Katie Bauer!

For the uninformed, a matrix org is one where ICs effectively have two managers. Typically, one is responsible for their career and competency in their functional role (the data science manager) and one is responsible for day-to-day task allocation (such as a PM or TPM)

To start: for each business area your team is matrixed into, you’re more or less managing a whole new team. That team will have different needs, strengths and weaknesses, and it may require you as the manager to be a different archetype from one of the other business areas' teams

And even if you are the right manager archetype to match a team or IC's situation, you might be spread across more business areas than you have time to meaningfully engage with

Having overlap (beyond technical skillset overlap) between the areas your ICs work in reduces some of the pressure on you as the DS manager because means they have more ability to support each other. But it doesn't stop you from getting pulled between the breadth and depth poles

You can only do so many things at once. Or at least, you can only do so many things WELL at once. Which is to say: I'm not inclined to believe that thinly spread matrixed orgs generate the best outcomes unless all the ICs are pretty senior

And even then, this is another situation where a lot of pressure is being placed on a single person or too few people. Teams are guardrails that allow folks to take vacation, get sick, or move on to something else without leaving everyone else in a lurch

Maybe the real takeaway here is that DS teams are usually understaffed. Matrix orgs are an attempt to make the most of a difficult situation, but IMO they require a lot of skill and humility to manage well. The fact that they're so common for DS teams is nuts