Disciplined Agile

The majority of testing, and in simple situations all of it, is performed by an agile delivery team itself. This is because we strive to have cross-functional “whole teams” that have the capability and accountability to perform the activities of solution delivery from beginning to end.   For organizations new to agile this means that you embed testers on agile teams and for organizations experienced in agile that you’ve managed to motivate agile team members to gain testing skills over time (often via close collaboration with other people who already have those skills).

This blog is organized into the follow topics:

• Parallel independent testing
• Why parallel independent testing
• Questionable reasons to adopt independent testing
• But that isn’t agile!

Parallel Independent Testing

The whole team approach to development where agile teams test to the best of the ability is a great start, but it isn’t sufficient in some scaling situations. In these situations, described below, you need to consider instituting a parallel independent test team that performs some of the more difficult or expensive forms of testing. As you can see in Figure 1, the basic idea is that on a regular basis the development team makes their working build available to the independent test team. This may happen several times an iteration – we’ve seen teams do so at the end of each iteration, on a nightly basis, as part of their continuous deployment (CD) strategy. How often this occurs needs to be negotiated between the two teams.

Figure 1. Independent testing with an agile team.

The goal of this independent testing effort is not to redo the confirmatory testing that is already being done by the development team, but instead to identify the defects that may have fallen through the cracks. The implication is that this independent test team does not need a detailed requirements speculation, although they may need updated release notes including a list of changes since the last time the development team sent them a build. Instead of testing against the specification, the independent testing effort typically focuses on:

1. Investigative/exploratory testing. Confirmatory testing approaches, such as test-driven development (TDD), validate that you’ve implemented the requirements as they’ve been described to you. But what happens when requirements are missed? What happens when requirements are narrowly focused on a point-specific solution, but not the overall ecosystem? User stories are a great way to explore functional requirements but defects surrounding non-functional requirements such as security, usability, and performance have a tendency to be missed by teams new to this approach.
2. Production readiness testing. This is sometimes called pre-production system integration testing. The system that you’re building must “play well” with the other systems currently in production when your system is released. To do this properly you must test against versions of other systems that are currently under development, implying that you need access to updated versions on a regular basis. This is fairly straightforward in small organizations, but if your organization has dozens, if not hundreds of IT teams underway it becomes overwhelming for individual development teams to gain such access. A more efficient approach is to have an independent test team be responsible for such enterprise-level system integration testing.
3. Difficult testing. Many forms of testing require sophisticated skills and sometimes even expensive tooling. Security testing is a classic example.

Why Independent Testing?

There are several reasons why you should consider independent testing:

1. Regulatory compliance. Disciplined agile teams that find themselves in strict regulatory compliance situations, typical in systems engineering and life critical environments, may need to perform independent testing by law. Having said that, only a portion of your testing efforts will need to be performed independently. We have yet to run into a regulation that says that all of your testing needs to be performed independently, although we have seen several organizations that choose to interpret the regulations that way.
2. Production readiness testing is exponentially expensive in multi-team environments. Development teams may not have the resources required to perform effective production readiness testing, resources that from an economic point of view must be shared across multiple teams. For example, if there are 5 other development teams in addition to my own then chances are that each team can do the work required to integrate and test against the builds of the other teams. But what about if there’re ten other teams? Twenty? Fifty? It becomes exponentially expensive for each team to do this integration and testing work as the number of teams increases. The implication is that you will discover that you need an independent test team working in parallel to the development team(s) that addresses these sorts of issues.
3. The average cost of fixing defects is exponentially expensive the longer you wait. For close to four decades Barry Boehm, and other researchers, have gathered data showing that the average cost of addressing a defect rises exponentially the longer it takes to fix. This is depicted in Figure 2. The implication is that we want to find defects as early as we can, in fact ideally we want to build quality in from the start and not inject the defects to begin with. In traditional environments we would have left some forms of testing, in particular system integration testing (SIT) and user acceptance testing (UAT) to the end for the convenience of the people doing the testing (“we need to have everything in place before we can test it”). This results in very expensive and risky defect repair. Parallel independent testing often proves to be a bit more complex for testers, at least at first, but results in much more economical repair efforts.
4. Complex technical environments. When you’re working with multiple technologies, legacy systems, or legacy data sources the testing of your system can become very difficult. System integration testing (SIT), performance testing, load testing, and security testing become more complex and more important in these situations.
5. Large or geographically distributed teams. Large agile teams or geographically distributed agile teams are often subdivided into smaller teams, and when this happens system integration testing of the overall system can become complex enough that a separate team should consider taking it on. In fact, SAFe prescribes this with their System Integration Team strategy (which is virtually identical to this strategy). Granted, the reason why you have such teams is because you’re facing either significant domain or technical complexity.
6. Outsourcing. Teams that are organizationally distributed, for example when some of the work is being outsourced to another organization, will very likely want to perform independent testing to validate the work being performed by the other company(es).  Read this article about agile outsourcing strategies.

Figure 2: Average cost of change curve.

Questionable Reasons to Adopt Independent Testing

We’ve run into a few rather poor excuses to justify independent testing over the years:

1. Your testing/quality staff prefer to work in a traditional manner. They may insist on testing from a detailed requirements specification, testing that is better performed by the team. They may insist on waiting to test the entire solution once it’s ready, instead of incrementally testing the system while it’s being built (enabling much cheaper defect fixing as discussed earlier). They may insist on all testing being done independently, instead of embedding people with testing skills on solution delivery teams. All of these strategies are choices that reflect a traditional culture, not an agile one. The real solution is to overcome these cultural challenges and help them to gain the skills and mindset required to work in an agile manner.
2. Testing is outsourced. Some organizations will choose to outsource their testing to an external organization that is focused on testing.

But That Isn’t Agile!

Bullshit! Please excuse my language. Disciplined agile is pragmatic, going beyond the limited approach promoted by many agile purists.  These purists will claim that you don’t need parallel independent testing, and in simple situations this is clearly true. The good news is that it’s incredibly easy to determine whether or not your independent testing effort is providing value: simply compare the likely impact of the defects/change stories being reported with the cost of doing the independent testing. In short, whole team testing works well for agile in the small, but for more complex systems and in some tactical scaling situations you need to be more sophisticated.

Lean thinking is important for scaling agile in several ways:

1. Lean provides an explanation for why many of the agile practices work.  For example, Agile Modeling’s practices of light weight, initial requirements envisioning followed by iteration modeling and just-in-time (JIT) model storming work because they defer commitment to the last most responsible moment.  These practices also help to eliminate waste because you’re only modeling what needs to be built at the point in time that it needs to be built.
2. Lean offers insight into strategies for improving your software process.  Lean principles such as optimizing the whole and delivering quickly motivate you to look beyond your existing specialized processes to explore how everything fits together and to streamline it.  Identifying and understanding the sources of waste in your IT processes can motivate you to improve the way that you work and thereby eliminate the waste.  The lean principle of building quality in
3. Lean thinking provides a philosophical foundation for scaling agile approaches.  No methodology, process, procedure, or framework is ever complete.  Nor can they be because you can always capture more detail for a wider range of situations.  Because of this incompleteness you need a collection of higher-level principles or philosophies to guide people when their process/procedure/… proves to be incomplete for the situation that they face.  Lean thinking has proven to be a very good source of such philosophies, as do other sources (Steven Covey’s principles come to mind, as does the work of Peter Senge).
4. Lean provides techniques for identifying waste.  Value stream mapping, a technique common within the lean community whereby you model a process and then identify how much time is spent on value-added work versus wait time, helps calculate overall time efficiency of what you’re doing.  Value stream maps are a straightforward way to illuminate your IT processes, providing insight into where significant problems exist.  I’ve created value stream maps with several customers around the world where we analyzed their existing processes which some of their more traditional staff believed worked well only to discover they had efficiency ratings of 20-30%.  You can’t fix problems that you are blind to.

In Implementing Lean Software Development, Mary and Tom Poppendieck show how the seven principles of lean manufacturing can be applied to optimize the whole IT value stream. These principles are:

1. Eliminate waste. Lean thinking advocates regard any activity that does not directly add value to the finished product as waste. The three biggest sources of waste in software development are the addition of unrequired features, project churn and crossing organizational boundaries (particularly between stakeholders and development teams). To reduce waste it is critical that development teams be allowed to self organize and operate in a manner that reflects the work they’re trying to accomplish. Walker Royce argues that the primary benefit of modern iterative/agile techniques is the reduction of scrap and rework late in the lifecycle.
2. Build in quality.  Your process should not allow defects to occur in the first place, but when this isn’t possible you should work in such a way that you do a bit of work, validate it, fix any issues that you find, and then iterate.  Inspecting after the fact, and queuing up defects to be fixed at some time in the future, isn’t as effective.  Agile practices which build quality into your process include test driven development (TDD) and non-solo development practices such as pair programming and modeling with others.
3. Create knowledge.  Planning is useful, but learning is essential. You want to promote strategies, such as iterative development, that help teams discover what stakeholders really want and act on that knowledge. It’s also important for a team to regularly reflect on what they’re doing and then act to improve their approach.
4. Defer commitment.  It’s not necessary to start software development by defining a complete specification, and in fact that appears to be a questionable strategy at best. You can support the business effectively through flexible architectures that are change tolerant and by scheduling irreversible decisions to the last possible moment. Frequently, deferring commitment to the last most responsible moment requires the ability to closely couple end-to-end business scenarios to capabilities developed in multiple applications by multiple projects.
5. Deliver quickly.  It is possible to deliver high-quality systems quickly. By limiting the work of a team to its capacity, which is reflected by the team’s velocity (this is the number of “points” of functionality which a team delivers each iteration), you can establish a reliable and repeatable flow of work. An effective organization doesn’t demand teams do more than they are capable of, but instead asks them to self-organize and determine what they can accomplish. Constraining these teams to delivering potentially shippable solutions on a regular basis motivates them to stay focused on continuously adding value.
6. Respect people. The Poppendiecks also observe that sustainable advantage is gained from engaged, thinking people. The implication is that you need a lean approach to IT governance that focuses on motivating and enabling IT teams—not on controlling them.
7. Optimize the whole.  If you want to be effective at a solution you must look at the bigger picture. You need to understand the high-level business processes that individual projects support—processes that often cross multiple systems. You need to manage programs of interrelated systems so you can deliver a complete product to your stakeholders. Measurements should address how well you’re delivering business value, because that is the sole reason for your IT department.

A common question that customers ask us is how do you measure productivity on agile teams. If you can easily measure productivity you can easily identify what is working for you in given situations, or what is not working for you, and adjust accordingly. One way to do so is to look at acceleration, which is the change in velocity.

A common metric captured by agile teams is their velocity. Velocity is an agile measure of how much work a team can do during a given iteration. Velocity is typically measured using an arbitrary point system that is unique to a given team or program. For example, my team might estimate that a given work item is two points worth of effort whereas your team might think that it’s seven points of effort, the important thing is that it’s consistent. So if there is another work item requiring similar effort, my team should estimate that it’s two points and your team seven points. With a consistent point system in place, each team can accurately estimate the amount of work that they can do in the current iteration by assuming that they can achieve the same amount of work as last iteration (an XP concept called “yesterday’s weather”). So, if my team delivered 27 points of functionality last iteration we would reasonably assume that all things being equal we can do the same this iteration.

It generally isn’t possible to use velocity as a measure of productivity. You can’t compare the velocity of the two teams because they’re measuring in different units. For example, we have two teams, A and B, each of 5 people and each working on a web site and each having two-week long iterations. Team A reports a velocity of 17 points for their current iteration and team B a velocity of 51 points. They’re both comprised of 5 people, therefore team B must be three times (51/17) as productive as team A. No! Team A is reporting in their point system and B in their point system, so you can’t compare them directly. The traditional strategy, one that is also suggested in the Scaled Agile Framework (SAFe), would be to ask the teams to use the same unit of points.  This might be a viable strategy with a small number of teams although with five or more teams it would likely require more effort than it was worth.  Regardless of the number of teams that you have it would minimally require some coordination to normalize the units and perhaps even some training and development and support of velocity calculation guidelines. Sounds like unnecessary bureaucracy that I would prefer to avoid. Worse yet, so-called “consistent” measurements such as FPs are anything but consistent because there’s always some sort of fudge factor involved in the calculation process that will vary by individual estimator.

An easier solution exists. Instead of comparing velocities you instead calculate the acceleration of each team. Acceleration is the change in velocity over time. The exact formula for acceleration is:

(NewVelocity – InitialVelocity)/InitialVelocity

For example, consider the reported velocities of each team shown in the table below. Team A’s velocity is increasing over time whereas team B’s velocity is trending downwards. All things being equal, you can assume that team A’s productivity is increasing whereas B’s is decreasing. At the end of iteration 10, if we wanted to calculate the acceleration since the previous iteration (#9), it would be (23-22)/22= 4.5% for Team A and (40-41)/41 = -2.4% for Team B.  So Team A improved their productivity 4.5% during iteration 10 and Team decreased their productivity 2.4% that iteration.  A better way to calculate acceleration is to look at the difference in velocity between multiple iterations as this will help to smooth out the numbers over time because as you see in the table the velocity will fluctuate naturally over time (something scientists might refer to as noise).  Let’s calculate acceleration over 5 iterations instead of just one, in this case comparing the differences in velocity from iteration 6 to iteration 10.  For Team A the acceleration would be (23-20)/20 = 15% and for Team B (40-45)/45 = -11.1% during the 5 iteration period, or 3% and -3.7% respectively on average per iteration.

Normalizing Acceleration for Team Size

The calculations that we performed above assumed that everything on the two teams remained the same. That assumption is likely a bit naïve.  It could very well be that people joined or left either of those teams, something that would clearly impact the team’s velocity and therefore it’s acceleration.  Let’s work through an example.  We’ve expanded the first table to include the size of the team each iteration.  We’ve also added a column showing the average velocity per person per iteration for each team, calculated by dividing the velocity by the team size for that iteration.  Taking the effect of team size into account, the average acceleration between the last five iterations for Team A is (1.9-1.8)/1.8/5 = 1.1% and for Team B is (5-5)/5/5 = 0.

Similarly, perhaps there was a holiday during one iteration. When there are ten working days per iteration and you lose one or more of them due to holidays it can have a substantial impact on velocity as well.  As a result you may want to take into account the number of working days each iteration in your calculation.  You would effectively calculate average acceleration per person per day in this case.  Frankly I’m not too worried about that issue as it would affect everyone within your organization in pretty much the same way, and it’s easy to understand why there was a “blip” in the data for that iteration.

What Does Acceleration Tell You?

For how you use acceleration in practice, there are three scenarios to consider:

1. Positive acceleration. This is an indication that productivity may be rising on the team, although it does not indicate the cause of that increase.
2. Zero acceleration. This is an indication that the team’s productivity is remaining flat, and that perhaps they should consider doing retrospectives regularly and then act on the results from those retrospectives. Better yet they can “dial up” their process improvement efforts by adopting something along the lines of the Disciplined Agile Framework.
3. Negative acceleration. If the acceleration is negative then productivity on the team is going down, likely an indicator of quality and/or team work problems.

Of course it’s not wise to govern simply by the numbers, so instead of assuming what is going on we would rather go and talk with the people on the two teams. Doing so you might find out that team A has adopted quality-oriented practices such as continuous integration and static code analysis which team B has not, indicating that you might want to help team A share their learnings with other teams.

Monetizing Acceleration

This is fairly straightforward to do. For example, assume your acceleration is 0.7%, that there are five people on the team, your annual burdened cost per person is $150,000 (your senior management staff should be able to tell you what this number is), and that you have two week iterations. So, per iteration the average burdened cost per person must be $150,000/26 = $5,770. Productivity improvement per iteration for this team must be $5,770 * 5 * .007 = $202. If the acceleration stayed constant at 0.7% the overall productivity improvement for the year would be (1.007)^26 (assuming the team works all 52 weeks of the year) which would be 1.198 or 19.8%. This would be a savings of $148,500 (pretty much the equivalent of one new person).

Another approach is calculate the acceleration for the year by comparing the velocity from the beginning of the year to the end of the year (note that you want to avoid comparing iterations near any major holidays). So, if the team velocity the first week of February 2015 was 20 points, the same team’s velocity the first week of February 2016 was 23 points, that’s an acceleration of (23-20)/20 = 15% over a one year period, for a savings of $112,500.

Advantages of the Acceleration Metric

There are several advantages to using acceleration as an indicator of productivity over traditional techniques such as FP counting:

1. It’s easy to calculate. We worked through two common variations earlier, you’ll need to experiment to determine what works for you.
2. It is inexpensive. Acceleration is based on information already being collected by the team, their velocity, so there is no extra work to be done by the team.  Assuming that the team hasn’t decided to take a #NoEstimates approach
3. It is easy to automate. For example, most agile management tools (e.g. VersionOne, Rally, Jira, Microsoft TFS) calculate velocity automatically from their work item list/product backlog and do velocity trend reporting via their team dashboard functionality. This trend reporting is effectively a visual representation of the team’s acceleration (or deceleration as the case may be).
4. You can easily adjust for changing team size.
5. You can easily monetize this metric.
6. It is unitless. The “units” are % change in points per iteration, or % change in points per time period depending on the way that you want to look at it. Because it’s a percentage you can use it as a basis of comparison.
7. You apply this across a department. It is fairly straightforward to roll up the acceleration of project teams into an overall acceleration measure for a larger program or portfolio simply by taking a weighted average based on team size. However, this is only applicable to teams that are in a position to report an accurate acceleration (the agile and iterative teams) and of course are willing to do so.

Potential Disadvantages of Acceleration

Of course, nothing is perfect, and there are a few potential disadvantages:

1. It can be gamed. Acceleration is derived from velocity which in turn is derived from manually-collected measures, and anything gathered manually can be easily gamed.
2. Management must be flexible. For this to be acceptable senior management must be willing to think outside the “traditional metrics box”. Using a non-standard, simple metric to calculate productivity? Preposterous! Directly measuring what you’re truly interested in instead of calculating trends over long periods of time? Doubly preposterous!
3. The terminology sounds scientific. Terms such as velocity and acceleration can motivate some of us to start believing that we understand the “laws of IT physics”, something which we doubt very highly that as an industry we understand.

We hope that you’ve found this blog post valuable.

The Disciplined Agile (DA) toolkit describes strategies for how an organization’s IT group can support a lean enterprise.  An important part of this is to have an effective IT operations strategy, and to do that the people involved need to have what we call a “lean IT operations mindset.”  The philosophies behind such a mindset include:

1. Run a trustworthy IT ecosystem.  At a high level the goal is to “keep the lights on.”  At a detailed level anyone responsible for IT operations wants to run an IT ecosystem that is sufficiently secure, resilient, available, performant, usable, and environmentally friendly.  Part of running a trustworthy ecosystem is monitoring running services so as to identify and hopefully avoid potential problems before they occur.  For some systems, and perhaps for your IT ecosystem as a whole, you may have service level agreements (SLAs) in place with your end users that guarantee a minimum level of trustworthiness.
2. Focus on the strategic (long-term) over the tactical (short-term).  Anyone responsible for IT operations needs to have a very good understanding between the long-term implications of a decision versus the short-term conveniences.  A classic example of this right now is the preference of people building micro-services to use what they believe to be the best technologies for each service.  This makes a lot of sense from the narrow viewpoint of that service and it often proves to be incredibly convenient, and fun, for the developers because they often get to work with new technologies.  However, from an operational point of view you end up with a mishmash of technologies that must be operated and evolved over time, resulting in a potential maintenance nightmare.  Yes, you will still make some short-term decisions but you should do so intelligently.  Too great a focus on the long term results in a stagnant IT ecosystem, too great a focus on short-term decisions results in operations teams who spend all their time fighting fires.  The long-term technical vision for your organization is developed by your Enterprise Architecture efforts and the long-term business vision comes from your Product Management activities.
3. Streamline the overall flow of work.  This arguably should be part of everyone’s mindset, but it is particularly important for people doing IT operations work.  IT operations has traditionally been a bottleneck in many organizations, often the result of the need to run a trustworthy ecosystem and to focus on long-term considerations, hence the need to focus on streamlining the overall flow of work. BUT, this isn’t just operational work that we need to streamline, but the overall flow of work into, within, and out of IT operations.  In this case we need a disciplined approach to DevOps that takes all aspects of the development-operations lifecycle into account, including the support of multiple development lifecycles (not just continuous delivery), the release management process, and the operational aspects of data management.  Of course, streamlining the flow of work goes beyond development-operations and is an important goal of your organization’s continuous improvement strategy.
4. Help end-users succeed.  An important goal of people performing operations activities is to ensure that your end users are successfully using your IT systems.  It doesn’t matter how well your systems are built, or how trustworthy they are, if your end users are unable or unwilling to use them effectively.  End users are going to need help – you need to be prepared to provide a support function.
5. Standardization without stagnation.  The more standardized your IT ecosystem is the easier it will be to run, to release new functionality into, and to find and fix problems if they should arise.  However, too much standardization can lead to stagnation where it becomes very difficult to evolve your ecosystem.  You will need to work very closely with people performing enterprise architecture and product management activities to ensure that you understand the long term vision and are working towards it.
6. Regulate releases into production.   Most DevOps strategies reflect the viewpoint of a single product team.  But what about the viewpoint of your overall IT ecosystem, which may comprise hundreds of products?  An interesting question to ask is what is the WIP limit for releases across your overall ecosystem?  In other words, what rate of change can your infrastructure, and your stakeholder community, bear?  In the Disciplined Agile (DA) toolkit this philosophy is an important driver of the Release Management process blade.  Furthermore, some regulatory compliance regimes call out a separation of concerns pertaining to release management – the people building a product are not allowed to release the product into production, someone else must make that decision and do the work (even if “the work” is merely pressing a button to run a script).
7. Sufficient documentation.  Yes, there will be some documentation maintained about your IT ecosystem.  Hopefully this documentation is concise, accurate, and high-level.  Common documentation includes an overview(s) of your infrastructure, release procedures (even if fully automated, there’s still some overview documentation and training), and high-level views of critical aspects of your infrastructure including security, data architecture, and network architecture.  Organizations that operate in regulated industries will of course need to comply to the documentation requirements of the appropriate regulations.  When infrastructure components are discoverable and self-documenting there is a lesser need for external documentation, but there is still a need.  Any documentation that you do create should be maintained under configuration management (CM) control.

Future blog postings in this series about IT operations and support will explore topics such as why you need IT operations and support, what activities you perform, and the workflow of doing such.