Five years ago AI leaders at Fortune 1000 companies asked me questions like this:
We’re building our data science capabilities and debating organizational options. Should we centralize our data science team and embed data scientists into product teams? Or should we hire data scientists directly into product teams?
I used to answer this question by presenting the pros and cons of these options. I described situations where both solutions are effective. But today I can say with absolute certainty that the debate is over: you need to centralize your data science team.
The organizational options
The debate between centralizing or decentralizing specialists has been going on for my entire career. Should we organize our developers as a single team or embed them in project teams as part of a matrixed structure? And should we do the same with our designers, product managers, project managers, operations specialists, lawyers, accountants, or HR professionals?
Every organization debates the pros and cons of these structures as they attempt to improve organizational efficiency. Here are the general trends over the past decade:
- Product managers, project managers, and developers are usually organized decentrally because they have foundational, generalizable skills.
- Designers and operations specialists are usually centralized and then embedded into projects.
- Supporting roles like lawyers, accountants, recruiters, and HR professionals are almost always centralized. They’re assigned specific tasks without any matrix relationship.
This structure for supporting roles is rarely debated, so I won’t address it further.
With this general framework in mind, consider the following general options for organizing your data scientists.
Centralized data science teams
In a centralized system, data scientists report to an experienced data science manager. The manager assigns data scientists to work on specific organizational projects and reviews their work. The data scientist participates in standup meetings, strategy meetings, and other day-to-day coordination activities of the teams they work with.
Decentralized data science teams
A decentralized data science organization has no data science manager. Like developers, data scientists are assigned to project teams or product teams. The product manager or project manager gives them specific tasks.
Why you should centralize
From an organizational perspective, I could give you a long list of pros and cons for centralizing your data science team. But from an AI talent perspective, the choice is obvious:
A centralized data science team led by an experienced data science leader provides the best environment where your data scientists can thrive.
The market for data scientists has become insanely competitive. To make headway in this market, you need to become an employer of choice. If you don’t, you’ll experience high turnover, poor productivity, and higher costs to retain talent. For this reason alone, you should centralize your data science team.
Data scientists want to work with a leader who understands their work and who can help them develop new skills. They want the opportunity to transition between projects where they can apply new techniques and unlock the value in new data sets. They want to work with people who understand and value the experimental process.
In a decentralized setup, data scientists report to a traditional product manager or IT manager. The working environment provides none of the benefits you’ll need to attract and retain data scientists.
Curious to learn more? Take a look at Prolego's report, "Win the War for Data Science Talent." Get your free copy here.