Good & Bad Software Projects...© Quantitative Software Management, Inc. Good and Bad Software Projects: by the Numbers Quantitative Software Management, Inc. 2000 Corporate Ridge,

© Quantitative Software Management, Inc.

Good and Bad Software Projects: Good and Bad Software Projects: by the Numbersby the Numbers

Quantitative Software Management, Inc.Quantitative Software Management, Inc.2000 Corporate Ridge, Suite 9002000 Corporate Ridge, Suite 900

Mclean, VA 22102Mclean, VA 22102Tel: 703 790Tel: 703 790--0055, Fax 703 7490055, Fax 703 749--37953795

Email: Email: [email protected][email protected]: Web: www.qsm.comwww.qsm.com

© Quantitative Software Management, Inc. #2

Framework

• Correlation is not Causation100% of Convicted Felons Have/Had a Mother and a FatherNo Causal Relationship Between Parenthood and Felony

• “There are three kinds of lies: lies, damned lies, and statistics” – Benjamin Disraeli

• What is Measured Tends to be OptimizedFocusing on a Single Factor (Variable) May Lead to Unintended and Undesirable Results


Quotation

The Government is extremely fond of amassing great quantities of statistics. These are raised to the nth degree, the cube roots are extracted and the results are arranged into elaborate and impressive displays. What must be kept in mind, however, is that in every case, the figures are first put down by a village watchman and he puts down anything he damn well pleases.

Sir Josiah Stamp, Her Majesty’s (Queen Victoria) Collector of Inland Revenues, more than a century ago.


Outline

• Project Selection• “Normal” Variability for Schedule & Effort• Best Performing & Worst Performing

Projects Defined• Best & Worst Projects Compared• Differentiators• Non-differentiators• Conclusion• Questions?


Project Selection

• Business IT (Information Technology) Projects

• Completed in Last 5 Years• Confidence Level of Average or Better in

Quality of Metrics• 1509 Projects• 66 Distinct Organizations (Many Divisions

within Companies)


Normal Variability, Time/Cost Trade-off

Service Time Frame Cost

1st Class 1 to 3 days $0.42Priority Mail 2 days $4.80Express Mail Overnight $14.55 - $23.40

U.S. Postal Service 1 Ounce Letter

• You pay a premium for guaranteed quick delivery

• Software functions the same way:

• The relationship between Cost/Effort and Schedule is non-linear


Normal VariabilityProject Duration vs Size

1 10 100 1,000Size (thousands)

0.01

0.1

1

10

100

Sch

edule

(Month

s)

Project Effort vs Size

1 10 100 1,0000.01

0.1

1

10

100

1,000

10,000

Perso

n M

onth

s

Size

Dur

atio

nE

ffort


Normal Variability, Schedule

Project Duration vs Size

1 10 100 1 0000.01

0.1

1

10

100

+1σ

Average

-1σ

2/3 projects between boundaries

4.52 months

16.69 months

Schedule varies 369% from -1σ to +1σ

Size

Schedule


Normal Variability, Effort


1 10 100 1 0000.01

0.1

1

10

100

1,000

10,00

+1σ

Average

-1σ

25.8 effort months

231.3 effort months

Effort (cost) varies 890% from -1σ to +1σ

Effort

Size


Best & Worst Defined

• Projects Often Optimize Schedule at the Expense of Effort (Cost) or Vice-Versa

Time/Effort Trade-off

• Some Projects Optimize Both; Others Fail on Both Counts

• Best Projects are Defined as Being One Standard Deviation or More Better than Average for Both Schedule and Effort

Worst Projects are Just the Opposite


Best & Worst DefinedProject Duration vs Size


0.01

0.1

1

10

100

Sch

edule

(Month

s)



0.01

0.1

1

10

100

1,000

10,000

Perso

n M

onth

s

Best projects 5.4% of total

Worst projects 4.5% of total


Best & Worst Compared: Cost/Effort• For a 10,000 line of code project, the Worst projects average 30.6 times as much effort

For a 100,000 line of code project the Worst projects average 23 times as much effort

• Effort is usually the largest cost component in software development


1 10 100 1,000

Size (thousands)

0.001

0.01

0.1

1

10

100

1,000

10,000

Pe

rson

Mon

ths


Best & Worst Compared: Duration• Worst projects’schedules are 7.3 times as long for a 10,000 line of code project

5.8 times as long for a 100,000 line of code project

Project Duration vs Size

1 10 100 1,000

Size (thousands)

0.01

0.1

1

10

100

Sch

edu

le (M

on

ths)


Best & Worst Compared: Avg. Staff• Worst projects had significantly higher staffing levels

4.2 times greater for a 10,000 line of code project3.8 times greater for a 100,000 line of code project

• All Best Projects had average staff less than 6

Average Staff vs. Size

1 10 100 1,000

Size (thousands)

0.001

0.01

0.1

1

10

100

1,000

Avera

ge S

taff


Best & Worst Compared: Quality 1• Worst projects have far more defects

13.3 times as many for a 10,000 line of code project29.5 times as many for a 100,000 line of code project

• Quality difference increases with project size

Pre-Implementation Defects

0.1 1 10 100 1,000

Effective SLOC (thousands)

0.001

0.01

0.1

1

10

100

1,000

10,000

100,000

Defe

ct Cou

nt


Best & Worst Compared: Quality 2• Few Worst Projects report post-implementation defects• Best Projects trend parallels entire data set; but is slightly better

MTTD (Post-Implementation Quality)

1 10 100 1,000

Effective SLOC (thousands)

0.001

0.01

0.1

1

10

100

1,000

MTT

D 1

st Mo

nth

(Da

ys)


Best & Worst Development Type

% Best Projects by Development Classification

(Unknown)

Conversion (


Best & Worst Development Type• Best projects likely to be New

Development or Minor Enhancements• Worst projects are disproportionately

Major Enhancements


Best & Worst Phase Effort Percentages

Best Projects Effort % by Phase

30.9%

69.1%

0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% 70.0% 80.0%

Analysis/Design

Build & Test

Worst Projects Effort % by Phase

10.4%

89.6%

0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% 70.0% 80.0% 90.0% 100.0%

Analysis/Design

Build & Test

• Best projects allocate nearly 3 times as much effort to Analysis & Design on a percentage basis


Best & Worst: Difficulty

Average Value of Metrics

Overall Difficulty

Data Complexity

Customer Interface

Ext Sys Interface

Integration Complex.

Doc Rqmts

H/W Stability

Sys s/w stability

0 1 2 3 4 5 6 7 8 9 10

Average

5.3

5.7

3.1

2.3

4.0

4.4

9.0

8.4

7.8

5.2

5.8

3.7

4.2

7.2

8.7

7.2


Best & Worst: Personnel


Overall People

Mgmt Eff.

Training Avail.

Staff Turnover

Dev Team Skill

Knowledge

App. Experience

Motivation

Cohesiveness

Communication

0 1 2 3 4 5 6 7 8 9 10

Average

6.9

7.5

4.0

1.8

7.4

7.5

6.3

8.7

7.8

3.3

6.8

5.2

5.3

4.8

5.3

5.3

5.2

5.0

5.5

5.7


Best & Worst:ToolsAverage Value of Metrics

Overall Tools

Testing Tools Cap.

Program. Tools Cap.

Config Mgt Tools Cap

Proj Mgmt Tools Cap.

Dev Std Robustness

Dev Std Adherence

Dev Std Exp.

Dev Std Adaptability

0 1 2 3 4 5 6 7 8 9 10

Average

6.0

4.3

8.2

5.3

5.9

5.6

7.2

7.0

7.6

5.8

4.7

6.2

6.8

6.8

6.7

5.5

6.5

6.0


DifferentiatorsAverage Value of Metrics

Mgmt Eff.

Staff Turnover

Dev Team Skill

Motivation

Cohesiveness

Communication

Knowledge

Overall Difficulty

Customer Interface

Doc Rqmts

Program. Tools Cap.

0 1 2 3 4 5 6 7 8 9 10

Average

7.5

1.8

7.4

8.7

7.8

3.3

7.5

5.3

3.1

4.4

8.2

5.2

4.8

5.3

5.0

5.5

5.7

5.3

7.8

5.8

7.2

6.2

People, Communication, Knowledge

Complexity

Tools


Non-Differentiators


Data Complexity

Integration Complex.

H/W Stability

Sys s/w stability

Overall Tools

Proj Mgmt Tools Cap.

Dev Std Exp.

0 1 2 3 4 5 6 7 8 9 10

Average

5.7

4.0

9.0

8.4

6.0

5.9

7.0

5.2

4.2

8.7

7.2

5.8

6.8

6.5


Non-DifferentiatorsNumber of Projects vs Primary Lang

(Unknown)ABAP

ACCESSAdvantage:Gen

ASPASSEMBLER

C#C++

COBOLCold Fusion

DCLDELPHI

HTMLINGRES

JAVAJava Script

LOTUS NOTELOTUS ScriptManual Test

ORACLEORACLE DEV 2000

PL/1PL/SQL

SQLTRANSACT

VB in ACCESSVBScript

Visual BasicVisual Page

VPF

Prim

ary

La

ng

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Number of Projects

8511111

6 13611 31 31

321111 3311

2 612

Number of Projects vs Primary Lang(Unknown)

ASPBaanC++

COBOLCOGEN

IEFIEF/COOL:GEN

JAVAJCLKSH

MONKOpenROAD

ORACLEORACLE DEV 2000

PACBASESMALLTALKSource File

SQLVB

Visual Basic

Prim

ary

Lan

g

0 5 10 15 20 25 30 35 40

Number of Projects

3611

37

211

411111111111

2


Conclusions

• Projects of the same size and complexity can vary dramatically in cost and schedule

• Major enhancements are a “mine field”: comprising half of worst performing projects

• The amount of effort spent in Analysis & Design is a key differentiator between Best and Worst projects

• Social and leadership factors seem to contribute more to project success or failure than technical ones

• Programming language is not a key differentiator


Questions??

Good and Bad Software Projects: by the NumbersFrameworkQuotationOutlineProject SelectionNormal Variability, Time/Cost Trade-offNormal VariabilityNormal Variability, ScheduleNormal Variability, EffortBest & Worst DefinedBest & Worst DefinedBest & Worst Compared: Cost/EffortBest & Worst Compared: DurationBest & Worst Compared: Avg. StaffBest & Worst Compared: Quality 1Best & Worst Compared: Quality 2Best & Worst Development TypeBest & Worst Development TypeBest & Worst Phase Effort PercentagesBest & Worst: DifficultyBest & Worst: PersonnelBest & Worst:ToolsDifferentiatorsNon-DifferentiatorsNon-DifferentiatorsConclusionsQuestions??

Documents

Good & Bad Software Projects...© Quantitative Software Management, Inc. Good and Bad Software Projects: by the Numbers Quantitative Software Management, Inc. 2000 Corporate Ridge,