Lead Time

WIP=

Applying Little’s Law

From observed capability

Treat as a fixed variable

Targetto

achieve plan

Calculated based on known lead time

capability & required delivery

rate

Determines staffing level

Complicating factors here:

Dark matter“Z-curve effect”

Scope creep

Complicating factors here:Variety of work item types and

risks

TestReady

S

RQ

P

ON

F

A Few Words About Projects…

H

E

C

I

G

D

M

DevReady

5Ongoing

Development Testing

Done3 35

UATReleaseReady

∞ ∞

ProjectScope

Lead time data andobserved/measured delivery

capabilityat the feature/user story level

are critical to forecasting projects

The project initiation phase is a great time to

builda forecasting model and

feedback loops

New Kanban Book

Mike Burrows

Influencers

Troy Magennis Dimitar Bakardzhiev David J Anderson

Dan Vacanti Dave White Frank Vega

Discussion 4: What Now?

Alexei Zheglov