Friday, December 5, 2008

Kings and Servants - collaborate on the highest priority item

Some time ago I heard a great suggestion on how to view the idea of a team “swarming” on the top-priority item of the sprint backlog – i.e. how the team should see themselves with regard to who works on what story. As you know, following the priority order is a key mechanism in Scrum affecting your ability for timeboxing (having a flexible scope).

The tip was to view the people who work on the topmost priority item in the sprint backlog as “Kings”. Everyone else (consequently working on lower priority items) are Servants. All the Servants help the King(s) immediately as soon as the Kings need anything. Also, everyone wants to be King.

I’m working on spreading this idea in our organization. I had barely finished speaking to one of our teams before I found a printed picture of a crown put up on the Scrum Board alongside the top priority story :-).

Here's an image that I found with Google that you can print and put magnetic tape on the back of and use on your own Scrumboard(s).



Tuesday, December 2, 2008

Done or Complete? How to develop in iterations.

This topic is closely related to that of “Continuous Refactoring” which I am constantly nagging about. It is also related to “Definition of Done”, which is another favorite topic of mine :-).

When are you “Done”? The right answer is: When the Story meets the definition of done. Let’s say your Definition of Done says “Releasable”. So, the Story is done when it is Releasable. Great. But, wait… What does that mean, “Releasable”?

Coders are creative people. I touched this subject in my previous post “1-day remaining forever”. We want to do an excellent job, not just an average one. We want everything we ship to be just right, because bad solutions or sloppy code reflects directly on us and our abilities and creativity. So it’s easy to get stuck in the “almost done” mode, coz we want to polish that extra method or just fix that last design flaw or just rewrite that last class which was poorly written two years ago... So when faced with the goal line (i.e. the point when we’re “Done”) we really want to go to lengths to make sure everything we’ve done is really “ready” and fully thought through, and future-proof, and so on.

But the point I am trying to get to here is that there is a difference between “Done” and “Complete”. Every story needs to be Done in the sense that it fulfills our Definition of Done. But that does not mean “Complete”: it does not mean e.g. “future-proof” or “with architectural perfection”. Those states are achieved over several stories – over several iterations. This is why we call Agile development Iterative. So, think of it like this: “Done” is something we do with each story and each sprint, and “Complete” is something we reach over several stories and several sprints.

I’ve seen it several times; developers dig down into the system and try to solve problems that are of much less priority than the story at hand, e.g. by designing or preparing for some story that is much further down in the product backlog. This behavior builds dependency between stories, it breaks prioritization and it adds a lot of insecurity to the estimates. The proper approach is to focus on the immediate story only, and effectively staying within its boundaries. Trying to stay within the boundaries of a story requires discipline and (not least) a good understanding of Scrum and continuous refactoring.

To sum things up: Aim for “Done”, not “Complete”.

Until next time..!

Tuesday, November 25, 2008

Flexible scope - A magic bullet?

Everyone agrees that you can’t both have a rigid deadline and a rigid scope. We’ve been there, Done that, Got the g***amn t-shirt. Software Development is not mass production and last time I checked I wasn’t able to foretell the future. We want to avoid projects where both the scope and the date are fixed. That’s one of the first lessons every developer learns… And that is one of the key things that Scrum brings: you develop in iterations, you timebox and you ensure you have something to deliver all the time, especially release day. Yay!

Timeboxing means the date should be regarded as fixed. Wow, there’s an easy sell to upper management…! But as a consequence we
have to allow the scope to flex… Simple enough? Well, allowing the scope to flex might not be as easy as it seems. “Flexible scope” as a concept is easy enough to understand, but it’s not something that happens by itself just because you say “we do Scrum”, develop in iterations with beautiful sprint planning sessions and sprint reviews.

I’ll illustrate with a scenario that I’ve seen in reality several times:

Let’s say you’ve started using Scrum, and one of the key selling points to upper management was: “In Scrum we never move the deadline, we promise to try to always have something releasable at the deadline. I can’t promise exactly what, but we won’t ever need to move the date”. Good start!

So, now you start working. The team is working on a release of (let’s say) two major features (each consisting of several smaller “parts”). The Scrum Product Owner and others work on breaking down the scope(s) into User Stories. But they’re not really familiar with Scrum and the stories they produce are poorly formulated, and are done without them really understanding what a User Story is and what to think of.

Still, eventually a product backlog is ready. But, again since the Scrum Product Owner might not be that familiar with Scrum there is not much effort or thought put into the prioritizations of the stories in the backlog. Anyway, the team starts their sprint with a sprint planning meeting and grabs whatever is on top of the backlog.

Once rolling, the team execute the sprint with little or no regard to the priorities. Everyone grabs a story each and start working.

Work continues along those same lines sprint after sprint. And stuff actually do get done. Progress is OK, people feel OK. So far so good?

Now, when you approach the intended release date, you look closer into what’s been done so far and you conclude that there are still several critical things that aren’t ready yet, or that maybe haven’t even been started yet. In addition, you conclude that there are critical tasks remaining for each of those stories which you previously thought were “Done” – tasks such as integration testing, string translation, code reviewing or similar. Ouch.

At this point, what are your options?
  1. The "easy" way out is to postpone the release date, i.e. effectively breaking the timebox rule. …Scrum breaking down. Oh misery…! And you had such an easy sell to upper management with the fixed-date thing... And all you do is move the date and then have the same problems once more, but that time you've sortof used up your goodwill and no longer have the luxury (?) of being able to postpone the date...
  2. Crunch! There's that word again... But you think "Just this time… This release was special...". You make up excuses about why you’re where you are, and call for overtime. Have people come in a little earlier and stay a little later. Have them come in those last few weekends. Not to mention those last days before deadline. So, the team works overtime, they put in extra hours, and not surprisingly quality suffers, people suffer, shortcuts are taken everywhere… Anyone recognize this?
  3. Flex the scope by lifting out the entire feature from the release. Since there are still critical stuff remaining you don’t have the option to just abandon parts. All you can do is lift out the whole feature. That doesn’t give such a good impression of the whole “flexible scope” mechanism, now does it? “Yeah well, in Scrum we don’t move the date, so we have to let the scope change, and that means removing a feature from the release. We have to allow that. That’s Scrum.”. A job well done. No, this is not what “flexible scope” means.
Those are really the only three options. And what’s even worse is that this is probably not discovered until very late. With only days or a few weeks (best case?) remaining until the scheduled release date… This is Scrum failing big-time!

So, the point I want to make here is that we need to be very careful so that we really maximize scope flexibility. The main mechanisms in Scrum which are affecting this are:
  1. Strict prioritization
  2. Ability of adhering to Definition of Done == Releasable.
  3. Skillfulness of breaking down a scope into user stories
To me “strict” means that we both prioritize AND stick to it, by actually working on things in priority order. If we don’t have a prioritization we can’t really say that we have done all in our power to have everything “critical” or “most important” (or whatever value you want to put in the prio) ready by the time the timebox ends. If you’re not careful, the “completed” part of the scope may be any combination of important and less important things, and (what’s worse) the same is true for the stuff that isn’t done yet.

Prioritizing the backlog needs to be intelligently done, and only 1 person should be allowed to have the final decision about a story’s priority: the Scrum Product Owner. I recommend thinking of the importance in terms of “Business Value”. But be careful; it doesn’t only mean value as in dollars and cents (or Swedish Krona and Öre ;-)). The keyword here is Business. Think of it in terms as Value to your Business, including your own internal business of developing stuff. So things that for example are “
architecturally significant” (I’ll post an article separately about this topic, some day) might get a high Business Value because it is important for the architecture and the consecutive development of other stories - even though the story itself may have less of an obvious value to the end user, the customer, your accounting department or your shareholders (but remember: if it is architecturally significant it is most likely valueable to the customer and end user too, it's probably just a matter of thinking about it hard enough and being able of formulating it in words).

The Definition of Done applies to how you “complete” stories. I’ve posted entries here in the past about this and I won’t go into details here. I just want to point out that adhering to the Definition of Done affects your ability of flexing the scope, because you ensure that you stay in control over what’s remaining; you stay on top. The things that have been done are really “done” and are really releasable, there are no hidden piles of work under the rug. I hate surprises and Definition of Done is my best friend :-).

A largely contributing factor both to your ability of adhering to the Definition of Done, and to your ability of following a prioritization, is how good you are at breaking the scope down into User Stories. Don’t regard the scope as a big monolithic lump of functions (e.g. the waterfallish term “Requirements”). Instead, you need the stories to be independent, finishable, valuable and estimatable. A good guide on the way of thinking you want to have is the I.N.V.E.S.T. acronym (stories should be Independent, Negotiable, Valuable, Estimatable, Small and Testable). Writing stories is a skill that you have to work actively with and continuously try and improve.
Don’t be alarmed though; it’s not rocket science. Some breakdown is better than none, and you shouldn’t make such a big deal out of it. Learn by doing. Try different approaches to formulating stories, try the INVEST acronym, try other ways. As long as you understand that the “quality of the breakdown” will affect a lot.

So, to wrap things up: with intelligent scope breakdowns, by following the prioritization and the Definition of Done, we can be more than reasonably sure that the e.g. 85% of the scope which is “ready” when the timebox ends, is really ready, and the remaining 15% is not half-started and we don’t have dependencies between the ready 85% and the remaining 15%. We know we can release
something – and that something is actually something valuable (hopefully even the maximum value).

There. Long post this time. But I like this subject. And it's an important one. I’d be glad to learn about other peoples’ experiences. Cheers! My food is getting cold.... ;-)

Wednesday, November 19, 2008

Have you ever looked at a burndown, really?

The topic today is about burndown charts. I’d like to make some remarks about what it is used for and why we have it.

First of all, the Burndown Chart is a pretty obvious artifact of Scrum. If you use a wall or whiteboard as your “Scrum Board”, you probably have the burndown close-by and you probably (you should) update it daily. As with most artifacts and mechanisms of Scrum, the basics of the burndown chart is pretty easy to understand. So, at a glance, the burndown chart is self-explanatory.

But is it?

Introduction to Burndown Charts
Let’s start out with discussing what the burndown chart is, mathematically/practically; The burndown is the sum of remaining time for all those items in the sprint backlog, spread out over time along the sprint. So along the X-axis you have your timeline, typically measured in calendar days (at least work-days). Along the Y-axis you have remaining time. You can measure that in Days, in Hours, in Gunships or whatever you want. I recommend that you measure it in (people-)days, though. So, on the daily scrum (the standup meeting) the team goes through all the items in the sprint backlog and they figure out the current remaining time (e.g. by summing up all tasks, or whatever level of detail you have on your scrum board). The sum is plotted in the burndown chart. The result is a plotted line that (hopefully) moves from the top left corner, to the lower right corner. By observing this line the team and everyone else can clearly see how well “on track” the team is: will we reach 0 by the time the sprint is ended?

Can the burndown be "ugly"?
Another important remark (that maybe and unfortunately isn't so obvious) is that the burndown chart is intended as a tool for spotting problems early (by displaying progress daily), which means that it has to be reliable: it has to be the real progress, the hard cold truth - not some made-up half-truthful data that makes the chart "look good". Ever thought of a burndown as "ugly"? I know I have. Sorry, but it is not the burnchart that is ugly (don't kill the messenger): it's the reality. It's the context in which the sprint is executed that is "ugly". As soon as you start "dressing up" the burndown you start hiding the truth. And who will end up biting the dirt for it in the end? Yes, the team...

So, a precondition for having any use what so ever of the burndown chart is that it is as truthful as it is current/up-to-date.

A deeper use of Burndown Charts
But why do you do this at all, track progress and plot it in a burndown? Why display it in this way? Or more specifically: what do you (you, the team, the scrum master, everyone) do with this information?

I know for my own part that I’ve observed hundreds of burndowns by now, and so have all of my colleagues. But I’m not sure that everyone has really spent time reflecting over what the burndown charts mean, other than the immediate, most obvious: that we can see if we have more work remaining than what will be reasonable to finish by the end of the sprint. So, that’s why I wanted to bring this topic up: to get a chance to reflect over how to use the burndown in a deeper, more analytical way: more effectively.

Also, I think most people just reflect over the single, immediate burndown infront of them, without specifically observing or thinking about the burndown in the context of other burndowns from past sprints.

The burndown, in combination with the Scrum Board is an excellent and powerful tool that tells a lot about the immediate climate and environment of the team and the ongoing sprint. If you have knowledge about burndowns and velocity of past sprints, then you have even more powerful information. The burndown – and a set of several burndowns from past sprints – is like a flashlight revealing problems. As soon as there is a problem with progress it is shown in the burndown. But equally important: if you have problems with planning, with product backlog maintenance, with outside disturbance or even with your Scrum implementation as a whole, it will show in those burndowns – if you just know what to look for, and if you realize that this is possible.

So, I guess my advice is: take time and reflect not only over the current burndown, but also reflect over the last 2-3-7-whatever sprints and their burndown, at the same time. I suggest you take a photo of the sprint burndown chart at the end of each sprint and archive it together with records of the estimated and the actual velocity. Then bring that out now and then, let the team get their hands on it and spur discussions about how it looks. Discuss patterns and findings.


Scenarios to reflect over
Below are a couple of made-up burndown charts that I made in Visio (yay!) to visualize some key points. I intended them to represent two consecutive sprints each.


Two sprints showing an interesting tendency...

What would it mean if you see a pattern like above? Here, already from the first few days of each sprint, it is possible to see where the burndown is headed. It is however left like that for too long time, but at least the team acts at some points, by (probably) removing stories from the sprint. So that’s good. But the next sprint shows the same tendency. What if this happens a third time, and a forth? It is clear that the team is over-committing every sprint. I.e. they are not following their actual velocity carefully enough. They are ignoring reality (velocity) and convincing themselves that “next sprint it will be different” – but it never is. So, by observing the burndown over a few sprints you can see that (1) the team acts by removing stories (great), but (2) the team keeps over-committing. It is shown in the burndown as well as in the actual compared to estimated velocity.



Two sprints that show flatlines...

What about the pattern above? Well, the team is unable to complete items – for some reason. It is not resolved (or new problems with the same result keeps appearing) over several sprints. So, the obvious issue here is that the impediment perseveres and renders the team unable to finish items. The not-so-obvious problem here is that nobody seems to act to resolve it. Two sprints is a long time. Remember that Scrum says that impediments should (optimally)be removed in a day. So you really have to issues here that you need to resolve: (1) fix the impediment stopping the team from finishing items, and (2) find out if and why nobody is addressing that impediment.

The two examples above are easy ones, but still I’ve seen them both. I think a lot can be saved by observing both the current burndown and really reflecting over it, but also (not least) by now and then observing the burndown of several sprints in a row.

Thursday, October 30, 2008

Allow slack, don't allocate 100%

A classic approach to resource allocation is to aim for 100% allocation. Anything else would be cost-inefficient, right? Having people sitting around doing nothing can’t be a good thing, right? …Well, wrong…!

Aiming at 100% allocation would be desired only if it was possible to know and plan every detail that every person has to do, in advance. And you can’t. Allocating less than 100% doesn’t mean that the unallocated time is expected to be spent on youtube (no offence, youtube is great! ;-)). No, it means that the unallocated time is expected to be available for all unplanned things that has to be done in order for everyone to do their planned work and keep the pace up.

It’s important so let’s repeat it: You should allocate less than 100%. The remainder, let’s call it Slack, is expected to be available to do useful things that can’t be planned for. What are those “useful things”? Well, helping your teammates for example. If you don’t allow for slack the team members won’t have time to help each other. So the team’s productivity will actually decrease because the cooperation effect is lost.

Usually the purpose of trying to allocate to a 100% is to make sure that productivity is maximized. But, in fact increasing allocation towards 100% will risk giving the opposite effect.

Obviously, the trick is to have just enough slack.

So, slack is good. You want slack, and you need to stay weary of any attempts or implied desires to the decrease slack time.

There are several methods you can use to ensure you introduce the slack time as a mechanism in your planning:
  • Continuously measure and use “Focus Factor”. A typical focus factor might be somewhere between 0.6 and 0.8 I would say, but there are no rules and differs for every team. So don’t focus too much on this number: it’s a crude measurement and it has much more to do with how large/many impediments and how much disturbance the team has, how good your user stories are and how good your backlog is, than e.g. the team’s efficiency.
  • Expect no more than e.g. 6 hours of "available work-time" every day. We’ve been experimenting with values as low as 5.5 and up to 6.5 or 7. If you use a project tracking tool it should be possible to set “Working hours” somewhere as a global setting for your project. Either as a percentage of a working day, or as a start- and end-time of each day (which was the option in our case).
  • Measure actual velocity and compare to actual, continuously and let the team help out to give feedback on their sense of “load”
We’ve tried all three, and I think a combination of them is the way to go, depending on situation. Key is to understand why you need slack. If everyone understands that, then you’re able to use either method whenever it feels right.

For another, shorter and more pedagogical way of saying what I say above, I recommend Henrik Kniberg’s slide that you can find here (see page/slide number 26): http://old.crisp.se/henrik.kniberg/presentations/2008-10-13-HiQ/Introduction-To-Lean.pdf

Tuesday, October 28, 2008

Fixed price contracts: Money for nothing, Change for free

Just a tip: If you haven't seen or heard of it already, read this recent blog post by Jeff Sutherland about the concept of "Money for nothing, Change for free" which is his name on the approach of setting up development contracts; Agile Contracts.

I haven't applied this approach myself, but it is really a fascinating topic. I come from the consulting business so I've experienced first-hand the horrible reality of Waterfall classic fixed price contracts. I'd be very interested in trying to apply Agile and Scrum to such a situation...

Thursday, October 23, 2008

Bug-fixing inside inside sprints or outside?

Today’s post is about how to manage bugs.

First some background. We’re a product developing company and we maintain our own products as they go live. A lot of our work focus on new feature releases for the existing products. We have continuously updated lists of live bugs that has to be dealt with.
Even if you’re not a product developing company like we are, I’m sure that you can relate to the situation.

Ok. Let’s take a minute and look at what a “bug” is, and its difference to “a story”. A bug is something that is often unestimateable. Trying to estimate it may take longer than actually fixing it. Or estimating it may even mean fixing it. A “bug report” is a description of an unwanted behavior of the system or, not seldom, a question about a possible unwanted behavior. Anyway, hopefully it also has some type of “priority” relative to other bug reports (and hopefully there’s a single person who prioritizes – or you’ll have a mess). Of course, bugs are unwanted and we need to strive towards continuously improving so that we avoid bugs altogether. The day we have zero bugs I’ll erase this blog post ;-). But until then we need a way to deal with them.

Since recently we use Trac to manage our bug lists. We used to use Hansoft. We moved to Trac because we felt it was more suited for our needs, and it has an excellent and highly configurable web interface. It is also possible to integrate to Subversion, which is the version control system we use.

Now to the options: Do you bring those bugs into the sprint backlog during sprint planning? Or do you leave room in the sprint backlog to be able handle bugs during the sprint, and maybe even set aside time and resources for it? Or do you have “Bug-fixing Fridays”? Or do you have dedicated teams for it? Or maybe you have “Bug-fixing Sprints”? The options are many, and we of course haven’t tried all. I’ll try here to summarize some of the most obvious methods that I can think of though;


A team of dedicated resources work in parallel to the development teams
This has been the approach in one of our products, for almost a year now. We have two dedicated full-time developers who work with bugs. It works great but it requires these resources to be highly multi-talented. These guys don’t work in sprints (at least not with the bug fixing) and they set aside time for the bug fixing activities; on a few specific days of the week they work with bug fixing and release patches to the live product. The other days these guys work with other “urgent” things which we want to shield from the development teams – much of these things have to do with supporting other internal functions such as sales, advertising, marketing, customer support, etc. We call this team “Client Services Team”.

You’ll need to be careful so that these guys are kept happy. If they want to circulate into development you should try to accommodate for it. Working with bugs can be tiring and stressful. Another downside to this is that the developers who originally cause the bugs seldom get to fix or even see the problems so the bugs doesn’t really serve as an incentive for them to continuously improve. On the upside though the release cycles are short and the bug fixing guys often see immediate result of their work. Also, they get to see and learn to work with a huge variety of technologies and parts of the system.


Dedicate an amount of time during each sprint for bug-fixing
This is our current approach in another one of our products, for a couple of months now. We (most often) do 1-week sprints, so the cycles are short anyway. Not sure this approach would work if we had longer sprints. But we let the development team assume the responsibility for dealing with bugs. They do this by means of two mechanisms;
  • Bugs as stories: every week the Scrum Product Owner and the Scrum Masters meet to discuss what’s currently critical or high urgency in Trac, and appropriate issues are taken out and put into the product backlog as stories. They attempt to (guess)estimate them using the StoryPoint value (Gnome Days, see other blog posts below). These stories are merged into the backlog just like any other story, and the prioritization is made relative to any other stories in there.
  • Dedicated time every week for “other” bugs: every week (decided on the sprint planning) the team appoints a person and a specific day of the week which will be used for bug fixing (bugs that are not already in the sprint as stories). On that day, Trac is checked for the current buglist and bugs are fixed from top to bottom. Bugs here are not fixed in order of priority but rather in order of size (smallest first). “Smaller bugs” get done faster, regardless of their priority. The idea is to avoid having bugs in Trac open for ages which are really annoyances but that are not critical. I think maybe this sorting on size instead of priority is a sign of the prioritization mechanism not working properly…
This approach seem to work well too, but it requires that the Scrum Product Owner and Scrum Masters really take time continuously to look at the buglist and update the product backlog properly.


Dedicate a whole sprint every now and then
We haven’t tried this approach. I think there are some problems with it. For one, the team will lose pace and we’ll break the Agile principle of “maintaining a constant pace indefinitely” (see http://www.agilemanifesto.org/principles.html). Secondly, the cycle time from bug report to fix will most likely be longer. This approach doesn’t work well for critical bugs that has to be fixed “right away”, unless the sprints are ultra short. Those critical things would most likely end up leaking in to the development sprints. The upside could be that the developers would get to fix their own bugs. It would also force a wider spread of competence throughout the development team(s) instead of e.g. limiting it to certain individuals.


Bug-Fixing Fridays
We haven’t tried this approach either. It sounds interesting enough though. I'm curious to find out if anyone's tried it. It would mean that you regularly dedicate a certain day(s) for everyone to do bug fixing. Maybe every week, maybe more seldom. The problem with this approach though is that it may be hard to have a feeling of continuity in the bug fixing process. And bugs that take longer than a day to fix will end up open for a long time.

Wednesday, October 22, 2008

On IDG's article about critisism against Scrum

Here's a Swedish article from IDG about that "the critisism against Scrum is growing".

It has something of an eye-catching headline, but the article is otherwise thin. The only really valuable thing in there are the statements from Tobias Fors about having to understand Scrum before trying to adjust it. But the topic as such and the angle of the article is just rediculous. Also, I think Tobias was somewhat misquoted or misunderstand with regard to what is written in the article about retrospectives.

The author quotes Ivar Jacobsson about the teamsize and iteration length limits of Scrum being a downside, as opposed to other methods. The problem with that argumentation is that Scrum is then blamed for not allowing "large teams" or "long iterations". You have to stop and ask yourself why. Why do Scrum limit the team size and the iteration length? Obviously because statistics and years of experience in software development projects has shown that those are factors directly affecting the outcome of a project.

That Scrum doesnt suit "large systems that are based on a service oriented architecture or large organizations" as a general statement is gong too far. If Scrum works or not is in my opinion much more a matter of the organization's willingness and capability to try - not a problem with Scrum itself.

Sure, some of the advantages of Scrum are lost if you are strictly controlled by a ruleset that is waterfall-based, such as public fixed price tenders, or if you work against the FDA or similar. In those cases the problem isn't with Scrum. The problem is with the very preconditions for the projects. In that sense this Ivar Jacobsson really do has a point. But for the author to write an article under the topic "critisism against Scrum" on that is to me a sign of news drought.

Monday, October 20, 2008

Traditional roles vs Scrum roles

Ok, you have an "ordinary" organization with traditional roles, and you want to do Scrum.
You have someone with “Project Manager” or “Producer” written on their business card. Maybe you have someone who’s “Technical Project Manager” too. You have your team, obviously. And last but not least you have some Customer. In our company we do our own products, so we don’t have external customers in that sense. We do have users who are extremely important of course, but I regard them as end-users rather than assigner in the sense of who’s ordering work from us in production. In our case the assigner is the person with "Product Manager" written on the business card – she’s the one making the business calls and the one outmost responsible for the direction and the result of the product. In the world of consulting the customer is obviously the external party who orders and pays for the work done. Anyway, the team does the work. The Project Manager is usually responsible for the high and the low; from high-level planning and risk management down to resource allocation, problem-solving, task assignment & follow-up, etc. The customer makes the order and expects delivery. Below you’ll see a schematic of the traditional roles.


Ok, so we’ve ironed out how things usually look. The above would apply to organizations with external customers (like consulting companies) as well as organizations with internal customers (like us, who own and distribute our own products) – it’s just a question if the Customer is an external or internal role.

Now, one simple approach when introducing Scrum is to put the Scrum Master-hat (SM) on the project manager. He’s the one doing many of those duties anyway, like helping the team move forward, keeping track on who’s doing what, the one controlling the status meetings, etc. Let’s say you do that, so then what roles are left to figure out? Yeah, the Scrum Product Owner. Ok, that sounds simple enough, at least in a company like ours where we already have someone internally who’s “owning the product” and making the business decisions: the Product Manager becomes the “Scrum Product Owner” (SPO hat). Even the name of the Scrum role suggests this construction. Now, in a consulting company, with external customers, I’m not sure it’s as easy as that. I’ll leave that open for now. Anyway, here’s how it would look:



Wow. You now have a team, the person who used to be “project manager” who is now called “Scrum Master” instead, and the person who used to be Product Manager is now called Scrum Product Owner. Sounds alright, right?

Wrong! But what’s wrong with this picture?

A lot of things. One of the most obvious things with this setup is that the Scrum Master isn’t someone within the team. The ever-so-important purpose and work of the Scrum Master is missing or at least the efficiency will be lower because of it. See previous blog post below about why the Scrum Master shouldn’t be a formal leader and needs to be someone within the team. Another factor here, if you have some history in your organization, you’ll need to remember that the project manager has been the one putting the strain on the team, the person who’s been breathing down the team members’ necks. The one with the whip. So it’s a person that the team will most likely never see as “being on their level”; a fact which will make it hard to perform the duties of the Scrum Master to any meaningful extent. So what can you do to change that? Maybe what you want to do is give the Scrum Master hat to someone in the team? Voila, you have your roles. And the project manager is out of a job. Or?



But wait, so far we haven’t discussed the appropriateness of giving the product manager the SPO hat. We just assumed it’s best because the names of the roles are similar :-). I argue that you need to carefully consider whether or not the product manager should really be your PO. In fact, I think that instead the project manager is very suited to take on that role. Because what happens is that the “project manager duties” are smeared out across (A) the team and the Scrum Master, and (B) the SPO. The daily stuff, the things that have to do with who does what, when and why, is handled by the team and the Scrum Master, and the planning, prioritization, risk management and so in is done by the PO. And so, the product manager in this scenario (or in the case of a consulting company – the customer) is perfect for being regarded simply as “Customer”, nothing else.

Now, I know there is some people arguing out there about whether or not “Product Owner proxies” should be used. Here's a link to a Swedish article by Tobias Fors on that subject. As I understand it, the argument from those opposing it is; if the person with the outmost responsibility (e.g. the assigner – the customer) for the product is not deeply involved with the team and instead those duties are taken over by the former project manager – now called the PO – then the PO becomes just a “proxy” for the real product owner. And a key point in Scrum is to involve the stakeholders early, with the team, and get the feedback directly and quickly. Hence, relying on a proxy for it will negatively affect that. So the argument is to forget about proxies and instead let the real product owner be the one with the Scrum Product Owner hat. Well, I have to agree with that. But there’s a reality too, where the product manager is loaded with business decisions, meetings, research, spreadsheets, budgets, administration, etc. In an ideal world I would change all that and remove all those things from the product manager’s back so she could focus on the important parts; what goes in the product – and thereby give the Scrum Product Owner role to her. But we don’t live in an ideal world and I will have to make do with what I can get. And the next best thing is a really engaged project manager (who I call “Scrum Product Owner”), who has the direct link to the product manager (who I call “Customer”).

So, finally, my recommended setup is as follows;


With this setup, some of the advantages are (not in any particular order);
  • The Customer/ProductManager can focus on other things too, and does not have to spend all his/her time on helping the team get the job done. He/she still has to be around on a regular basis and provide feedback early, at least on or around every sprint review.
  • The Scrum Product Owner is a full-time role who's job it is to (a) know the customer and (b) know the team. The challenge here lay (among other things ;-)) in how well this person understands what the customer wants, and how well that understanding is reflected in the prioritization of the product backlog and the breakdown into stories.
  • The Scrum Master is someone within the team, not a formal leader. Also, with this construction the Scrum Master can really focus 100% on being the one sweeping a clean path for the team ("the curling parent"), without any specific further responsibilities outside those of the SM and those of the whole team as a collective.
In an ideal world, if I practically had the choice (e.g. if I were to build up a new organization from scratch and got the power to decide every aspect of it) then I would try and let the "Product Manager/Customer" have the SPO hat. In that sense I do agree with the concerns about "Proxy Product Owners". But if that is not possible - which I don't see that it is in our case, and probably in many other cases - then I chose the pragmatic approach. I think the above is the better than... Well, what's the alternative? To have a Scrum Product Owner or a Scrum Master that for various reasons aren't focusing 100% on their roles.

Monday, October 13, 2008

"1 day remaining" forever

Let’s say the team is working on some story and need to finish a task which is originally estimated to 4 days. Once the team starts working they continuously decrease the estimate until they reach 1. The task then remains on “1 day remaining” for several days – even though the team actually do work (and hard, too) on finishing the task. You ask the team what’s going on, why the task never reach 0, and the reply you get is “Well, things keep popping up that we didn’t think of which has to be done too in order to finish this task. We can’t finish the task unless we also handle those things.”. Let’s say you don’t only see this occasionally. You see it over and over again, Story after Story and Sprint after Sprint.

What is the problem? (Please leave me some comments below if you’ve seen this problem). I have some likely suspects:

Estimation technique
I.e. your estimation process. If you keep having this problem then you probably need to consider if it is because you keep missing something in the estimation. Maybe it is always the same “thing” that keeps holding the task up from being finished? Could you incorporate that “thing” as an item in a checklist which is always considered during estimation meetings? Do you have the right people attending the estimation meetings? Do you have someone with deep enough, and right, knowledge about the system and technical domain to give his/her input to the estimate?

Technical Debt
As a system ages, the shape of its code, architecture, design etc deteriorates – its natural, and every coder knows it. The problem is that the rate of decay is so much higher if the people working on the system continuously take shortcuts. There’s a saying that I keep coming back to; “The dirty remains long after the quick is forgotten”. Think of that “dirt” as taking a loan. Eventually you will have to pay it back, and in the meantime you will have to pay interest on the remainder. “Interest” in our world is in the form of us having to deal with problems when developing new things; things inevitably take longer because the system is so complex. Paying back the loan will mean having to take time to eliminate those shortcuts (that dirt).

If your team tends to stay on “1 day remaining” for ages then you’ll need to consider if the problems they have (the unknown stuff that appears) are due to technical debt. If it is, you’ll want to prioritize very high some efforts for paying back some loans. There’s a business reality out there so, most often is simply not possible to halt development for six months and have all resources work only on refactoring. But at least you can ask yourself; are we increasing the debt still as we build new stuff? And if so, why? Most likely the developers do not want to take shortcuts, but they are more or less forced to because of pressure (implicit or explicit) from stakeholders. You’ll want to look at your team’s environment and see if you can somehow remove the feeling of them not being allowed to take longer time to do things “right”.

A mechanism to emphasize this is to introduce a bullet to your Definition-of-Done: “Not increased technical debt”. It should mean that it is not OK for a task/story to be finished until the implementation is such that it doesn’t increase the technical debt. I.e. it WILL inevitably take longer to finish, but the shape of the code will be better than otherwise. Next step is to replace that bullet with the following: “Decreased technical debt”. That will mean that with every story that is finished there has been some effort put in to refactor (even if ever so slightly) what’s already there.

Formulation of User Stories
If your stories are formulated in a bad way then the team will definitely have a hard time finishing. Maybe the stories are too big? Maybe there’s too much detail and no room for flexibility? If your stories contain every little detail then you’re really back to waterfall again. You’ll spend enormous amounts of time writing specifications, leaving no room for the developers to think for themselves and no way for them to build it in any other way than what’s specified. And hence, it will be naturally hard to develop in iterations. The problem is that the flexibility of the scope has been lost. You essentially end up with a fixed scope.

“Fixating a scope”, to me, is OK. Wow, that’s a controversial statement. Let me try and explain what I mean. As long as it is done on Theme/Epic level (i.e. something along the lines of what’s traditionally called “Function Specification” level) – and not on Story level (i.e. not like a Requirement Specification), then it is OK for it to be considered fixed. If you have someone (e.g. a customer) ordering something (e.g. a feature or a whole system) then it’s OK for that person to really think through what he/she expects. It’s even OK if that is written down.

But as the feature(s) come closer to the development team then it is important that it (however detailed) is broken down into User Stories which are flexible enough to be implemented in iterations, so that you are able to deliver something working at the end of every sprint.

If you’re the Scrum Product Owner, then spend some time thinking about how the stories are formulated. Is the level of detail “just right”? Is there enough flexibility in there for the team to be able to finish it (i.e. without being extremely restricted to an exact certain scope)?

Badly overlapping User Stories
Obviously there’s a million ways to derive (Mike Cohn calls it “trawl”, by the way) User Stories from a vision of a feature or system. I’ve stated previously that I believe the approach should be to divide the feature(s) into verticals (we call them “slices”). Regardless of if you listen to my advice or not :-), you risk running into the problem with overlapping stories. Note that I do not mean “overlapping” as in overlapping code/subsystem: on the contrary, continuous refactoring is all about iterating and reiterating the same code over and over again. What I mean is overlapping functionality; the problem comes when you can’t finish a Story because it intertwines with other Stories, in a way that you can’t tell which story you’re really working on. That is a badly overlapping User Story.

If the team keeps remaining on 1 day, then you should consider if maybe badly overlapping stories is contributing. Try reformulating stories with this in mind. Can you think of other ways to write the stories? Think of the acronym “INVEST” as a rule of thumb; stories should be
  • Independent
  • Negotiable
  • Valuable
  • Estimateable
  • Small
  • Testable
Finished does not mean Complete
This is, I think, as much a result of all of the above as it is a fundamental attitude in the people working on and around the system. When a Story is “Finished” it doesn’t mean that what has been done is in any way “Complete”. I keep coming back to it: Continuous refactoring, iterative development – this is what it is all about.

When unforeseen problems appear and a team stay on “1 day remaining” for ages, can it be that they put the wrong meaning in the word “Finished”?

“Finished” should mean that the story is implemented in some way. It doesn’t mean that the implementation has to be perfect, and it doesn’t mean it needs to be completed and never again returned to. I’ll post a separate blog some day about whether or not this means increasing technical debt or not (in my opinion it does not).

This, I think, requires a change in attitude among the developers, the product owner and the customer/stakeholders. A sprint demo/review doesn’t mean showing Complete stuff. It means showing Stories finished. “Finished” as in “Done-Done”, yes. But that doesn’t mean that we don’t go back immediately next sprint and add, change or remove stuff in order to finish another Story.

Also, another related cause may be that the sprint is too long. The longer the sprint, the more and the heavier the feedback will be from stakeholders. The more seldom the stakeholders see the result of a sprint the more they will feel that they have to “get all the comments out there” and thereby the bigger deal the sprint review will be to the team. And the bigger deal the more it implies “Complete”. So, consider shortening the sprint length (a lot) and get the customer and other stakeholders involved much more often. It is a matter of building confidence between team and customer – confidence that the result in the end is what was expected (or actually even better).

Thursday, October 9, 2008

Beyond the buzzwords

What is “Agile”? And what is “Scrum”? And “XP”? Oh.. And Lean? What is that? Can you “work with” Agile? Or what exactly is it that you do? Are all just different and c00l buzzwords meaning the same thing?

Oh, the confusion… =)

Well, I think of Agile as a “value package”. The word “Agile” is simply the name of a set of core values and principles about how “development projects” is viewed. So, you can’t really “work with Agile”, but you can have processes and project methodologies that are Agile – they fit the Agile core values. Here’s what’s written on Wikipedia about Agile Software Development: “Agile methodologies generally promote: A project management process that encourages frequent inspection and adaptation; a leadership philosophy that encourages team work, self-organization and accountability; a set of engineering best practices that allow for rapid delivery of high-quality software; and a business approach that aligns development with customer needs and company goals.”. There’s a manifesto that covers the core values, which you can find here: http://agilemanifesto.org/.

So, “Scrum” then, what is that? Is that a process? Or a is it also just a “value package”? No, to both questions. Scrum is an agile "process framework". It is not a process, it is a framework for processes to exist (be formed) within. Scrum is an interpretation of Agile Development. Right or wrong, I view Scrum as an “instance” of an Agile development approach. Scrum sets the boundaries and the general direction for the processes but it doesn’t state exactly what those processes should be or what steps and methods they need to hold. That is why it is said that if you “implement scrum” you have to find your own way: you have to create your own set of processes and practices that fits your organization and the Scrum framework. It is quite correct to say that you “do Scrum”: it means that you have identified and enforced a set of practices that cover all the aspects of Scrum. You can read more about Scrum on (lots of places) for example http://www.scrumalliance.com/ and on Wikipedia.

Ok, and “XP” (eXtreme Programming)? Just like Scrum, XP is an “instance” of an Agile development approach. So XP too needs interpretation and adaptation to be implemented in your organization; it is not an out-of-the-box ready set of processes. The difference between Scrum and XP is that Scrum is more focused on project management and team aspects of software development, while XP focuses almost entirely on the hands-on development practices. Scrum for example talks about how to plan your iterations, using Sprint Planning meetings and Sprint Reviews. It doesn’t say much about how to actually write code. XP on the other hand doesn’t say as much about managing your project (although there is some overlap) but focuses much more on how the developers need to sit together and work in pairs, or that you need to have continuous integration, automatic testing, Test Driven Development (TDD), and so on. Scrum and XP fits really nicely together, and together they form a great and whole-covering framework for a really agile development approach. In our organization we have yet to adopt XP to any extent. But we’re getting there.
You can read more about XP on (lots of places) for example Wikipedia.

And finally, what is “Lean” in all of this? Well, it too is a way of observing the world - the world of manufacturing to be precise. According to Wikipedia, Lean is “the practice of a theory of production that considers the expenditure of resources for any means other than the creation of value for the presumed customer to be wasteful, and thus a target for elimination.”. It originates from the automotive industry (e.g. Toyota in the 1950s) and was later adapted to software development (Lean Software Development) by Mary and Tom Poppendieck. Just like Agile it covers a set of core values, and sets a direction for how to go about “manufacturing”, in our case Software Development. Lean is all about spotting and removing sources of “waste”, decreasing cycle time, continuous improvement, respect for and personal growth of individuals, etc. Lean and Agile fits very nicely together.

Confused? Yeah, me too ;-)

Wednesday, October 8, 2008

Slices (Verticals), User Stories and Scrum

We try to divvy our stories in a way that they span from top to bottom; through all layers of the architecture. We call such an isolated part of a function or project, a “Slice”. It is also commonly known as “Vertical”.

But why develop in slices, or verticals, why even care about it, why not just do the User Story regardless and forget about if it is a horizontal or vertical or whatever…? Well, we’ve tried various approaches here. The one thing I am certain of is that this is hard. It requires skill to be able to produce “good” user stories. And not only is it hard, it is also very important: because badly formed User Stories can wreck velocity. Properly shaped and formulated User Stories is one of the basics, a requirement, for a well-functioning sprint execution. So, Hard + Important means that we have to focus on learning it and really spend time thinking about it – and continuously evaluate our approach and view of the matter.

Ok, so then why vertical and not horizontal?
If you do things horizontally, isolated to each architectural layer, then it will be hard for you to actually deliver working, testable code. By doing Slices you are forced to deal with every layer right from day 1. This foundation will then be built further upon, as you iterate the software with every new slice. Also, working vertically – meaning involving every layer in every slice – will often require different competencies (for example UI/web, C++ and SQL/db). Suits nicely with our cross-functional Scrum teams, doesn’t it?

To use Lean terms, it’s about decreasing the “batch size” and ultimately minimizing cycle time; i.e. the time it takes until you deliver something valuable to the customer. If you work horizontally it will take longer until the customer has something that brings value. A complete user interface, but empty of all functionality, is rarely valuable to the customer (and rarely “complete” for that matter, until you’ve built the underlying stuff). And it is definitely not something “releasable”. Remember that you want to strive towards having something releasable every sprint. Sometimes (often?) it doesn’t make sense to release “half a feature” (e.g. “Add User” without “Edit User” functions), but at least you need provide that option and let the customer decide what he wants and doesn’t want: don’t make the decision for him by implementing horizontally.

So what forces are in play here, with regard to stories, architecture and Scrum? Well, let’s recap;
  • We’ve just concluded that we have a requirement to build vertical increments of the system: Slices.
  • We also have a requirement to “finish” (Done-Done) things, with each sprint.
  • We even have a requirement to be able to “Release” at the end of every sprint.
  • And we have a requirement to think iteratively, i.e. we should come back continuously and refactor our code (continuous refactoring – see blog post below)

Tuesday, October 7, 2008

Should the Actual Velocity measurement include “Unplanned stuff”?

A question that keeps popping up here is whether or not the Actual Velocity (i.e. the sum of the Story Point values of all Done-Done stories from a sprint) should include “Unplanned stuff”. I.e. things that were not expected and which are not really part of the Sprint Backlog. The question assumes that the team somehow estimate (in Story Points) that “unplanned stuff” as it appears, and put it into the Sprint Backlog, and consequently track it to Done-Done state and as a consequence include it in the Actual Velocity measurement.

First of all, I think the way to go is to minimize the amount of unplanned stuff. Period. If you have lots of unplanned stuff then you should consider shortening the sprint length, or possibly look at how well the Product Backlog is managed and prioritized.

But there may still be unplanned stuff that appears. There’s no way around it. It’s a reality and we have to be agile enough to be able to handle it.

Let me just emphasize that if you were to estimate the unplanned stuff, you would have to estimate it using the same unit and estimation reference as the regular User Stories; otherwise the Story Point values would not be comparable or possible to sum up. In our case we estimate all our User Stories using “Gnome days” (se blog posts below), so the Story Point value of the unplanned stuff would have to be estimated using that unit too.

So in every sprint we have two alternatives, right: either you do include unplanned stuff (in Actual Velocity) or you don’t. Let’s take an example where we have a team doing a sprint starting out with 30 SP of Estimated Velocity. Let’s also assume that the team so far has had no unplanned stuff appearing in previous sprints and the Estimated Velocity of 30 is pretty accurate and stable.

Case 1: the team decides to include “unplanned stuff” in the Actual Velocity measurement
The team works for a sprint and there’s a bunch of “unplanned stuff” that appears. The team estimates it and includes it in the Actual Velocity measurement at the end of the sprint. The team’s Actual Velocity will probably be close to the Estimated Velocity, even though parts of the work Done-Done is actually “unplanned stuff” that has replaced User Stories that were originally part of the sprint backlog. As a consequence the team will have completed less User Stories than expected, because it had to deal with that unplanned stuff. So the understanding of the team’s velocity at the beginning of the sprint has proven incorrect, in the sense that it incorrectly indicates that the team can perform 30 SP of User Stories from the Product Backlog. That is true only if there’s no new “unplanned stuff”.

Case 2: The team does not include “unplanned stuff”
The team works for a sprint and the unplanned stuff that appears during the sprint is not included in the velocity measurement, so when the sprint ends the team’s Actual Velocity (25 SP for example) will be lower than the Estimated Velocity (30 SP in the example above) – because the team was distracted from the original sprint backlog doing that unplanned stuff. As a consequence, on the next sprint planning the team will observe its past Actual Velocity (25 Story Points in this example) and they will thereby commit to less. This way, they automatically adjust for a “similar amount” of unplanned stuff to appear in future sprints.

Conclusions
Given that you have a relatively steady inflow of unplanned stuff (as opposed to very rare occurrences, or spikes) in your sprints then you do not want to include Unplanned Stuff in your Actual Velocity, because it decreases the quality/reliability of the measurements. As a consequence you risk constantly taking in more User Stories than you’re actually able to complete.

Note that I’m not saying that you shouldn’t estimate or do the “unplanned stuff”. If there’s unplanned stuff that appears that has to be done, then sure you can include it in your scrum board, estimate the remaining time and track it in the burndown. But do not give it a Story Point value and do not include it in the measurement of Actual Velocity.

Monday, October 6, 2008

What metrics should be collected and how to measure and use "Velocity"?

What should be measured during a sprint? Obviously, you can measure a million different things. My suggestion below is tilted towards the simplistic approach. I like to keep measurements and maths to a minimum. Don’t get me wrong; I love Excel, metrics and complicated statistical formulas... But as far as the team goes, and what they need to care about and focus on, I suggest sparing them the nittybitty details and just collect the bare essentials.

  • Estimated Velocity: The sum of the story point value of all those User Stories taken into the sprint backlog (at start of sprint). You can read some more about the concept of "Velocity" here, on the Scrum Alliance site.
  • Actual Velocity: The sum of the story point value of all those User Stories that are determined to be Done-Done at end of sprint.
  • Average Actual Velocity: This should be the Actual Velocity averaged out over a number of sprints (provided the team i relatively stable – or this will have little or no meaning).
  • Mandays in the sprint: This is simply the number of persons in the team multiplied by the number of workdays available in the sprint.
  • Focus Factor: The Actual Velocity divided by the number of Mandays in the sprint.

Example: Let’s say I have a team consisting of 4 full-time individuals and one person working half-time with some other duties. Let’s say they work in 2 week sprints. Let’s further say that the team takes in a total of 35 story points at sprint start. Let’s say they manage to complete (Done-Done) a total of 28 story points. So;
Estimated Velocity = 35
Actual Velocity = 28
Average Actual Velocity = Avg(28) (So far only 1 sprint to average over)
Mandays in sprint = 4.5 persons * 10 days = 45.
Focus Factor = 28 / 45 = ~0.6.

So, what values do you use during Sprint Planning? I suggest the following:
  • Average Actual Velocity, or at least last sprint’s Actual Velocity, to compare the new sprint’s Estimated Velocity to, as a sanity check.
  • Probable Velocity = Mandays available in the new sprint X Focus Factor.
Example (continued from the example above): If the team and the sprint lengths remain unchanged then the team should probably not take in more than approximately 28-30 story points into the new sprint. If the new sprint is, for example, 15 days instead, you can use the previously measured Focus Factor to calculate the probable velocity; (4.5 persons * 15 days = 67.5 mandays) * 0.6 FocusFactor = ~41 story points.

The most important thing is that the team feels comfortable with the commitment on the sprint planning. The sum of the stories selected by the team should however be verified against previous velocity to make sure that the team hasn’t substantially under- or over-committed.

A team’s velocity is a measurement of the reality. It can’t be “too low” because the reality is what it is; making up excuses to why an actual velocity is lower than its estimated is not meaningful. If you see an actual velocity that is lower than your estimated then you will want to lower your estimated velocity next sprint. “But”, you say, “the velocity needs to be higher, it is too low, the project will take too long”. … Yeah, right. But it is not the velocity that needs to be higher, it is the environment in which the team operates which needs to be different. You will only fool yourself (and the stakeholders) if you make up excuses and count on an optimistic, higher velocity than your actual; all you’ll manage to do is to make your plan look good, but it will then – inevitably – be a lousy, incorrect, optimistic and unrealistic plan. The one thing (except resources) that affects velocity is impediments. Remove impediments and the team’s velocity will go up. This should be your approach if you feel the velocity “is too low”.

Friday, October 3, 2008

Scrum Master role vs Leadership vs Project manager

Is the Scrum Master "a leader"? Yes, an informal one - not a formal one. The Scrum Master should not perceive the role as a leader role. The Scrum Master should be someone the team trusts and looks up to - but should be someone within the team, on the same hierarchical level as the rest of the team members.

I'm pointing this out because I have noticed that being appointed "Scrum Master" may imply "becoming the team leader". Which is incorrect. I see the Scrum Master as a representative for the team - the "curling parent". A person sweeping a clean path for the team to go forward. That does not mean leading per se.

If the Scrum Master is "the leader" I think you risk losing some of the cooperation effect and the "tightness" of the team, and with a strong leading Scrum Master the team risks relying too much on that single person to take decisions and lead the way. Or at least it will be easy for the Scrum Master to fall back in old tracks and take charge: thereby undermining the self-organization and collective committment of the team.

So, when you appoint Scrum Masters, make sure it's someone within the team, who the team trusts, knows and looks up to. Most often you'll find there is one or a few informal leaders in the team, and either of those would be perfect for the Scrum Master role.

An important note here is that the Scrum Master should not be the tech lead of the team. Don't make the Expert the Scrum Master. Why? Because the Scrum Master's first and foremost priority will be to remove impediments and be "Scrum Police" (sorry, I mean "Advocate" :)). That means taking time away from other tasks, such as coding. So if you put your tech lead as Scrum Master you will lose pace and you will hinder the rest of the team (because the tech lead will be busy removing impediments and won't have as much time to help others).

Finally, a pitfall which we fell into - which I guess is not totally uncommon for organizations and people who are new to Scrum - is to mistake the Scrum Master role with the classic project manager role. When we started our Scrum implementation we thought that the project manager (called "Producer" in our world) should be the Scrum Master. That was indeed a mistake. Just read the above :-). A lesson learned here is that the project manager role is sortof altered and spread out across Scrum Master and Scrum Product Owner. In our implementation, we regard the Producer as the "Scrum Product Owner" nowdays. I'll write some more about that in future posts.

Thursday, October 2, 2008

Don't document more than that you have to ask to understand

Jeff Sutherland writes in his most recent blog entry about Agile Specifications. Interesting stuff!

I have rule of thumb that I use and try to teach everyone, when it comes to documentation: Don't document more than that you have to ask to understand. It is, I think, something to remember when it comes to writing agile specs, user stories, etc. Spending lots of time writing fancy specs and try to catch every little detail is just a waste of time, because I think it causes a false sense of trust in the spec. You won't succeed just because you discribe every little aspect of some feature. What will let you succeed is if the developers manage to understand it well enough to build it. And communication is key here. So don't write fancy specs just for the sake of it. Communicate instead! Ask! But as Jeff describes in his article above, of course you should document enough. We do write specs too (I'll describe our "scoping process" some day). But the above rule of thumb should always be kept in mind...

Picking stories at sprint planning = Drawing The Line

My thoughts today is about how the team picks stories at the sprint planning meeting.

One point of the sprint planning is for the team to be allowed to pick the amount of work they feel they can cope with, and commit to, for the coming sprint. As a preparation for the sprint planning meeting, the product owner goes over the backlog and revises the story prioritizations based on most recent feedback/input from sprint reviews etc (whatever affects the priorities).

Our approach up until recently was to let the product owner present and make available (e.g. on a whiteboard) the X topmost stories in the backlog. Those stories would be considered "up for grabs"on the sprint planning, but without relative priority. The team could freely pick any of them, in any order. It sounded so good on paper; the team has complete freedom over which stories they pick, yet the product owner has a say about it by means of him prioritizing the backlog prior to the sprint planning meeting.

The pitfall with this is that since the team picks any stories from the top of the product backlog, the stories enter the sprint backlog without priority; or at least the priority has less of a meaning now. And it is therefore tempting to just regard the set of stories in the sprint backlog as one big unordered set of stories - all which must be completed. And as a consequence you lose one of the most valuable concepts of Scrum & Agile: focus on the most important thing first. Plus, it leads to people in the team isolating themselves with one story each, which means the cooperation effect is lost.

Don't do what we did!

So how should you do it? Well, the product owner should prioritize the stories in the backlog, and they should be ready-ready (see previous blog post). On the sprint planning meeting these stories are put in a list and the team can not just pick any of them. Instead, what the team does is to draw a line somewhere. That's it. This drawing of The Line represents their committment. Everything on one side of the line is what they commit to delivering, and the rest is left for consequtive sprints; or whatever the product owner choose to do with them. In our case, we should have the stories printed out on index card papers, and put on a whiteboard from left to right in descending priority order. The team then draws a line using a whiteboard marker. Simple enough! :-)

Wednesday, October 1, 2008

Ready-ready: the Definition of Ready for User Stories going into sprint planning

You all know what “Definition of Done” means. Just for the hell of it I’ll recap it anyway; What it refers to is a state of a chunk of work. The state is also referred to as “Done-Done”. The purpose of the Done-Done mechanism in Scrum is to ensure that those things completed in the sprints really are completed in a commonly acceptable, controlled, premeditated and agreed sense. For example, nothing should be considered “completed” until it is e.g. tested, reviewed, committed and merged – e.g. “Releasable”. Without the definition of done we risk building up a mountain of stuff that we push ahead of us, which eventually (e.g. release-day) will kill us – or at least give us a hell of a fight. What’s even worse, without the definition of done we not only build up such a mountain of stuff, we also have no way of knowing how big that mountain is until we start getting our hands dirty and digging in. And what’s even worse is that many of the things inside that mountain will take so much longer to do later than if performed immediately day 0, because the longer it waits the harder it becomes to fix, to understand, to find, to remember, etc.

So, I think we’re all in agreement that we need a definition of done for the work we ship out of a sprint. Now I think we need a similar mechanism for the work taken in to the sprint. A mechanism which ensures that the stories taken in is really ready to be taken in. Let’s call this state “Ready-Ready”.

First of all, what does “taking in” mean? Well, it means “available and up for discussion on the sprint planning meeting”. So the definition of ready needs to describe a state of stories which they must be in, in order for them to be eligible for discussion on the sprint planning. Before they have reached the Ready-Ready state the stories do not exist and will not be considered for sprint planning. I.e. the top part of the product backlog has to contain only stories which are ready-ready.

Why? To save time of course: to spare the team tedious discussions about stuff that isn’t sufficiently thought-through. To let the team focus their valuable time and energy on things that are as “safe” as possible.

So, the definition of Ready should be;
  • A user story exists which points out the actor, describes the targeted feature and the purpose of it. The story should be formulated like this; As a I want to because . The first two are often easy to pinpoint but the purpose is often overlooked, because it feels obvious. But is it? The idea with writing down the purpose is to give the reader a good feel for the context of the feature – what problem it solves. Not only what it is but also why. This way, the reader can form his own opinion on how to implement it. It gives the reader a deeper understanding for the story.
  • The formulation of the user story is general enough for the team to have the flexibility of delivering it in increments. Don’t describe every detail. Describe roughly what you want but leave the rest for discussion.
  • The story is small enough to fit inside a sprint. Stories larger than that need to be broken down. Breaking down and reformulating stories have to be done before the sprint planning. Obviously, a story cannot be completed in a sprint if it is larger than a sprint. So care should be taken to ensure they are small enough.
  • Each story has its own unique priority in relation to every other story in the product backlog. This is to ensure that the team can work on the most important things first, and to force the product owner to make intelligent decisions about the order of stories.
  • The story has a description of how it can be tested (demoed). The description need to give the reader a good sense of what determines if the story is completed or not. It should be e.g. the acceptance test that is described here. It can be used as a description of how to demo the story too.
  • The story has been estimated (in story points). This should preferably be done by the whole or parts of the team, but at the very least it should be done by individuals with the same understanding of the system and the domain as the team who will eventually be working on it.
So in our case we should have meetings prior to the sprint planning meeting, where the product owner meets the teams and sits down and discuss the product backlog. Don’t forget to time-box the meeting. The goal of the meeting is to deem stories as ready-ready. It will require that some stories are broken down, reformulated, etc – and estimations are done using poker planning. A tester or test leader is required to attend here too, so that the formulation of the “how to demo” is really testable and meaningful.

Tuesday, September 30, 2008

Version control with multiple scrum teams

Earlier this year Henrik Kniberg wrote a very interesting article about version control and scrum, in a multiteam environment. It doesn't cover every aspect, but it is a good source of inspiration and interesting to read anyway. You'll find it here: http://www.infoq.com/articles/agile-version-control

Cross-functional teams (“Feature teams”) in game development is not straightforward

Our teams are average about 5-6 persons in size. Some are somewhat larger and some are smaller. But in general they are all a mix of people with varying competencies. For example for one of our games which is a somewhat “3D intense”, the teams consist of web developers (PHP), 3D developers (C++), 3D artists and 2D artists. The idea with Scrum is that the team collectively commit to the sprint backlog and help each other out to focus on finishing (Done-Done) stuff in sprint backlog priority order.

The problem here is that the 3D artists cannot really contribute to the PHP or C++ coding, and the Web developers aren’t really capable of helping with the 3D programming. So the feeling of collective commitment is weakened, and the cooperation-effect is lost; what remain to help out with are things like testing, perhaps code-reviewing and just generally “wiping the floor” (removing impediments) for whoever is working on the top-most priority stuff (e.g. being “Servants” to the “Kings”). It works sometimes, but it is not optimal. It would be interesting to know how others do it.

The alternative would be to split the teams up according to these “hard borders” of competence, so that e.g. the artists would be in a team on their own, and leaving all the programmers in one team. I don’t have empirical evidence to back this up, but I have a feeling it is a choice between two less-than-optimal alternatives, and putting them all into one team is less bad than to split them up…

Thursday, September 25, 2008

Iterative, Incremental development - Continuous refactoring

It's hard. Certainly much easier to say than to do, right? We keep failing at this. Actually, "failing" is maybe a bit harsh so let me rephrase; we're not as good as I think we should be.

The problem is in how we look at the code we write and the work we do; and what values and expectations we lay in the term "Done" (or "Releaseable"). We want to finish what we start. Even the Definition of Done (DoD) tells us that it should be "finished". Our Definition of Done says "Releaseable" (more or less). This should be your DoD too, by the way. So the team sits there with a bunch of stories to produce and knows that the DoD says that it should be "releasable". A natural reaction, given every developer's pride in work, is to focus on delivering something that is really fully "releasble", meaning "great". Or often "perfect". When all the DoD really (read it again if you don't believe me :-)) says is "releasable".

But here's the problem: "Finished" doesn't mean "Complete".

What we fail to do is to remember that the Definition of Done only requires us to deliver something that is possible to release. No more and no less. It doesnt say anything about the code being in super shape, future proof or completed-once-and-for-all-for-all-eternity-never-to-return-to-again. On the contrary; Scrum and Agile development is all about iterative, incremental work. It is about continous refactoring. It is OK, it is even expected - even required - for us to go back to code and add new stuff and kill off or change things done in previous sprints.

Forget about trying to do everything right from the start. Try and look only as far as the sprint length is - as far as your current sprint goal. That is what we need to focus on. We do what we need to now, in order to make what we're currently committed to - then we come back next sprint and add to and refactor what we previously did. It's ok - it is not something to avoid: it is something to strive for!

Another thing to consider is the following: isn't it great to be able to refactor your code, continuously? I know since my programming days, I really loved it when I got a chance to refactor my own code. This, I think, because I learned stuff when I originally wrote it - so I know how to make it better when I rewrite it. And I think most people love the feeling of making something better.

Finally, I also want to mention something I picked up when speaking to a scrum coach the other day: refactoring a few years ago really could be a pain, due to the poor development environments. But today most major IDE's and code-writing tools have great support for refactoring; in a matter of minutes you can make changes to tens or even hundreds of files. This is something that I will definately look into in our organization; what support for refactoring do we have in our development environment(s), and what better tools are out there?

Monday, September 22, 2008

Shock Therapy!

I'm glad I'm not the only one with this experience... On the topic of what I wrote abot doing Scrum right from the start instead of cutting corners; I learned the other day that Jeff Sutherland (one of the "inventors" of Scrum) calls this approach "Shock Therapy". There's a very interesting article on his blog about this: http://jeffsutherland.com/scrum/2008/09/shock-therapy-bootstrapping.html

Thursday, September 18, 2008

Estimating at Sprint Planning

When do you estimate? And what?

We estimate stories using Gnome Days, which becomes the story's initial "Story Point" value. I know Scrum teaches to estimate the story points as a relative complexity or size, not taking time into account.

So you have a bunch of stories in your backlog, all estimated using Gnome Days. Meaning, I have knowledge about the size of the stories in the backlog, and with my knowledge about previous actual velocity I can derive an estimate of how long those stories will actually take.

So I go to the Sprint Planning meeting with a bunch of stories, the top priority stories in the backlog. The team picks a bunch that they feel comfortable with.

Now, how much do they look at the story point values when they pick what they commit to? Do they even care about it? At this point? Or do they adjust completely to it in relation to knowledge about past velocity? We do something in between. The team picks a set of stories they feel comfortable with but use the story point sum to compare against past velocity, as a sanity check.

Next question: Do you estimate the un-estimated stories at the sprint planning or are you supposed to have that ready when you enter? If so; at what point do you estimate stories? Does the Product Owner estimate the stories himself? Or is it done by the whole team at some other point (other than the sprint planning)? Or do you have an "estimation team" that meet regularly to estimate the most important currently unestimated stories in the backlog?

The latter is my preferred approach, until someone gives me a good argument as to why not. I like the idea of a relatively stable team of mixed competences who estimate stories ahead of time, in priority order. That way, that is done when the team enters the sprint planning, and the team can focus on picking stories and instead revise the estimates, breakdown into tasks where needed and/or sanity-check the commitment. How do other organization do it?

Wednesday, September 17, 2008

Start off by the book, if you're new to Scrum!

We started our agile journey a year ago. I knew that Scrum was increasingly popular and I read a lot about it (and other Agile methods). One of the great things about it was (and is) that it is so "common sense". Every description of the method(s) explain that you have to find your own way and that for example Scrum is just a framework, not a ready out-of-the-box method or toolset. This is great, but it was also (to us at least) a pitfall and contributed largely to one of the first really big mistakes (out of many ;-))...

Why? What happened? Oh, right.. I will explain. The pragmatic attitude of Scrum and the "laidback" sortof informal tone it gives when you read about it, made us feel like it was alright to cut corners; "Oh, yeah well we dont have to do it exactly by the book - Agile and Scrum says you need to find your own way right, an implementation and process set that suits you and your unique environment.. So then it's OK for us to skip the retrospections really, or to not have people appointed Scrum Master' [and so on]". We'd tend to use Scrum and Agile and its flexibility and humbleness, if you will, as an excuse to cut corners in the implementation. As a result, I think our implementation of Scrum/Agile has been suffering and consequently has taken a lot longer time than it otherwise could have. Don't get me wrong; the Agile approach has been meaningful and has improved our means of developing software from day one. I just feel it could have done us even more good even sooner. Many times throughout the past year we've discovered something new about Scrum that we've introduced (or had a problem that we've had to solve) which has really made us take one or several steps closer to a "real" Scrum implementation.

What I am saying is this: If you're new to Scrum and you want to introduce it in your organization or your team; do it by the book first. Don't cut corners or make tradeoffs! Set out to follow everything you read about it, 100%! Take that extra week of studying Agile and Scrum, or that extra $2000 CSM course. Do that, upfront from day one. No seriously. Do it.

I wish we had done that. You will want to learn and fully understand the method first and the thinking behind it, before you are capable of understanding how to make tradeoffs or cut corners. Yes, Scrum is a very flexible method (framework) and you definately need to adjust to your organizaiton in order for it to work fully for you. But you need to (really) understand it first before you're capable of understanding the full implications of shortcuts.

You will find a valuable checklist that you want to benchmark against, here, it's Henrik Kniberg's Scrum Checklist. I wish we had known about this sooner. Check it out.

Tuesday, September 16, 2008

Backlog manager tool

Just wanted to share our "Backlog Manager" excel file, which we use in all our teams.
I'll try to make some posts here with further explanations, information and thoughts surrounding it, but for now I'll just settle on providing the link to the file.

It should work both for Microsoft Excel 2003 as well as Microsoft Excel 2007. It is possible to open in OpenOffice Calc too, but I don't think the Macros work.

NOTE! Buttons doesn't work (nothing happens when you click them?)... Well then you have to enable Macros in the security settings. See the little notice that appears just below the menu/toolbar in Excel when you open it.

Here's the link:
BacklogManager-v8.xls
(Updated 2009-01-23)

Credit to Henrik Kniberg for the original, which I have incorporated into this sheet. The index card generator and the index card template is full credit to him (although I refined his version a little and edited parts of his VB code).

Tool for recording Planning Poker games

When we play Planning Poker I wanted a way of recording the results of each story in a simple way. I also wanted to be able to accept the fact that we won't always reach consensus. So if we have three people playing a 3, person playing a 2 and one person playing a 5, and all of them refuse to change their minds, what do you do? Do you count it as 3? Or 5? Or 2? After you've tried letting the extremes argue (defend) their choices and then had a couple of replays, it starts to feel silly. So being a fan of mathematics and formulas I wanted to be able to live with that difference in opinion and still have a clear way of counting the result. The obvious approach would be to count the average; (2+3+3+3+5)/5 = 3,2 (or rounded to nearest half; 3.0). It works. Do it if you want. I on the other hand wanted the formula to tilt towards "worst case" rather than letting the lower estimates affect it equally. So I used a formula I picked up in (I think) "Agile Estmiating and Planning" by Mike Cohn; Estimate = (1*Min + 2*Avg + 3*Max)/6. The result in the example above would be: (1*2+2*3.2+3*5)/6=3.9. Or rounded to nearest half (which we always do); 4.0.


So for every play we do we record the final estimates from each player. We enter them in the formula above and that is what we the put as the estimate or story point value.


I have a tool I made in Excel, which we now use. Feel free to download and use if you like:
EstimationSheetTemplate-v4.xls

Planning Poker and tasks vs stories

We've used Planning Poker for over a year now. Our experience is that it is a great and easy way of coming up with estimates. If you don't know what it is, you can read more about it here.

When we started it seemed natural to us then that the values of the Planning Poker deck represented hours; 1,2,3,5,8 hours and so on. We had a rule of thumb that said that no task/requirement/story should be larger than e.g. 18-24 hours. If estimators felt like playing a higher card than that then it indicated that the story needed to be split up into smaller bits.

So, this seemed fine, especially on paper. The problem was that many sessions became extremely lengthy and we often got stuck in discussions about details and solutions. I know this is to be avoided in Planning Poker, but that's what I wanted to point out here: we made that mistake. It wasn't as easy as just saying "No let's not talk about design details now"... We said that. Over and over :-). People felt they needed to discuss the details, in order to decide between e.g. a 5 or an 8, or between 8 and 13... It was often possible to force a play and ask people to follow gut-feeling and intuition first, but the discussions kept popping up over and over again.

Our approach on avoiding this and speeding up the sessions was to simplify: forget hours. We estimate days instead. Our smallest estimate now is 0.5 (days) and then we follow the typical Planning Poker deck series; 1,2,3,5,8 and so on. A typical task/story takes between 0.5 and 5 days, I would say.

In my experience estimating in hours causes a false sense of security and leads to over-confidence in the plans. One of the arguments against using days as the estimation unit may be that we then miss the opportunity of really thinking through what parts/activities/tasks a certain story/task consists of and thereby miss possibly important aspects in the estimate... Maybe it is so. I dont know yet... So far I think it's a good tradeoff; I trade waste of time and false sense of security for a possible decreased accuracy... Well, I've already accepted to sacrifice accuracy given the 2-3 minutes per story and the whole idea behind Poker Planning to begin with: a fast and simlpe way of estimating.

My understanding of the common practice in Scrum is that you estimate two sorts of things;
1. Estimate stories in Story Points, which is a relative estimate of complexity (regardless of time)
2. Break stories down into tasks and estimate tasks in hours.

Our implementation (so far :-)) is to simply estimate stories using Gnome Days. Sometimes we have to break stories down into tasks during a sprint planning session, but certainly not always. If we do, we obviously estimate the tasks individually, but we still use days as the unit, not hours.

I'm curious to hear about other people's experience in estimating days vs hours in Scrum.