Effective Machine Learning Teams

Chapter 11. Effective ML Organizations

I’ve given all I can,

It’s not enough

“Karma Police” by Radiohead

Have you ever worked on a great team that nonetheless struggled to have an impact because it was being asked to do the wrong sort of work, or focus on too many things at once, or was constantly waiting on other teams to get things done?

On our journey so far, we’ve looked long and hard at what happens inside ML teams, including the complementary mechanics of work practices, how we configure our technology, and the humanistic elements of teamwork. We’ve also looked at the interplay between product and ML delivery in teams. This chapter takes us a step further, examining how team effectiveness is moderated by organizational factors and, in turn, how effective organizations are built from effective teams.

This organizational perspective is critical to building effective ML teams, because even the best team—a team with finely tuned dependency management, continuous delivery, automated testing, supercharged IDEs, trust, communication, diversity, and purpose—will be beaten by a bad system of poorly shaped work or excessive organizational dependencies.

From a systems perspective, a team transforms work into outputs. The work may come directly from interactions with a customer or it may be defined by another dependent team in the organization. The team’s outputs either directly serve the customer or serve dependent teams. Poorly designed work and output dependencies between teams ...