When to centralize data work
I think one of the core questions of companies is whether they should form a central data team, including domain experts, that covers everything or decentralize parts of it by, e.g., hiring analysts
, in sales & product or even analytics engineers inside the domains.
This is a core question for machine learning teams, too. Should we have one central data science team or put ML engineers in different product teams?
For me, this question has a three-part answer: First, you’ll need to change your approach as your company changes, your environment changes, and your teams mature.
Second: Start-ups often start in one single office, one room, really. Because it’s better! Consolidation and centralization are better and cheaper in the beginning. At some point in time, there will be a tipping point. That’s what you got the third part for…
Third, It’s all about the axis of change. That’s a terrific concept, but it’s also hard to understand. The data work your company does changes in a certain way. Usually, two important influences are there: the data source side and the use case side. On the one side, your products that produce data evolve at a certain speed, grow, shrink, and change. On the other hand, use cases evolve, a new recommendation system emerges, and the sales department might change its reporting system every quarter, etc.
Now, you want to get a rough idea of the axis and the general direction of change and make sure your data work is organized to align with that. If your sales department is the most important thing in your company and is so data-driven that it needs new data reports every week, you might put your complete data team into the sales department, completely decentralized into that one function.
There’s no one recipe, but this three-part answer should help you evolve your data work.