Day Care

Alias: "Progress Team / Training Team"

external image DayCare.jpg


... the project has just brought on several new people.

✥ ✥ ✥


Your experts are spending all their time mentoring novices.

You begin to hear things like "We are wasting our experts," or "A few experts could do the whole project faster." Indeed, the experts are not proceeding at the rate you or they would expect, because training the new people is draining their energy, time and concentration. But the new people must be trained, by experts, of course.

At the same time, you must make progress on the project itself.

Therefore:

Put one expert in charge of all the novices, let the others develop the system.

Separate an experts-only "progress" team from a training team under the tutelage of one or more mentors. Select the mentors for their ability to teach design and programming (object-oriented design and programming, for example) to novices. Let the progress team design 85-95% of the system, let the training team focus on quality training, delivering only 5-15% part of the system. Transfer people to the progress team as they become able to contribute meaningfully.

Make sure that the training team does not simply do training exercises, but actually contributes to the final system in an ever-increasing way.

If you have many people to train (more than, say, six), you will have to design a series of tasks for them to attempt. Otherwise you may give them a small, real part of the main system to design.

If the people in the training team are the ones who know the domain, you will have to make some further adjustment, or else the division may cause conflict.

✥ ✥ ✥

The result is that most of the experts can continue to make progress on the project. The novices contribute a small part of the project, that grows as they gain experience.

In extreme cases, though, you eventually have too few people to constitute a progress team.

How many people can one mentor train, if training results and not running software are his/her deliverable? A small, reasonable number is five. I have heard of one person mentoring 15 people on five concurrent mini-projects.

Related patterns:

This pattern is a cross-specialization of several given in this chapter: OwnerPerDeliverable, SomeoneAlwaysMakesProgress, TeamPerTask, SacrificeOnePerson.

Principles involved:

The principles are synergy vs. distraction, the synergy of having a novice learn directly from an expert vs. the distraction to the expert. Experts having to answer novice questions are reduced to a fraction of their productivity, without particularly raising the productivity of the newcomers. Adding one novice to an expert may cut the expert's productivity in half, adding two may cut it to a third, adding three may prevent all productivity altogether.

Assume there are X experts who work at productivity 1 each, a larger number of N novices who work at n productivity each, with n much smaller than 1, on the order of 1/10. If the experts could work together, they would have, in this simple model, a total productivity of
    (X)    for the experts working together.
 
If one of them is sacrificed to train the novices, that person has zero productivity (except training novices), so the group's total productivity is
    (X-1) + N*n    for Day Care (upper curve in figure 8.1).
 
If they are all mixed together ("Even Mix"), m=N/X novices per expert, each expert's productivity falls from 1 to something like 1/(m+1). The group's total productivity is now
    (X*X/ (N+1)) + N*n    for Even Mix (lower curve in figure 8.1).
 
external image DayCareVsEvenMixFigure8.1.gif
Figure 8.1 shows the productivity of DayCare versus EvenMix, novices assumed to work at 1/10th the productivity of the experts. This shows the total productivity for the team in units of experienced people's productivity. As the number of novices increases, the EvenMix line shows the effect of training them. Let us check that the assumed productivity difference is not skewing the results. Figure 8.2 shows the ratio of DayCare to EvenMix, for different productivity assumptions. Note that with five experts and five novices, the ratio is actually just below one, meaning that the experts are absorbing and making use of the novices. By two novices per expert, DayCare is already considerably more effective.

external image ProductivityDayCare.gif


The nature of the training does not matter. Design and teaching are antagonistic tasks (as described in TeamPerTask), and better split into separate teams.

Treating the delivery of trained people as separate from the delivery of running software gives you access to OwnerPerDeliverable. SomeoneAlwaysMakesProgress protects the delivery of running software.

Sample situations:

A. Mentoring. The standard recommendation in the industry is to put 1-5 novices under each trained expert. The consequence is that the experts spend the prime part of their energies training, halfheartedly. Besides being drained of energy for designing the system, the experts typically do not have the personality, background or inclination to actually teach the novices how to do design. They are caught between trying to get the maximum out of their trainees and trying to do the maximum development themselves. Thus, they neither develop the system, nor train the novices adequately.
Some companies have dedicated "Apprenticeship" programs, in which novices are put under the tutelage of a dedicated mentor for 2 weeks out of every 3 for 6 months.

B. Adding staff. Fred Brooks, in The Mythical Man-Month, talks about the training costs of adding people to a project. These new people drain productivity from the experts. The same suggestion applies: put the newcomers in a separate team to learn the system. Move them to the progress team as soon as they are up to speed.

Reading:

Brooks, F., The Mythical Man-Month, Addison-Wesley, 1995 [BibRef-Brooks1995].

In Situated Learning: Legitimate Peripheral Participation, [BibRef-Lave1991], Lave and Wenger describe the use of this sort of arrangement in apprentice-based work situations.

A version of this pattern first appeared in [BibRef-Cockburn1998].