Important Traits To Help You Become A Better Data-Science Manager

Important Traits To Help You Become A Better Data-Science ManagerManagement points, based on my experience of managing a team of two data-scientists.

Ori CohenBlockedUnblockFollowFollowingJul 3Pão de Açúcar, Rio de Janeiro, Brazil.

Ori Cohen.

The following points are based on my experience leading a small team of data-scientists, however, I believe that they can benefit any future or existing manager, not only for research and data-science.

These are the values of which I try to standby, they allowed me to improve, advance and successfully manage 56 projects, from small POCs to full-size projects, in the last year, with a team of 2 very talented data scientists.

General management:Learn from your great & worst past managers.

I learned to treat team members as colleagues rather than “my employees”, to empower, trust and allow them to lead their own projects.

On the other hand, I learned that micro-managing is a terrible idea as it doesn’t bring the best in people.

Maintain a good environment where you can make mistakes, research is not bulletproof, the end is unknown and we need to make mistakes in order to deliver a working solution.

Such an environment reduces stress & fear.

Therefore, your team member will never be afraid to approach you with a mistake.

In general, mistakes help us move forward, not to mention we can always correct them.

Trust your team members, follow their ideas and advice, allow them to be creative as much as possible.

The fact that you are a manager doesn’t mean you know everything.

Try to learn from everyone around you including your team, as it will help you become a better manager.

Be their wall when criticism is at the door, it prevents them from diverging and going out of focus.

Always credit everyone involved.

Acknowledging other people, especially if you did most of the work, doesn't take anything away from your achievement or investment.

Data-Science ManagementManage DS projects with Agility (Data-science? Agile? Cycles? My method for managing data-science projects in the Hi-tech industry)Maintain a high level of communication and redundancy with all your team members by doing daily sit-downs.

When everybody is in sync, everyone can cross-communicate and fill in for you when the need arrives.

‘Early stopping’ – stop a project seems like it has no chance of success or it has reached a saturation point.

Package your deliverables, make sure they are tested and ready for production.

If possible, try to produce a ‘baby’ deliverable mid-project and allow your data-engineers to prepare a solution in parallel to the completion of your project.

There are times when someone else in the company will have an idea and you will be in a conflict or be obligated to do it.

Suggesting an A/B testing can reconcile the competitiveness nature of stakeholders, allow you to test both ideas or indirectly persuade the other party that the time and effort for their idea is too expensive.

When a stakeholder requests a certain metric, such as accuracy, ask if this is really the right metric for them.

When someone asks what kind of success level can you achieve for a new undefined project, tell them a one to two-weeks POC is needed in order to give them an answer.

Pursue knowledgeAlways be on top of current research (TDS, Arxiv Sanity, ML review, Ruder.

io)Always integrate new technology and research into existing projects, it builds experience for you and your team, alternatively, create short POCs in order to validate new research ideas.

Maintain a knowledge base (mine is 250 pages long) in order to have O(1) time-complexity when you need to learn something old.

Good links should be summarized and saved, searching google for the same thing a second time is a waste of time, it can easily shorten your work.

Allow your team time to acquire knowledge.

Reading new research is not a waste of development time, it makes us better at our job.

Allow this to happen in general and at every step of the project.

If we don’t stay on top of advanced research, new methodologies and ideas, we will be stuck professionally.

When a new research methodology proves itself, all team members should learn about it.

Publish your results, either on Medium, TDS, or Academic outlets, and always accompany code-based articles with a Github repository.

Try new thingsUnderstand product management in depth, it will make you a better data-scientist especially when you are working on a new feature with Product.

Collaboration is key!.be open-minded for external and internal collaboration with other companies or teams.

Collaboration multiplies your experience in a short time and makes for positive working relations.

Collaboration strengthens your position as a leader.

Try to look ahead and see where your company is heading and if that makes sense to you, try to be one of the first to follow this path forward.

Constantly consult with your friends regarding research ideas and methodologies, there are people out there that have a different angle and know more than you.

Be in touch with what other teams in the company are doing, by joining their meetings, going to lunch with them or setting up joint team meetings.

Invest time in mentoring, this can be done by building an internship program.

Mentoring sharpens teaching skills that are often overlooked.

I would like to thank my fellow colleagues, Samuel Jefroykin & Yoav Talmi for their invaluable advice.


Ori Cohen has a Ph.


in Computer Science with focus in machine-learning.

He leads the research team in Zencity.

io, trying to positively influence citizen lives.


. More details

Leave a Reply