Team Per Task


... a big diversion hits the team, threatening to disrupt the ongoing work, and temporarily halt progress.

✥ ✥ ✥

Large distractions (usually called crises) must not be allowed to stop a project, even for a short time. Crises are inevitable, and they are legion. If the project takes time to respond to each, its members will soon find themselves spending so much time responding to each crisis that the real work doesn't get done.

Of course the diversions are real. A previous release needs an emergency bug fix. New people must be trained. The ISO audit will happen. But they must be handled in a way that the project still moves forward.

The temptation is to take this high-priority item and throw everything you have at it: let the whole team work the issue until it goes away. However, such an approach confuses urgency with amount of effort; some problems require only a small amount of attention, although it should be immediate attention. A stitch in time saves nine.

Therefore: Let a sub-team handle the diversion, which allows the main team to keep working.

One approach is to split the team. Sort the activities so that each team has a primary task with additional, sympathetic activities. Sitting in meetings, answering phone calls, writing reports, for example, are non-sympathetic to designing software. Arrange it so that each team can focus on its primary task, and each task has a dedicated team member.

✥ ✥ ✥

The result is that the important distractions are handled pretty much entirely by specialized teams, thus allowing the main team to continue uninterrupted.

However, one must be careful not to overdo it. Carried to extremes, it results in single-person teams. In addition, while solving a crisis is important, be careful not to heap praise too lavishly on the crisis teams. Otherwise, addressing crises becomes the glamor job, and the focus of the team becomes putting out fires rather than building the building. (See CompensateSuccess)

Related patterns:

This pattern treats each task both as an activity and as a deliverable. Therefore:

OwnerPerDeliverable - the general form of ownership and accountability.
FunctionOwnerAndComponentOwner - team for each artifact, as well as the task of designing it.
SomeoneAlwaysMakesProgress - the general distraction management pattern.
SacrificeOnePerson - specialization to lose only one person.
DayCare - addresses training as a separate deliverable from the software.

Principles involved:

Increase flow time and decrease distractions, thus trading personnel parallelism for time slicing. "Flow" is the quiet time in the brain when the problem flows through the designer ([BibRef-Csikszentmihalyi1990], [BibRef-DeMarcoLister1976]). It is when the design alternatives are weighed, and decisions are made in rapid succession as mental doors open. The problem, the alternatives and the state of the decision process are all kept in the head. It is a not only a highly productive time, it is the only time when the designer feels comfortable making decisions.

It takes about 20 minutes to reach the internal state of flow, and only a minute to lose it. Beyond getting into flow, the designer must have time to make actually progress, which may be another 10 minutes. Any significant interruption within the half hour minute essentially causes the entire half hour to be lost. As it takes energy to get into the flow, a distraction costs energy as well as time.

To increase flow time, distractions have to be reduced. Certain pairs of activities are more mutually distracting that others. Fixing a bug requires flow in the old system, hence distracts from flow in the new system. Sitting in meetings, answering questions and time on the telephone are major distracters to design flow. Therefore the recommendation to group tasks into sympathetic sets. Requirements and analysis involve meetings, reading, and writing. Design and programming require concentration on the implementation technology and keeping a great number of details in the head.

Parallelism vs. time-slicing. Time-slicing can be more attractive in terms of job satisfaction - each person will do design some part of the time. The significant time to switch between tasks causes parallelism to be preferred in this case. Some of the people may adopt the new task as their profession (see SacrificeOnePerson, DayCare, and FireWalls for examples).

Sample situations:

A. Concurrently gathering requirements and designing software.

Project Winifred tried having each person do requirements, analysis, design and programming. We thought the developers would enjoy the change of activity, that this would reduce the meetings and bureaucratic documentation exchanged between people.

What happened was that the first two activities were so different from the latter two that people were unable to switch easily between them. After having attended and documented meetings for much of the day, it was difficult to start working on the design and programming. As with bug-fixing / new-development, every time a designer was pulled away from her or his work, it cost an additional hour to recover the train of thought.

We applied TeamPerTask, and split the teams along task lines. Requirements gathering and analysis went with designated people in each team, and design and programming went with the others. The result was that the requirements/analysis people sat in meetings, read and wrote specs, examined interfaces and the like. They communicated their findings to the designer/programmers - orally, for the most part, since they were closely linked on the same team (HolisticDiversity). The designer/programmers stayed in their train of thought, getting fresh input from their requirements colleagues. Some of the people put onto requirements really wanted to program, so this was quite a sacrifice for them (SacrificeOnePerson).

Two things we did not do. We did not put the requirements/analysis people into a separate team (HolisticDiversity again). A team was jointly responsible for a section of the system, from requirements to delivery. The splitting was within each team. We also did not require the requirements group to document their decisions for the designers benefit (they did document for the project's benefit). The requirements and design people were in close contact at all times, and most information passed orally. There was, therefore, no "throw it over the wall" effect. These were both important teaming decisions made earlier, which we were intent on preserving.

B. Training is distracting the experts. See DayCare.

C. Other examples under SacrificeOnePerson.


See [BibRef-Csikszentmihalyi1990] and [BibRef-DeMarcoLister1976].
A version of this pattern first appeared in [BibRef-Cockburn1998].