Reliable Machine Learning

Chapter 13. Integrating ML into Your Organization

Integrating any significant new discipline into an organization often looks more like an exercise in irregular gardening than anything else: you spread the seeds around, regardless of whether the ground is fertile or not, and every so often come back to see what has managed to flourish. You might be lucky and see a riot of color in the spring, but without more structure and discipline, you’ll more likely be greeted by something barren.

Getting organizational change right is so hard for plenty of general reasons. For a start, an effectively infinite amount of material is available on how to change organizations and cultures. Even choosing from this plethora of options is daunting, never mind figuring out how best to implement whatever you settle on.

In the case of ML, though, we have a few domain-specific reasons this is true, and arguably these are more relevant. As is rapidly becoming a cliché, the thing that is fundamentally different about ML is its tight coupling with the nature and expression of data. As a result, anywhere there is data in your organization, there is something potentially relevant to ML. Even trying to enumerate all the areas of the business that have or process data in some way helps to make this point—data is everywhere, and ML follows too. Thus ML is not just a mysterious, separate thing that can be isolated from other development activities. For ML to be successful, leaders need a holistic view of what’s ...