

This article is part of the How Do YOU CM2? blog series in collaboration with the Institute for Process Excellence (IpX). Although I receive compensation for writing this series, I stand behind its content. I will continue to create and publish high-quality articles that I can fully endorse. Enjoy this new series, and please share your thoughts!
While researching my article on Conway’s Law, I came across a study from 2008 by Nachiappan Nagappan, Brendan Murphy, and Victor Basili. They tried to predict which parts of Windows Vista would fail after release, and the best predictor they found was not in the code but in the organization.
They looked at 3,404 binaries with more than 50 million lines of code, and compared models based on code metrics such as churn, complexity, and test coverage with a model built from eight organizational metrics, for example, how many engineers touched a binary and how many of them had left the company.
To read the results, you need two terms as the authors define them. Precision is the share of binaries the model flagged as failure-prone that really were failure-prone. Recall is the share of all failure-prone binaries that the model managed to find. A model can score high on one and low on the other, so you need both.
The organizational model reached 86.2% precision and 84.0% recall. In other words, of every 100 binaries it flagged, about 86 really were failure-prone, and of every 100 failure-prone binaries, it found 84. Code churn, the best traditional model on recall, reached 78.6% and 79.9%. The authors quote Conway’s Law themselves and are honest that this is one software system at one company. Whether the same holds for hardware changes, I don’t know.
Still, it made me think about how we triage changes. A change carries technical and organizational risk, and I believe we mostly focus on the first. When we decide between fast track and full track, we ask whether the item is safety-critical, whether it involves Form, Fit, Function, or Interface (F3I) changes, and what it costs.
Basically, the organizational signals are already in front of us. How many functions, sites, and suppliers show up in the impact matrix? Is the engineer who designed the item still in the company? How many teams changed its datasets since the last release?
Imagine two brackets with the same fastener change. The first is built in one plant by the team that designed it. The second is built in three plants, comes from two suppliers, and the engineer who designed it retired five years ago. Technically, it is the same change, but the two do not deserve the same route.
A few questions to test your own process:
- Do your fast-track criteria contain any organizational signal, or only technical ones?
- Do you know which items are changed by many teams and owned by none?
Which organizational signal would you add to your fast-track criteria? Let me know what you think.
