Agile is a broad church. It includes a lot of tools and techniques, some applicable to some teams and environments and others elsewhere. Anyone who thinks hard about how to measure Agility quickly realises it cannot be measured by adoption of practices, it needs to be considered on outputs and abilities.
Agile is sometimes simply defined as ‘not waterfall’. This is a poor, if understandable, definition. Unfortunately, this means that any process that doesn’t strictly follow the classic waterfall methodology can be considered Agile. Adding to the confusion ‘Waterfall’ can cover a number of different approaches, such as stage gate models like DoD 2167 and 2168 and all encompassing methods like SSADM.
In companies where strong, documentation centric, procedures have been hoisted on development teams, Agile is sometimes seen as a ‘get out of jail free’ card. Simply saying ‘this project is Agile’ is seen to exempt work from company procedures. Unfortunately, this card is also used as a cover for cowboy development.
In truth there is a spectrum with strict-waterfall at one end and ‘pure Agile’ at the other – see Figure 1. Since waterfall never really worked that well, very few teams are at the strict waterfall extreme. In his analysis of software development projects over 20 years, Capers Jones suggests that, in general, requirements are only 75% complete when design starts, and design is a little over 50% complete when coding starts [Jones08]. He goes on to say that, as a rule of thumb, each stage overlaps by 25% with the next one.
The spectrum form Strict Waterfall to Pure Agile: almost everyone is somewhere in between.
It would seem reasonable that the pure Agile end of the spectrum is equally sparsely populated. Whether because few teams need to be so extremely Agile, or whether because experience and tools have yet to allow such a degree of Agility, some staged elements exist in many projects.
More than one software development team has encountered the situation when the team want to be more ‘Agile’, the organization and management might even be asking them to be more ‘Agile’, but there are still many ‘requirements’ in a big document and the expectation is that all these will be ‘delivered’. Experience and anecdotal evidence suggest this scenario is faced by many teams.
This mismatch arises when the organization is largely waterfall but the development team are trying to work Agile. I have consulted with companies where senior managers believe Agile is only a delivery process for developers. Business case, requirements, design and even testing is waterfall, just the bit in the middle is Agile.
This article attempts to both understand the different degrees of Agility and provide teams with a way of resolving the requirements-delivery mismatch.
On close inspection Agile has at least three styles: iterative, incremental and evolutionary, shown in Figure 2. These are largely governed by the development team’s relationship with the requirements, and whether the organization wants work defined in advance or prefers goal directed working.
Development team work in short iterations.
Requirements emerge as team incorporates feedback and discovers new opportunities
Testing almost entirely automated
Development team work in short iterations
Minor releases frequent
Formal requirements document start development with change request incorporated
High degree of automated testing
Development team work in short iterations
Major releases infrequent
Formal requirements document (Salami Sliced)
Formal change request process in place
Limited automatic testing
As we shall see in a moment, these three styles occupy different places on the spectrum. But, in truth, there is no clear cut divide between iterative and incremental, incremental and evolutionary, or even iterative and evolutionary. The three styles all overlap and fade into one another.
Iterative development – salami Agile
Working in bite-sized chunks from predetermined requirements with one big delivery at the end.
Iterative Agile refers to the practice of undertaking projects in small, bite-sized chunks. Every two-weeks (or so) an iteration completes and the total amount of work is burnt down on a chart. Customers will probably be shown the latest version of the software at the end of the iteration, although this is little more than a demo. Most likely there will be a single software release at the end of the work – followed by several ‘maintenance’ releases.
At the start of work there is a big requirements document – the work to be done is, at least in theory, defined in advance. Someone, perhaps a previous project, perhaps external consultants, has created a list of the features and functionality the new system must, or should, have. The development team are expected to deliver all of it, or nothing.
The approach here is to see the big requirements document as an uncut sausage of Salami (long and dense). Someone on the team – preferably someone with Business Analysis skills but it could be a developer, project manager, or someone else – needs to slice the requirements into thin pieces of salami (story) for development.
There is no point in slicing the whole salami in one go. That would just turn a big requirements document into a big stack of development stories. The skill lies in determining which bits of the document are ready (ripe) for development, which bits are valuable, and which bits can be delivered independently.
Some slices of salami will be thicker than others but that’s just the nature of the world. Over time, with more skill at slicing salami, it will improve and slices will be thinner.
Working in this fashion opens up the ability to accept change requests relatively easily. But because the work has been set up as a defined project with ‘known’ requirements these opportunities probably aren’t exploited to the full. Similarly, opportunities to remove work will also appear – some slices of salami may be thrown away – but again this will depend on how rigidly the project seeks to stick to the defined work.
As well as the requirements document there are probably some estimates somewhere – maybe even a Gantt chart, which has to be updated to maintain the illusion that it is useful.
However, this is the land where the burn-down chart reigns supreme. There is a nominal amount of work to be done and with each iteration there is a little less. Such empirical measurement is likely to provide a good end-date forecast.
Salami Agile is the basis for incremental development and occurs somewhere about the middle of the spectrum. To go further towards pure Agile, work has to be based less on a shopping list of features and more on the overarching overall objective of the work.
Working in bite-sized chunks from predetermined requirements with regular deliveries and accepting changes
Salami slicing is still prevalent in incremental development, at least during the early stages. Work is completed in bite-sized chunks and periodically delivered to customers to use. These events might, or might not, occur in tandem. While a team might work in two-week iterations, deliveries might only occur every two months.
The pieces of salami are delivered to the customer early, and over time customers start to realize they don’t need some things in the original requirements document, so some slices can be thrown away and some salami left unsliced and unused.
This model capitalizes on the flexibility provided by eating salami rather than steak. Requirements which were not though of can be easily incorporated, others can be changed, enlarged or shrunk.
The iterative approach still assumes the original requirements are correct, so not implementing them all, or changing what is done, is a sign of earlier failure. In incremental development changes are seen positively and reductions in scope are seen as savings – a sign the model is working.
That real live users are getting access to the software early is valuable to the business. It also means user insights and requests are inevitable. Still, there is a major requirements definition somewhere, and while the team can accept change requests easily, it is still expected that one day the team will be done.
Burn-down charts might still be used to track progress, but at times they may appear as burn-up charts as work is discovered.
Tensions arise when the team are instructed to refuse changes, or themselves insist on continuing to salami slice the original requirements document, but users and customer are asking for changes based on their experience. In other words, the users and business have changed their understanding, but the team do not, or are not allowed to, change theirs.
There is no hard and fast line between iterative and incremental, they are just points on the spectrum – with incremental to the right of iterative by virtue of delivering more often. Perhaps the hallmark of incremental is that the team delivers on a regular schedule. When each delivery is a big deal, a special occasion, then things are really just iterative with occasional drops.
Evolutionary Agile – goal directed projects
Working in bite-sized chunks from emerging requirements with regular deliveries
Evolutionary Agile takes this to the next level and is the natural home of goal directed projects. Teams start work with only a vague notion of the requirements. Over time the needs, practices and software evolve. As the software is released to customers the needs are reassessed, new requirements discovered, existing ones removed and new opportunities identified.
The team has a goal, will determine what needs doing (requirements) and do it (implementation) as part of the same project. The team is staffed with a full skill set to do the complete work – analysts, developers, testers and more. The team is judged and measured by progress towards the goal and value delivered, rather than some percentage of originally specified features completed.
Even goal directed Agile needs to start by establishing a few initial requirements. Some teams call this period ‘sprint zero’ in which a few seed stories are captured, from which product development (coding) can start as soon as possible. From there on, requirements analysis and discovery proceed in parallel with creation. Those charged with finding the requirements (Product Owners, Product Manager, Business Analysts or who-ever) work just a little ahead of the developers.
Burn-down, even burn-up, charts have little meaning for goal directed work because the amount of work to be done isn’t known in advance. Work to-do and work done are better tracked with a cumulative flow diagram showing the progress in both discovering needs and meeting needs.
Governing goal directed work is superficially more difficult because it is not measured against some nominal total. Instead work needs to be measured against progress towards the goal.
These projects should be placed under a portfolio management regime that regularly – at least quarterly – reviews the progress and value delivered so far against the goal and the costs incurred. These figures should be produced within the team itself, and the team should feel confident enough to suggest its own end.
Adding these points to the spectrum gives Figure 3. For a team migrating to Agile the objective is to move from left to right. These three approaches might reflect three levels of capability but they may also reflect the nature of Agile in a particular organization. One size does not fit all: some teams are better off with one style of Agile, and some with another.
Many organizations, rightly or wrongly, consider any development process that is iterative in nature to be ‘Agile’. Therefore, in common parlance any method on the right of this spectrum is called Agile, while anything on the left is called Waterfall.
Waterfall approaches might split work into stages, work packages, or sub-projects which can make work look a little like iterative development. Although Waterfall development is associated with Big Bang releases many such projects are released in several Small Bangs. And after release ‘maintenance’ teams would continue to release updates.
Just as few teams actually embrace 100% evolutionary development, few teams ever followed a pure Waterfall approach. Indeed, I would argue that the Waterfall is so fundamentally flawed that a pure Waterfall was always impossible. (Before writing to take me to task please read the original Waterfall paper [Royce70].)
In my experience most development projects lie somewhere between these two extremes, mostly clustered around the centre. Although I don’t have any data to support my argument I suspect that a standard-distribution bell-curve could be laid over this diagram, reflecting that most teams following an interactive process, with a few teams more incremental and a similar number doing periodic releases on a Waterfall basis.
While there are no hard and fast rules about when a team is doing one style of development or another, there are some common traits visible by looking at the practices the teams adopt. These are summarised in Table 1. While these attributes are a useful way of describing and comparing different styles and different teams, they are not prescriptive.
|Planning||Start of project; revisions as needed||Regular 2–4 week iterations||Regular 2–4 week iterations||Regular 2–4 week iterations|
|Status reporting||Regular, against plan||Regular||Regular||Regular, against goal|
|Retrospectives||Sometimes at end of work||Occasional – more talked about than done||Regular||Integral|
|Demo ‘Show and Tell’||Occasional snapshot||Occasional||Regular||Only as information prior to release|
|Budget||Allocated at start||Allocated at start||Mostly upfront||Arrives in increments|
|Budget control||Monitored against plan||Monitored against plan||Value delivered v. cost incurred monitored|
|Releases||Once: at end||Once at the end, or at irregular intervals||Regular during project||Like clockwork|
|Automated unit testing||No||Maybe||Yes||Yes|
|Automated acceptance tests||No||No||Yes||Yes|
|Test first development (TDD)||No||Some||Lots||Everywhere|
|System integration tests||At end of project||During project||During project||Ongoing during project|
|User acceptance testing||Only end of project||At end of project||During project||Ongoing during project|
|Tracking charts||Gantt||Burn-down||Burn-up||Cumulative flow|
|Design||Big up front activity||Mostly upfront||Some up front plus refactoring||Little upfront; mostly emergent with refactoring|
|Goal||Requirements are goal||Requirements are goal||Mix of upfront requirements and goal directed||Governs project and directs progress|
|Requirements||Officially specified in advance||Specified in advance; salami sliced to developers||Specified in advance; salami sliced to developers||Emerge during project|
|User feedback||Minimum||Little||Plenty but little scope to change incorporate||Fundamental to project success|
|Change control||Traditional – changes seen as problems||Traditional||Relaxed traditional||None – changes are requests|
Interestingly, there is one area of software development where the goal-directed evolutionary approach has long been the norm: maintenance. Maintenance teams have the goal of keeping systems working, fixing bugs and, often, small enhancements. Work emerges over time and the highest priority work gets done and other work is left undone.
I remember working on a financial reporting tool called FIRE in 1997. There was no roadmap or even a plan for the product. The company had three, four, then five and even six customers. As each sale was made, new requirements emerged: port from Solaris to Windows, from Sybase to SQL Server, to Oracle, to AIX. And of course bugs.
These requests arrived with more or less noise and urgency. I introduced time-boxed iterations to the team: we released each month, and put a white board on the wall to show what we were doing. Each iteration had a collection of work: we delivered, and then reviewed what had arrived in the last month.
Evolutionary would be the best characterisation of FIRE. Requirements and processes emerged as the work progressed. The overall goal was never clearly stated and we only had elementary unit testing – but we had some!
Conversely, one of my clients in Cornwall is currently writing a completely new version of their flagship product in an iterative way. The feature list is almost entirely taken from the existing product. The team work in one-week iterations. At the end of each iteration their proxy-customer reviews the work and ticks it as done.
The work to do is grouped – physically – into monthly bundles – November, December, January, February. The original aim was of releasing in March but it now looks like it will be April. Nothing will be released until it is all done.
Of course once the first release is done working will change. Probably the team will take more of an incremental approach with monthly updates. They still have plenty of features – new or held over – to continue implementing for a few months. I expect that at some stage new requests and ideas will bring a more evolutionary nature to the work.
This team will need to revisit their overarching goal. As I write the goal is ‘Get a version released with a subset of the current features’. At some time in the near future they will need to question that goal lest they drift into a ‘find work, do work’ mentality.
A change model
It is useful to consider this spectrum as a change model. Assume a starting point somewhere on the left of the spectrum, a team doing some form of common waterfall with all the imperfections that suggests. Being Agile, by any definition means moving to the right.
As a first step the team can adopt an interative approach and use Salami Agile to manage requirements. In time, as they improve, they advance to an incremental approach. To go further the team needs to move away from salami and become goal directed. This requires more of the organization to embrace the Agile ways of the team. Some teams may stall here for this reason.
When a team has a proven track record at incremental delivery, the organization will come to trust the team, then opportunities arise for goal directed, evolutionary work.
Although Waterfall and Agile are often characterised as straight alternatives, neither is particularly well defined. It is better to view them as representing different areas on a continual spectrum from a strict phased approached to a no-phased approach.
On the Agile end of the spectrum there are different ways of approaching work. Many teams work with pre-determined requirements in a salami fashion. They deliver software iteratively or incrementally. A few teams work in a more goal-directed fashion where needs, solutions and processes are evolving.
Different techniques, tools, practices and processes are used at different parts of the spectrum, but there are no hard and fast rules as to what is used when.
Thanks to Paul Grenyer and Ed Sykes for reviewing an early draft of this article; and the Overload editorial team for their usual attention to duty.
Jones, C. 2008. Applied Software Measurement, McGraw Hill.
Royce, W. W. 1970. Managing the development of large software systems: concepts and techniques.
Overload Journal #102 - April 2011 + Process Topics
|Browse in :||
All > Topics > Process (52)
Any of these categories - All of these categories