Agile Estimation Using Functional Metrics Parts 3 and 4

Agile estimation using functional metrics is designed to cover the product and release rings of Cohn’s planning onion using a synthesis of parametric and Delphi estimation techniques with the emphasis shifting from parametric to Delphi as events dictate.    The technique leverages the ability to size requirements to develop parametric estimates and then dovetails collaborative techniques to refine those estimates based on memories and self-knowledge.

The process flow is as follows

Stage One

1. Identification of functional requirements (or stories)
2. Sizing using Quick and Early Function Points
3. Simple Parametric Estimation

Stage Two – Sprint or Iteration Planning

1. Break Requirements into More Granular Pieces (if needed) and Refine Size
2. Team Level Re-estimation of Requirements Using Delphi Techniques
3. Team Level Commitment

Size is a critical component for developing an estimate and for planning however size and estimates are not synonymous.  Size is merely a step along the path from point A to point B.  As we move along the path size will be revisited twice.

The sizing process begins by segregating the functional requirements from the non-functional requirements.  The functional requirements are then sized using Quick and Early Function Points (QEFP).  Function points for all their warts are the easiest way to consistently size software requirements.  This is accomplished by focusing on the basic building blocks of functionality found in all software projects. The QEFP method leverages the relationship between action verbs and transactions to identify the transaction functions found in function points and the subject of the requirement to identify the data functions found in function points.  The application of this technique is similar to sentence diagramming that you learned during grade school or high school.  This relationship between words and size has been observed and investigated over the past few years by a number of different people within the functional sizing community including myself (see “Turning Perfect Good Words Into Numbers” originally presented at the IFPUG Functional Sizing Summit at www.davidconsultingroup.com).  Function points or any functional metric has at its heart the goal of converting requirements or stories into a number in a consistent manner.  The number then must be interpreted based on the abilities of the team or organization.  At the level of the overall project, the technique described in this paper leverages parametric estimation.  A simple parametric estimation equation could be:

Y =-(7^-6*(X^2))-Z*X+26.587

Where:

X = Size in function points

Y = Productivity rate for the type of project

Z = Behavior or Process Index

The result is a productivity rate for the project.  Collection of historical data on a selection of projects will be required to build an equation.  I would further suggest that you will need to augment internal data with external data to increase the validity of the estimation equation.  Note the factor can be applied to a disaggregated requirement, epic or story.  This estimate is created for organizational planning purposes.

Stage two begins as sprints or iterations are kicked off.  The sprint teams (we will use SCRUM terminology) breakdown the stories into pieces that can be accomplished during the sprint, then re-size the pieces using the QEFP method.   The goal of using QEFP at this point is to take one source of variance out of the estimation discussion that will be had when the Delphi method is applied.  This focuses the group on expanding the team level self-knowledge needed to coalesce on an appropriate level of effort needed to complete the story.  For example the QEFP technique has been combined with planning poker into a process that was quickly learned by the sprint teams.  The results were a marked reduction in stories that were not completed during the sprint they were committed to in.  By removing size as a variable the team that initially piloted this method indicated they were better able to focus on discussing team capabilities and technical considerations when doing their initial sprint planning.

Real life examples will help drive how organizations synthesized what could be considered competing methodologies into something greater than the sum of the two parts.

Part 4

Case One:

• Firm:   Small custom technology organization
• Project Types:  Internal and external projects
• Culture: Highly collaborative
• Current Methodology:  Mixed waterfall and SCRUM

Other notes:  All external projects are bid with many using a fixed fee structure.  Internal projects were continually re-scoped to fit the internal development budget which changes as the economy waxes and wanes.  Prior to rewriting the estimation process and leveraging functional metrics approximately 30% of bids were successful and budgets tended to be a suggestion.  The lack of estimation success meant that there was a significant risk of losing money if the business was won and if the business was not won going out of business in the long run.

The firm adopted Quick and Early Function Points for sizing the backlogs for all projects, both internal and external.  Where backlogs were not being used to manage requirements they were developed.  A quick baseline was developed to determine a productivity factor.  The productivity factor was then used to translate individual stories into effort.  Each team spent a day reviewing how they worked together to generate a baseline of self-knowledge and trust.    Collaborative story level estimation was redone using planning poker.  It became apparent quickly that a lot of disaggregation was needed to actually estimate the backlogs.  After applying QEFP and the productivity factor to the in-flight projects the firm progressed to applying QEFP to all bids.

The results were that won bids increased 20% and negative misses were nearly eradicated.  A negative miss was defined as underbidding on a fixed bid contract.

Case Two:

• Firm:   Large software development firm
• Project Types:  Internal projects (software for resale)
• Culture:  Hierarchy, classic command and control
• Current Methodology:  Mixed waterfall  (but SCRUM recently introduced)

Other notes:  The methodology in the environment was predominantly classic waterfall with central PMO.  Just before we readdressed the estimation process a team had implemented SCRUM and some components of extreme programming (xP) this was done in sort of a guerilla fashion.   One very large project was consuming the majority of the organizations resources.  Significant requirements were still being discovered after construction had begun.   Estimates had been developed based on a bottom up process very early in the project and they were of questionable validity.  The top managers just returned from begging for more money from the board of directors.  The project was being capitalized.

The solution in this case was for the company to more firmly embrace the SCRUM framework for project management at the team level.   A product backlog was developed and Quick and Early Function Points were adopted to size the backlog.  The sizing process exposed a number of functional blind spots (function points can be leveraged as form of analysis).  Team members were trained in using QEFP which allowed them to size new stories or re-size changed stories at an individual sprint planning level.  The impact of these changes was to allow the PMO to size and preplan the backlog with development leaders and the primary product owner.  The full product owner team selected stories to formulate sprints during the sprint planning meetings.  Teams re-evaluated (sized and estimated) the selected stories and then committed to the stories they could do.

The initial result was improved product owner satisfaction, involvement allowed them to be part of the solution.  After the first few sprint there was an increased perception of estimate consistency both at the product backlog and sprint team level.  It was also noted that the teams that had been using the using SCRUM before adopting the new estimation methods had a reduced number of stories that had not completed at the end of the sprint.  The teams attributed this improvement during retrospectives to a better capacity to size and commit to work that they’re actually able to accomplish during the sprint.

Summary

Agile methods have matured and are now being integrated into many different approaches to the development of software.  Estimation has been problematic for all methods; agile to plan based therefore it tends to be a lightning rod for experimentation and synthesis such as is being described in this paper.   Agile Estimation Using Functional Metrics has presented a path for integrating the discipline found in functional metrics with the collaborative approaches found in agile estimation.

 

Agile Estimation Using Functional Metrics Part 2

Uncertainty:

A lot has been written about uncertainty, mostly from the point of view of requirements, however the impact of uncertainty extends further than requirements into factors that can be purely technical (whether specific coding languages can do the job) to the complexities of the real world (cue the changing economy as an example).  If we change our perspective to completion of the project, I propose that we will all admit that level of project uncertainty is substantially reduced the closer you are to completion of the project.  Moving back toward the beginning of the project where most estimation exercises occur one simple truth becomes apparent, knowledge dispels uncertainty.  We need to gather better knowledge whether by leveraging history, mathematical algorithms and/or project specific information to make better estimates.   Integrating agile techniques for knowledge capture in projects are tools for reducing uncertainty.  Techniques to use include incorporating a user or user proxy on the team, focusing on short pre-defined time horizons, implementing processes that foster communication and periodic re-planning.

Self knowledge:

Two psychologists, Joseph Luft and Harry Ingham developed a construct to understand personal awareness.  The tool named Johari’s Window[i] divides personal awareness into four different categories, as represented by its four quadrants: open, hidden, blind and unknown. The lines dividing the four panes are like window shades, which can move as an interaction progresses.  The concept is adaptable to teams. Team level blind spots complicate estimation, planning and ultimately performance.  Techniques to improve a team’s self knowledge include forming stable teams, fostering intimate communications and ensuring retrospectives actually happen often.  These tools minimize what isn’t known by the team and surface misunderstandings quickly.

Consistency of Method

There are several types of estimation that can be leveraged during any project.

Estimation Types:

Let’s quickly review the four most popular estimation techniques in very broad terms.  The four are:

• Analogy
• Bottom-up
• Parametric
• Delphi

Estimation by analogy starts will the selection of a similar project (building a new bathroom based on the results of building a bathroom last week) which acts as a central metaphor for the new project.  The estimator will then decide how closely the two projects resemble each other and whether there are any mitigating circumstances that will affect the effort required to finish the project, how much the project might cost and how long it will take.   Based on these differences the estimator will apply a correct factor and voila, an estimate is created.  The second general category of estimation techniques is the bottom- up estimate.  An estimate of this type typically starts by identifying a set of technical deliverables (a shower stall, sink and pipes for bathroom with a shower if building a house) or work breakdown structure (the tasks needed to build a bathroom for our house).  The identified low level deliverables are estimated based on some form of history and then rolled up to higher and higher levels until the cost or effort for the entire project is known.  The third category estimation is called parametric estimation.  This form of estimation builds and leverages statistical relationships between historical data and one or more variables that define scope such as functional size (the number of square feet in the bathroom times the productivity of the builders).  The fourth type of estimation techniques can be grouped broadly in to the category of Delphi.  The central theme of Delphi techniques is the use of collaborative techniques to leverage group think to decide upon an estimate (the plumbers, electricians and carpenters get together and use a process to come to consensus on how much time is required to build the bathroom).  These techniques work best when the requirements being estimated can be stated a level of granularity that can be understood by those participating in the estimation session.

Mike Cohn has described the planning continuum using an onion analogy[ii].  Where strategy is the outer layer followed by layers for portfolio, product, release and iteration segments as you approach the onion core.  I suggest that there is no one tool or technique perfectly suited for each level in the onion.

Integrating Cohn’s planning onion into our earlier conversation of budgeting, estimation and planning,   I would suggest that strategy and portfolio levels are budgeting tasks.  The product and release layers are estimation tasks where as tools like white parametric estimation make sense.  Iteration and day-to -day organization are planning tasks where planning tools like schedules, kanban boards and standup meeting make sense to direct activity.  The method we are introducing in this paper combines the use of functional metrics and tools in the estimation step in a manner that fosters usage in the planning layers of the onion.  This set of techniques provides a consistent strategy and answers the changing information needs as the project evolves.  All this, while providing a collaborative environment for both the team and the client.

---

[i] http://en.wikipedia.org/wiki/Johari_window, June 14, 2009

[ii] http://www.mountaingoatsoftware.com/presentations/51, June 14 2009

 

Which Size Metric Should I Use? Draft Part 3

The fourth step in our checklist for selecting a size metric is an evaluation of the temporal component. This step focuses your evaluation on answering the question, “Is the metric available when you need it?” When do you need to know how big a project is depends on what you intend to do with the data (that goal thing again). The majority of goals can be viewed as either estimation related (forward view) or measurement related (historical view). Different sizing metrics can be initially applied at different times during a projects life. For example Use Case Points can’t developed until Use Cases are developed, Lines of Code can’t be counted until you are deep into construction or at the very earliest in technical design.

Figure 4

The major dichotomy is driven between estimation needs and measurement needs. As Figure 4 suggests determining size from requirements (or earlier) will require focusing on functional metrics. Functional metrics can be applied earlier in the process (regardless of methodology) because they are based on a higher level of abstraction that is more closely aligned with the business description of the project. Developing estimates or sizing later in the in the development process opens the possibility of more physical metrics which are more closely aligned with how developers view their work.

 

Which Software Size Metric Should I Use? Part 2 (draft)

Knowing the measurement goal that the size metrics you are selecting actually supports is a big deal (extra credit points). The goal has to be a key contributor to the decision process.

While knowing the goal your size metric will support is important it will help you make a cut based on four macro attributes between organization specific and industry defined metrics and between physical and logical metrics. For example, if benchmarking against other firms or industry data is required to attain your measurement goal using organizationally defined metrics would be less viable. Similarly if you have a heterogeneous software environment then selecting a functional metric would make more sense than using a physical metric (logical metrics normalizes varied technology).

Figure 1: Focus

The second checkbox is whether the measure has an externally defined and documented methodology. Why is definition important? Definition is the precursor to repeatability and consistency which allows comparability. Consistency and repeatability are prerequisites for the ability to generate data needed to use the scientific method such as Six Sigma and tools used to support Kiazen. Finally, an external definition reduces the amount of effort that is required to construct and implement measurement programs.

Even where a definition exists a wide of nuances are possible. Examples of the range of definitions begin with the most defined, the functional precision of ISO functional metrics to the less defined methodology of Use Case Points which began with a single academic definition and has evolved into many functional variants. The variants seen in UCP are a reflection of having no central control point to control methods evolution which we will explore later in this model. The range of formality of definition is captured in Figure Two.

Figure Three consolidates the view of formality of definition with the delineation between logical and physical metrics. Each measure has strengths and weaknesses. The first two items in our checklist are macro filters.

Once the macro filter is applied each subsequent step in the checklist will narrow the field.

 

Which Software Size Metric Should I Use?

All jokes aside size matters. Size matters because at least intellectually we all recognize that there is a relationship between size of product and the effort required to build. We might argue over degree of the relationship or whether there were other attributes required to define the relationship but the point is that size at least would be a major part of the discussion. I will cede those points and promise to return to the topic later in the year however I believe we will all agree size is a major component of any IT measurement program. Size is important for estimating project effort, cost and duration. Size also provides us with a platform for topics as varied as scope management (defining scope creep and churn) to benchmarking. In a nutshell size matters both as an input into the planning and controlling development processes and as a denomination to enable comparison between projects.

Size matters but finding the specific measure of software size for your organization is part art and part science. The selection of your size measure must deliver the data need to meet the measurement goal and to fit within the corporate culture (culture includes both people and the methodologies the organization uses). A framework for evaluation would include the following categories:

· Supports measurement goal

· Industry recognized

· Published methodology

· Useable when needed

· Accurate

· Easy enough

The next installment of this essay will discuss evaluating whether the size metrics supports the measurement goals and the topic of industry recognition.

 

Discipline and GNGB Introduction Part 2

Plan Based, Agile or Just Doing ‘Stuff’:

There seems to be confusion in the some corners of the software development world. Those afflicted with ‘the confusion’ have concluded that plan based methods can be replaced with random activity, ‘winging it’. Winging it can not the means to address problems in the real world. Discipline is required to consistently address real problems. Real problems means someone is watching. Discipline requires saying what you will do then being able to execute (whether promising to pick up the kids or finger paint the Mona Lisa). The how is a discussion point whether you use plan based methods or function based methods is a market choice not a discipline choice.

Good Numbers Go Bad – Introduction Part 2

The impact of measures and metrics is dependent on how closely they are linked to business goals and organizational strategy. The closer the linkage between business goals and measures and metrics the higher the probability that value will be derived. Metrics that are specifically tailored to address the organizations business context must not only deliver information as well as add value (information does not equal value). For example, if the business’s goal is to reduce cost, the metrics must measure the reduction of cost of production and the attributes that drive cost. Measuring other areas might be more appealing intellectually (for example quality and productivity) however they will not produce the value that is sought. Identifying and ensuring proper linkage helps ensure the value of measurement thought support of the ultimate business goals

Realistically business goals are typically not as cut and dry as reducing cost. Goals of market leaders (or aggressive newcomers) are typically targeted on towards expanding opportunities (creating disruptive innovation) and/or sales (expansion). Linking metrics to these types of goals can be addressed a two levels. The first is a macro view, measuring output (innovations or sales) and at a micro level, measuring the critical steps that lead to innovation or sales. If you are going to use the micro level strategy, focus only on measuring the most (most critical = small number) critical steps in the process. Measuring too many items can lead to death by measurement, an affliction that can be avoided. The goal of measurement is actionable information not death by numbers even though the later might be tempting.

Measurement programs represent the scaffolding for the analysis and presentation of data. The basic goal of a measurement program is organizing data so it can be interpreted more easily. Frameworks bring structure and organization to an overloaded information worker with allows them to first notice the data then to extract its information content. The quality of the framework will act as a governor on the level and speed of information extraction. The choice then becomes not whether a framework is needed but rather the efficiency of structure and the filters that will be applied. A viable framework of structure and discipline provides an anti-structure. Lack of a frameworks or an anti-structure belief is a will cause a large number of measurement problems, a prescription for making “Good Numbers Go Bad”.