Agile Intelligence: Schedule Variance

Whenever I sit on a project review and ask the question on quantitative management, this is the first quantitative parameter that I hear: Schedule Variance. Across the projects, IT and non-IT included, this is a metric of concern. Or, may be this is the easiest metric to measure, only to discover that the results are dismal at large.

Riding on the old English anecdote that Time is Money, the importance of schedule variance is indeed paramount. However, when it comes to following the Agile method, it becomes a bit different. May be a bit interesting too. Let me explain why: Agile methods by definition, cannot have any schedule variance. All deliveries are contained within one sprint or other. Sprints do have a fixed end date. As a result, the schedule variance of all sprints are zero.

While the sprints end on time, the deliverables don't. The unfortunate fact is covered up by simply pulling the deliverable off the sprint, and pushing it to the next sprint. So, if the measurement is done sprint-by-sprint, the schedule variance is zero. If the measurement is done output-by-output (or by deliverable), the result is unlikely to be zero. Sad but so real!

At the risk of antagonizing the proponents of Agile methods, the metrication on this area needs a pragmatic approach. Sprints are fine, but their end dates are curved on stone. But the buyer get most benefit when the work products are delivered on time. The product owner may take the responsibility of listing down the macro level deliverables to define such packages. The product owner may also need to do a thorough homework in determining the realistic and beneficial date for each of these packages to become usable. For the consultants working for their clients, it would be useful to have an open meeting with the client to agree on these dates. Once the dates are agreed, measure the schedule variance of each of these deliverables, and aggregate those to the project level.

Many a times, while speaking to the buyer or sponsor about Agile methods, it was found to be drawing a good degree of enthusiasm simply because there was a promise of showing the work products early enough while in the making. The same enthusiasm get converted to frustration when the buyer gets to see a lot things frequently, but none of those make any sense to him/her. A delivery centric view covering only the deliverables that matter most would be good to sustain the enthusiasm, as well as would save some time for both developers and buyer. The same feeling can get echoed through the measurement technique described above for schedule variance. We cannot simply ignore this metric for sake of the definitions within Agile Programming method.

Agile Intelligence: Quantitative Management

The purpose of Quantitative Project Management is to manage projects by a set of defined number driven parameters to achieve the stated project objectives. Apart from having familiarity of various quantitative methods, such activity also demands planning, reporting, and analysis. Unfortunately, Agile methods don't support activities like planning and reporting in a formal way. It doesn't have any role for Project Manager. That leaves the entire space of quantitative management open for interpretation.

So, how should an Agilist address the need of Quantitative Management? The earlier proponents openly discarded the existence of such need, but no longer. As Agile methods are becoming mainstream, such questions are real, demand for demonstrating project performance through numbers are phenomenal. The only quantitative reporting available with Agile methods are burn charts. The chart shows the effort burned over a period of time, and how the budget gets consumed. The chart does not reflect anything with respect to the work product quality. The chart does not reflect anything with respect to the meeting project objectives. The chart does not reflect anything pertaining to the analysis and resolution. As a result, the seasoned Agilist is bound to look into the classical quantitative management techniques, devise his/her own plan and execute it.

This brings back to the same situation similar to my previous post: wearing a project manager's hat while still within the team. Somebody like the product owner should prepare the metrication plan, collect data during scrums/otherwise, report data through a frequency driven reporting, and perform necessary analysis. If the data collection becomes part of the scrum, then the scrum format needs to be tailored. Instead of sticking to the 3 question format, a fourth question may get introduced by asking 'Tell me your numbers'. Or, a standard form can be used to collect data in writing.

The good news is, many large projects today use some process support IT platform, even if the project follows Agile methods. Tools like Team Foundation Server, etc. becomes quite useful in storing and updating the sprint backlogs. The same tool can also provide quite a few meaningful metrics based on how much of the backlog tracking is being done through it. At a minimum, the three basic project performance criteria can be derived quite easily: schedule, cost, quality. There is a minor twist though. The definitions of these parameters within the scope of Agile Programming are different to that of what the classical project management techniques taught us. My next posts will start exploring those areas.

Agile Intelligence: Risk Management

Traditional software engineering has Risk Management as an activity by itself. It is recognized as a process area or knowledge area by the most project management standards. Interestingly, it was never been part of Agile Programming. The reason was easy: Agile Programming was conceptualized and devised by the programmers. Programmers don't like activities like risk management and many others. This worked well to the point where the team was largely formed by two (three max) seasoned developers, and they used to churn out some of the great codes every month. The problem started when large projects started following agile methods. The methods and practices started getting questions that were never thought out, nor addressed. Risk Management was one of those.

So, doesn't a project following Agile methods need any Risk Management? Certainly, it does. The project needs to take help from the established project management standards. There is a problem though: Risk Management is a PM's job. Agile projects don't keep any PM. Who would manage the risks in that case? The question has been so far addressed by the volunteers. The Product Owner or the Scrum Master put on the hat of the PM to manage the risks. That demanded the risk manager to come out of the project chores time to time and take a thirty-thousand feet view to identify and assess the risk.

The broad level steps are same for risk management in Agile: Risk Identification, Risk Assessment, Risk Planning, and Risk Monitoring. These steps are implemented through scrums, the Scrum Master facilitates the session. However, these scrums don't follow the traditional 3 question format. These are more akin to the brain storming interactive sessions to ensure that all risks are identified and their impacts are rightly analyzed and understood. These sessions also demand the Scrum Master and Product Owner to provide a good perspective on the project risks. Being part of the team and weathering the daily chores, the risk of missing the forest for trees are very high in such a situation. Many project teams try to address this by bringing in an external expert / observer into these sessions and take their inputs. Such attempts are undoubtedly good but deviates from the Agile principles to a good extent.

To summarize, IT projects following Agile methods do need risk management. To accomplish this, the project team must leverage one of the many established formal project management standards and guidelines. The project team must assimilate the role of the PM within one of them to ensure the risks are formally managed. External support, if available would do good to the entire project process of risk management.

Agile Intelligence: Good To Great!

This is no plagiarism. My intention was not to pickup some perfect lines from here and there, but to rationalize a thought process. A thinking that keeps me asking to myself: Are organizations becoming great by adopting Agile Programming?

There is no denying that in an environment of continuous change demand for more agility from the business support functions are at historic high. At the same time, dependence on IT to run business and deliver goods & services are increasing like never before. Combining these two, the flexibility of the IT systems and processes to respond to ever-changing demand becomes the contributor to the business success. Certainly, Agile Programming can help in removing the inflexibility of the IT systems and processes by adopting some of its proven practices.

However, the agility may bring some compromises as well. While Agile methods don't explicitly denounce any structured approach and the need for documentation, there is a subtle ignorance towards this. Programmers love Agile Programming because quite often it relieves them from doing the mundane documentation, or following some lengthy processes. This is a dangerous trap. Sooner, the organization would either become captive to those rogue programmers due to their know-it-all attitude, or it would succumb to the fact that the IT systems are no longer maintainable. Yours truly had a recent experience of the latter situation, where a great software application became miserable and painful just because it was no longer maintainable, and nobody knew the inbuilt functionality well enough to develop a new application. So, Agile Programming is not necessarily a path towards greatness.

Many companies have adopted the process standards like CMMI, and Six Sigma for their IT organization to ensure proper adherence to the processes and fostering an environment to improve. CMMI, if effectively implemented, can be helpful in having a great structure in place for IT work. Six Sigma can help in removing all the wasteful processes and steps thus making the CMMI structure quite efficient. None of these two have any conflict with Agile programming, and can very well embrace some of the proven best practices like pair programming and scrums. Such a combination may bring some amount of greatness within the IT organization.

Unfortunately, there are not enough takers of such a theory, let alone practice it. Of late, the proponents of Agile have started denouncing that there are conflicts between Agile and CMMI. They are also openly encouraging the need for documentation within an Agile process. All these are good sign. It also shows a desperation to become mainstream practice for software development. But to make an IT organization great, only Agile may not be enough.

Agile Intelligence: Peer Review

There is no peer review activity in Agile Programming. It's embedded into it. When the proponents of the Agile methods started practising it in a bigger way for serious IT projects, there was serious need to doing reviews. Traditionally, the programmers following Agile were all experienced hands and hence there was a little need for doing such reviews. As the projects following Agile started becoming large and complex, two problems started bugging the Scrum Masters and Product Owners a lot:

1. There is not enough experienced hands to code any more, since the team size is bigger.
2. Complex codes require comprehensive look through the second pair of eyes.

Pair programming principles came very handy this time. In this setup of work, no single developer is assigned a programming task. It's always assigned to a group of two programmers. The technique instructs this pair of programmers to sit in front of a single workstation (computer) and perform the programming corresponding to the assigned task. One programmer is supposed to do the typing of the codes: to be called as the driver. The other programmer is supposed to continue reading the codes being typed and review the correctness of the codes: to be called as the observer. Some books also term the latter person as the navigator. As a result of this setup, the codes are getting reviewed on the point of production.

This technique has been adopted in many IT projects and being implemented successfully, the simple reason being its effectiveness. A traditional software development method mandates review of the program after production, but is mute on the technique. Most often, a developer writes a piece of code, compiles and re-compiles it till it becomes error free (not bug free), and then leaves the program unit for review for an elderly programmer. By the time the reviewer steps in, the program unit may reach to a size of many thousand lines, with the implementation of several requirements. The reviewer on hot seat now has a daunting task of reading through the requirements, and then reading through the programs to make some review comments. There are a many good ways to make such reviews effective, but at large the task is no more a fun affair. This is where pair programming technique clicks. It makes the programming a fun. The programmers interacts continuously between themselves about the codes and implementation and through instantaneous iterations, make the codes better. No more pain and effort for documenting review comments, no more closing the review comments through a loop of comprehensive work flow. Sample this: How many of the traditional code review comments don't start with a statement on the code indentation or comments? The answer is probably "not many". While it's very important to produce properly organized readable codes while programming, quite often the reviewer gets lost within such show-off stuff instead of getting deeper into the question whether any alternative coding would make the implementation better. Pair programming technique allows such review comments to be made right on spot by the observer so that the driver can make the necessary changes right there.

Criticism is not few though. One big question is always put up is about productivity. Deploying two programmers to program on a single workstation perceivably reduces the productivity to half. However, the equation is not that linear. The proponents of pair programming submits two points in this respect:

1. Zero Review Effort: Review being embedded into the technique, no separate review activity needs to be planned and managed. This would save some good degree of time and effort. Each organization or project team maintains its own review effort benchmark, and it varies widely. So quantification may be difficult, but everybody agrees that it does not fully substitute a second programmer's time. Here comes the second point, a bit upright and direct.
2. Productive Hours: Some of the proponents of this technique openly challenged to walk by the cubicles of the programmers who are nested within their own self for programming. Their revelation was that about half of the population were doing something but coding: checking emails, preparing reports, web browsing, and even blogging like yours truly. According to them, this time would not go waste if two of them staring at the workstation for programming. In a pair programming process, the pair typically gets exhausted after 6 hours of work, simply because the intensity of the focus they need to put into work. On the contrary, according to the proponents, the single programmer programming is not worth more than 4 productive hours sans review.
   I admit that this is a questionable argument. Not all organizations foster similar culture, and hence the outcome would largely vary. However, given the quality of the output through pair programming, it may be useful at many times. If not for the entire set of programming tasks, but certainly for the complex and critical sets.

Pair programming also opens up an avenue that the proponents largely ignored: possibility to deploy junior resources. Agile methods don't have enough room for junior programmers, pair programming allows the formation of senior-junior tag team to perform programming and learning to do better coding on the way.