Globally distributed scrum

I'm on record as saying that globally distributed scrum teams can work, but sub-optimally. I'd like to retract and revise that position, so that something I'd said doesn't inadvertently lead a team down a poor path.

My team's scrum adventure lasted from March 2007 to April 2008. (Why does it have an end date? Buy me a beer, and we'll talk. But none of the people who were a part of or were affected by our efforts wanted it to end. The project was a success.) We started out with four developers in Austin, six developers in Bangalore, a Test Lead in Austin, and five testers in Hyderabad. I'll call this the Two Parallel Teams period. Team members were moved over time, so that we became two developers in Austin, eight developers in Bangalore, a Test Lead in Austin, and three testers in Hyderabad. This is the One Distributed Team period.

We had a good collection of team members, and I think we could have worked together very effectively, had the distance between our desks not been so great. I am not criticizing off-shoring; I am criticizing scattering your product owners, developers, and testers across different continents. Note who I included in that list there; if your dev team is co-located, but your product owner is far away, you have a distributed team.

We did not have a choice about where our teammates should be located, but we had influence over which project methodology we would use. So the choice was not "distributed scrum versus co-located scrum," but instead "distributed scrum versus distributed waterfall."

Do I still think distributed scrum is better than distributed waterfall? Yes, although with less unbridled enthusiasm. It required more meetings from all team members (including product owners) at crazy hours, but it still made it easier for the product owners to get the system they need; it still had us writing more code, more often, which is way more fun than fighting over contracts; and it still enabled productive, visible progress toward our goals.

We were happiest in our Two Parallel Teams mode. The developers in one city worked on one feature, and likewise for the other city. The only people who really suffered here were the product owners and the scrum master, who had to accommodate a ten-and-a-half-hour time difference to have their sprint planning meetings.

We offset our two-week sprints by one week, so that the testers could be shared between the two sets of developers. It worked okay because it did not require the "save state and hand off" process inherent in distributed teams. That process is, at the end of the day, you write an email explaining what you did, where your thoughts are about the work-in-progress, what you want your teammates to pick up, and what you want them to leave alone because it's currently too fragile to explain or share. It's like "hibernating" in Windows--save a snapshot of where you're at--and it takes time.

The more distributed the team, and the more they need to collaborate, then the more time will be spent handing off, and the team will be less efficient. In our One Distributed Team mode, the hand-off process overwhelmed the development process, so that more time was spent saying what you're doing than doing. This is the dangerous mode I want to warn people away from. If multiple people are collaborating on a feature, they have to be able to talk to each other, in person, real time.

There are still dangers lurking in the Two Parallel Teams mode. The most insidious is the way it undermines egalitarian, democratic, self-organizing decision making. For a team to decide things like designs, system architecture, even working practices, they have to be able to discuss and debate as a team. This requires trust, camaraderie, and personal safety. Those things are built in informal, day-to-day interactions--y'know, the way friendships are built. Over email and conference calls, they happen incredibly slowly, if at all.

You have to be comfortable with someone to be able to argue with him. The threads stitching together a global team are tenuous, and we operate with great delicacy to avoid snapping them. The communication channel is so anemic, you have to choose every word carefully, to ensure it does not accidentally offend. This is the opposite of having the confidence to know that, after we argue about this topic, we'll still be a team. Which means a lot of decisions go un-debated; they get made by default instead of by consensus.

Do I recommend globally distributed teams? No. Regardless of the project methodology, and no matter how talented the team members are, you will be dramatically less efficient if the members of your team are scattered over the planet. If you're off-shoring, then off-shore--put developers, testers, product owners, project managers, the whole team in one place. Software development is intrinsically collaborative. Structure your teams so that they may collaborate.

This experience gave me a taste, however, of how fun scrum can be. With a co-located team, you could seriously rock and roll.

3 comments:

Elizabeth said...

Hi Sharon,
This is such a great article! The IBM Scrum Community has been working on a book on distributed scrum. We just completed a few chapter drafts and put them up at www.distributedscrum.com. We'd love to get your feedback on the content and possibly include a quote from this article if you're willing.

Anonymous said...

Um .. I wonder if you would have a different conclusion if you had rotated guru's/product owners between locations, had more face to face interaction to build the trust you indicate was lacking. We have very successfully made distributed scrum work but also spent the time and money to create the trust. For example, our distributed team members are company employees, not contractors. Our product owners in the US do work crazy hours. Your writing style is great by the way ...

Sharon said...

Thanks, Vaughn. I agree with you that a greater investment in face-to-face time would help a lot. I wonder where the inflection point is where the savings on salaries makes up for the expenditure on travel, telecom, and misinterpreted requirements.