Wednesday, February 10, 2010

Chaos and SCRUM

Josh Nankivel at asked me to write about some of the challenges and obstacles that we face in our environment. In order to do that I should explain a little about that environment. Over the past 36 months we have transformed ourselves from being a software development team in crisis to a strong team that is focused on delivering on it's commitments. In 2005 we determined that the current approach to writing software wasn't working. We were the poster child for "garage-band" style of software development, having inherited 1 million+ lines of classic (poorly written) ASP code [aka 1MLOC - I'll let you guess what the "C" stands for], and having very little recognition from corporate management that anything was wrong with the way things had been done. When I came to lead the group in early 2007, I had put in countless hours trying to lead the business toward implementing a PMO. While we actively maintain the 1MLOC, we have been working closely with a third party partner to replace the application. In parallel, we have a number of other small LOB applications that we have written using accepted best practices and design patterns. Needless to say, this means that there is a significant amount of churn and chaos when it comes to determining what's going to be worked on and by whom. Enter SCRUM.

In early 2008 I approached the IT leadership and explained that an iterative and incremental approach to developing software was far superior, in our environment, to the code-and-fix triage approach to project management that had been promoted in the past and that this would be well suited to supporting the behemoth 1MLOC application while still allowing us to be agile enough to respond to the business' needs in a timely manner. They bit.

So, what is our environment like, how do we handle some of the obvious gotchas, and what are the advantages/disadvantages to how we do business? Well first, I'll say that this is what works for us, in our environment. I can't speak to your situation, although if I can help, please drop me a line. Well, we currently have over a dozen product backlogs that are somewhere in the development/grooming workflow. We have roughly half that many Product Owners [none have been "officially trained", all have gone or are pending admission to the school of hard knocks] and the expectations of what it means to be a Product Owner are communicated loudly and often. Some gravitate toward the role, others are a little thicker and require some additional coercion encouragement. Regardless, we do the classic planning poker estimating that I described here and then, based upon our standing team velocity we let the Product Owner how much of the backlog we will be able to knock out during the next sprint. At that point, for any one of a couple reasons [the Product Owner needs more functionality than can be delivered in a single sprint/the team has some groundwork to lay in one sprint in order to deliver the key business deliverable in a second or third sprint/etc.], the team may decide that they need to deliver the next increment of work over several concurrent sprints. In this case, the backlog of backlogs is queried and the next product owner is queried for budget and schedule requirements [we may even look at the current state of that PO's product backlog] and a reasonable determination is made [by my boss and the IT Steering Committee] whether it's acceptable to the business for the subsequent development to 1) be delayed 2) be preempted or 3) be split [deliver Sprint 1-Product A, then Sprint 1-Product B, then Sprint 2-Product B]...from the team's point of view, this decision is somewhat irrelevant as long as the sprint backlog is ready to be loaded into the SCRUM Dashboard [which is integrated into the Microsoft Team Foundation Server - our source control and overall ALM solution]. Once the tasks are entered into the Scrum Dashboard (which is a lot like SCRUMY except that it's tied to coding tasks, time tracking, check-ins, build events, etc.) the team tracks their own progress daily which automatically updates the Sprint Burndown, Product Burndown, BurnUp, and a hundred other reports that are hosted in the Microsoft SQL Server Reporting Services site that comes out of the box. The daily stand-up meetings are usually run by the team, as the ScrumMaster I sit in on them regularly; however, my prime directive is really just to make sure that they're happening and that there are no discovered impediments that are/have not been communicated to me that come up as an outcome of the meetings. We have a weekly code review meeting for the whole team to see where we are, and we usually use this meeting to demonstrate the application state to the product owner. As part of our development team, we expect for the business to provide testing and QA resources so that we can be sure that the feedback loops are as short as possible.

So, you ask, what are the gotchas in this environment? That's an easy list to start, but a very difficult one to call here's my "easy list"...the hard ones will quickly become evident in your environment.

1) Training Product Owners - especially in a "Command and Control" environment, it's difficult to convey the concept and importance of servant leadership. Similarly, it's hard to get "busy" business folks to commit their time to things that are commonly deemed to be unproductive

2) Training Product Owners- getting them to understand the concepts of story points, estimating, life without requirements documentation and use cases, life without earned value, and ultimately the fact that the ROI and the "fit for purpose" functionality is their #1 responsibility
Death by meeting - herding all of those backlogs, training all of those Product Owners, daily stand-ups, retrospectives, code reviews, "demo days", Planning Meetings, not to mention the other "normal" meetings that one usually gets pulled into [weekly staff meetings, employee meetings, annual reviews, etc, etc, etc]...leads to about 80% of your time being tied up in one meeting or really is hard to do a lot more than manage your schedule and check your business card to make sure it matches your driver's license
3) Leadership & Management - if your management doesn't back you up and get out of your way, you will be tripping over one another. They MUST support you and then get out of your way. The proof is in the pudding...deliver a few times and they'll get it.
4) Tools & Co-location - We are one team, in one location. We tried the distributed team thing for a while and, to be honest, the requirements for the team to be available to meet to review something on very short notice makes co-location extremely difficult. Likewise, much like there isn't really a "SCRUMBOK" because SCRUM is prescriptive as far as Engineering Practices, there are hundreds [if not thousands] of electronic tools for the team to use...a definite gotcha is the temptation to jump into the midst of the tools arena before you really understand the process that needs to be supported by a tool. We found TFS and SCRUM's what works for us....that means squat to you....make your process work analog, and then digitize it....I can't say this any more clearly...learn to love the whiteboard, post-it, and index cards...worry about the business now and worry about the rest later [it might actually take care of itself].

OK. So there's a start to a list...far from comprehensive, or even thorough. That list could go on for a while. So, to wrap up, you're probably interested in understanding what are some of the pros and cons?

PRO: We deliver working software, and we do it pretty frequently. This means that our Product Owners are pretty happy. Happy PO = Happy Business = Job Stability = Happy Employees = Better Software

PRO: We get to deliver at a sustainable pace and we LOVE what we do. I hope that the death marches of 2006 are gone for good. We value people over process. Happy Employees = Better Software = More Projects = Job Stability = Happy Employees
PRO: We get to be honest and transparent. With the PO on the team, there are no secrets about something taking longer than planned, or something not working as expected. The PO and the team get to figure out how to work thorough these challenges rather than worrying about doing damage control.
CON: Estimating is really difficult, especially at first. Then, to make matters worse, budgeting is even more vague. For some organizations, this may be too much to overcome. I hate budgeting ask me how much something is going to cost is to ask me for a complete and estimated product backlog, which isn't terribly different than a preliminary project plan and schedule. Ugh. Budgeting Sucks.
CON: If you are a "Command and Control", "Do first, ask questions later", "take the bull by the horns" kind of manager, you will really struggle with Agile. The CORE of Agile is in empowering the team, transferring that control to them, and settling into a support role....for a lot of folks, this will be impossible.

Saturday, February 6, 2010

Implementing SCRUM...the basics - part 2

Once again, I published this on but thought it would be appropriate to post it here as well.

In this post, I explained that SCRUM is not a silver bullet and that there some significant barriers to entry for organizations wanting to adopt SCRUM. In this post, I'm going to introduce Agile estimating, planning poker, fibonacci, and velocity.

Let's say that your organization is ready to go, you've turned command and control on it's ear, you've identified the Product Owner and you're going to be the ScrumMaster. You've put together a team of seasoned developers, a creative designer, and a QA Lead, but the designer and QA Lead have never worked on an Agile project and they also have never worked with this team of developers.

Let's say that the Product Owner has put together a pretty comprehensive list of User Stories [let's say that for argument sake that these have been validated and prioritized and you have verified that they follow the INVEST principles...more on that in another blog post]. What now?
First, you need to work with the "Team" to make sure that they understand what their responsibilities are, this includes the talk about being a self-organizing team, that they're responsible to one another, the Product Owner, and ultimately to the organization to do what they say they're going to do.

The next step is for the SCRUM Team [that's the development team, the Product Owner, and the ScrumMaster] to decide which items in the Product Backlog you're going to deliver in the upcoming Sprint. In order to accomplish this, you need to be able to quantify how much time or effort each item on the Product Backlog will require. This quantification of the Product Backlog is called estimation and is commonly accomplished by having the development team play a game called Planning Poker.

Game playing? Man this SCRUM stuff is wierd. OK, let's talk about Planning Poker.

The idea behind User Stories is to collect the requirements in bite-sized pieces. The size of those bites will inevitably vary. The Agile solution to this problem is to rate each User Story according to relative "size". The best analogy I've heard is to compare the stories to dog breed...compared to the other stories, if the current User Story a Toy Chihuahua or a Great Dane? Where in the middle...Jack Russell Terrier? Labrador Retriever? Once each person has an idea for the relative "size" of the item, each person has to play their hand, and the group consensus wins...highest and lowest have to explain why they think the feature is larger or smaller than the rest of the team. When I do this, as the ScrumMaster, I intentionally don't play a hand because I don't think I should influence the team's collective insight into how big or small a feature is. To each their own. The planning poker deck we use was purchased from Mountain Goat Software and was well worth the few bucks it cost. There are online versions and of course with a pair of scissors, some index cards, some markers, and a little free time you can make your own. The important part is that you understand the notion of the relative scale of accuracy and the variability that is introduced as a feature gets larger. You're probably familiar with the Fibonacci series of numbers. We use this series of numbers to incrementally scale the "size" of a feature. Why do this at all you ask? Well, if I ask you to estimate in inches, the distance from the tip of your index finger, where it currently is, to the tip of your nose, you will probably be able to estimate with better than 85% accuracy. Now if I ask you to estimate using the same criteria the distance from the tip of your index finger to the middle of the letter "O" on the nearest Stop Sign [depending where you are right now] you might be able to estimate with a significantly smaller degree of accuracy, and finally if I ask you to estimate, in inches, the number of inches from the tip of your index finger to the tip of the Empire State Building in New York City, you are likely to have an accuracy approaching 0%. With this in mind, we use the Fibonacci sequence because as a feature increases in size, our inherent ability to accurately estimate it's size diminishes. After all of the items on the backlog are estimated, the next step is to determine the Sprint Backlog...the estimated subset of product features that will be addressed in the Sprint. We had a hard time understanding how much we could bite off in our first few sprints [in other works, how many "story points"...or Fibonacci values we could consistently complete in a sprint].
In order to get around this, we took two decisive steps. First we committed to micro (one week) sprints and second we intentionally rigged the sprint backlog with items that we knew were small enough to complete in a week, yet big enough to validate [one way or another] our estimates. This process allows us to determine our "velocity". Velocity is the average rate at which a team can complete story points for a unit of time. Once we were able to determine out weekly velocity, we were able to extrapolate our bi-weekly velocity. Once we were able to determine the velocity for a "normal" sprint, we have become increasingly confident in our ability to deliver on that velocity. I will put in a big caveat...the velocity statistic is only a general guide, and as you adopt SCRUM, the accuracy of the velocity is only as good as your ability to limit the variability of external change within the team. In other words, if you determine the velocity for a specific team, that velocity will be affected by making changes to the team, and even in some cases by changes that the team may make from within.

For more information on these topics, there is a great book on this topic written by Mike Cohn called "Agile Estimating & Planning". I highly recommend it.