We'll respond shortly.
An informal survey conducted earlier at Pivotal Labs asked “how often do you rotate pairs on your project?” The answer for the Best Buy Remix team is usually every two days, but three days is the hard stop. It was about the same when I worked on the Palm Pre apps. But back on Mavenlink, we rotated ever single day.
Instantly I wondered why there was a difference. I really liked rotating every day on Mavenlink, and my gut feeling has always been that rotating that much on Best Buy wouldn’t fly at all. I think the difference is tied to story size.
At Mavenlink, a one point story is 10-15 minutes of work, like “Change the text to read…”. Two points is half a dayish of work, and four was a whole day plus a little kind of task. In comparison, on Best Buy we don’t really have many 10-15 minute tasks. Really the smallest story we see with regularity is about half a day’s worth of work – some of that is due to our client being off site and in a different time zone – so those get one point. Then two points for 1-2 days worth of work, and four points for something that might take nearly all week.
So looking at it that way, in both situations, we rotate after 2 to 3 points of work have been completed. That seems about right. Two points may equate to one day, it may equate to one week depending on your project, but I think if you use the Pivotal Tracker scale, two points is about the time you ought to think about rotating. Any more and a pair starts to “get married” and knowledge doesn’t get shared through the group. Any less and the pair can’t get enough knowledge around a problem to be able to solve it.
Despite our different rotation patterns and point values, there seems to be an underlying rule for rotation. In the middle of the scale, from trivial to monolithic, is a sweet spot where pairs can get things done, feel accomplished, and move on to a different task.