Where are Data Scientists in your organization? IT side, Business side? If Business side, how do you avoid having a "shadow IT" team growing?

The structuring of data science teams can vary significantly across organizations.

Centralized Model:

In this model, data scientists are brought together into a single, unified team that reports to a common management structure.
Advantages:
Career Paths and Mentorship: Data scientists benefit from being part of a team with peers from the same discipline. This facilitates career growth, skill development, and mentorship.
Alignment with IT & Engineering: Centralization fosters close alignment between data science and IT & engineering teams, promoting collaboration and efficient problem-solving.

Decentralized (Federated) Model:

Data scientists are embedded within specific product or business teams. Each team has its own data scientists who work closely with domain experts.
Advantages:
Domain Expertise: Embedded data scientists gain deep domain knowledge and context, enabling them to address specific business needs effectively.
Tailored Solutions: Teams can focus on use cases relevant to their area, leading to customized solutions.

Hybrid Model:

Combines elements of both centralized and decentralized approaches.

In summary, the best practice is to consider the organization’s size, culture, and business needs when structuring data science teams. Regardless of the approach, fostering collaboration, career growth, and alignment with business goals are key considerations.

In addition,  addressing data management challenges is essential to prevent the growth of “shadow IT” within or outside data science team, therefore following principle are key to manage
Collaboration: Both IT and data scientists must work together to develop data management strategies that meet each team’s needs.
Awareness: Increase awareness on both sides. Developers learn about security risks, while IT understands how to deliver tools matching developers’ needs.
Speed and Flexibility: Address pain points by ensuring faster databases and flexible data tooling.
Experimentation: Enable data scientists to experiment rather than merely pulling queries.
Security and Compliance: Educate data scientists about security risks and compliance requirements.

By fostering collaboration and addressing these challenges,  one ca maintain  a robust data science practice while avoiding the pitfalls of shadow IT

In my case, is into IT, to be sure is working for full company innitiatives. Other way, you will need one by business mature area, even with a different aproach to solve same business issues.

Create a collaborative/transparent working model/committee, that can provide IT oversight/guidance (trusted advisor), while ensuring the business has flexibility to deliver/innovate within framework