Data Science teams can provide immense value to an organization if built or it can provide no value at all. Sometime the difference in success comes down to the simple fact that you didn’t actually need a Data Science team to begin with. Other times it comes down to how you hire, manage, grow and nurture the team. In this post we’ll cover all these topics and more.
There’s many facets to evaluating a Data Scientist hire and team depending on your specific organizational needs so take these are suggestions rather than hard guidelines. You also may be tempted to hire someone who can do everything and anything but someone who can do that well is a unicorn. But unless you are very good at hiring you’ll just end up with a donkey that’s glued some plastic to it’s head. They may be competent but not in everything and not in the areas you really need them to be competent in since you didn’t focus enough during the interview process.
Good background reading is out previous post which goes over what a Data Scientist is and what other related disciplines (Data engineering, Data Analytics, Machine Learning Engineering, etc.) are.
Do you really need a Data Science team?
Data Science is not the only way to extract learnings out of data and in fact a successful Data Science team requires many other disciplines to work most efficiently.
- Should you hire a Product Manager? Do you have concrete deliverables, projects or even areas that you want looked into? If not then first you should hire someone who can help come up with those and often it’s not really a Data Scientist. Rather it’s someone who has extensive domain expertise, product management skills and the ability to seamlessly interact with stakeholders. These are the foundation requirements for defining a Data Science project and being able to push it to a successful business conclusion. That person may be a Data Scientists or a Data Science Manager but often it’s a separate role.
- Should you hire Data Engineers? Does your data reside in a well described way in a system that allows for easy access? If not then you should look at Data Engineers first to put in place the foundation that Data Scientists will later use. Otherwise your Data Scientists will spend their time acting as second-rate Data Engineers to everyone's frustration.
- Should you hire Data Analysts? If you have a simple data related question can you get it answered? If not then you likely have a lot of low hanging fruit that is better plucked by a Data Analyst. Data Analysts will be less expensive than a Data Scientists and less likely to be bored having to run simple SQL queries every day.
- Should you hire Machine Learning Engineers? Do you want to deploy machine learning models into production so that live users are impacted by their results? If so then Data Scientists aren’t the best first hire since they are not production engineers. While they can hack things into production the results will be less from optimal. Machine Learning Engineers have both the skills to build models and to do so in a way that’s most amenable to production deployment.
How to hire your first Data Scientist
So you do think you need a Data Scientist and have the foundation for them to work most effectively but who do you hire and how?
The first Data Scientist can make or break not only the team but also the idea of Data Science within a company so hiring the right person is paramount. The skills you need in your first hire are also not the same as the skills you need in your second of fifth hire.
- Don’t interview them alone: The best person to evaluate a Data Scientists is someone familiar with Data Scientists and Data Science teams. So go through your network and your company’s advisers to find someone who does and whose opinion you’d trust.
- Hire senior talent: Junior talent is cheaper and easier to find but that is for a reason. They will lack many of the organizational skills (stakeholder management, presentation, etc.) and long term technical planning skills essential for a new Data Science practice to succeed within a company. Moreover, if they are good then they will want mentorship and if you cannot provide that then they will jump ship as soon as their resume is built up enough.
- Generalists are the way to go: It’s unlikely that you know exactly what problems your Data Scientists will be solving even if you know the areas of focus and the types of data involved. As such it’s better to hire someone flexible enough to tackle different problems in different ways rather than a narrow expert in one area. You likely have so much untapped information in your data that you don’t yet need someone who can squeeze every last drop of information out.
- PhDs are for later: There is a lot of value in someone who has a PhD in a relevant field, less so if it’s not in a relevant field, however like the last point they are overkill for what you currently need. Instead look for someone who has more years of industry experience since that’s much more relevant to your problems than academic credentials.
- Willing to get their hands dirty: Your first hire, even if a manager, should be willing to do hands on work. Not only can you not afford to wait for them to hire other people to actually do the work but a non-practicing manager has a much harder time hiring quality talent for a small team. There is a reason the top tech companies vet their managers on technical skills even if they’re not going to use them day to day. Skill wise you’re going to want someone who can check off:
- Machine Learning/Statistics
- Cultural fit is a must: The first hire sets the direction fo the team and you need to be cognizant of that fact. In your first hire, don’t settle for someone who has the technical skills but doesn’t set the right tone for the team as a whole.
- Storytelling is important: To a large extent Data Science is not just a technical science but the art of presenting information to non-technical audiences. You need someone on your team who can do this and for your first hire it will be that person at least until you hire more so look for this in candidates.
How to cultivate your Data Science team
Now that you have a team what do you do to keep them functioning at peak efficiency, growing and being content with their jobs. Like any other team, a Data Science team requires attention and care to grow and stay strong.
- Don’t create a mono-culture: You want to make sure your team still gels together however you also don’t want a team that all thinks the same way. Diversity of thought is good and now that you have a team you can hire those specialists (optimization experts, economists, statisticians, etc.).
- Provide political support: Forcing the team to fend for itself is unlikely to help it grow or retain talent. Someone need to provide political cover and support for the team. The results of Data Science projects may upset some people and that shouldn't negatively impact the team.
- Do not manage them exactly like engineers: Strict adherence to scrum, agile, story points, etc. should be chucked into the waste bin. Data Science project are more open ended, more free form, and have different deliverables than engineering projects. They need to be managed as such or otherwise the result is very unhappy Data Scientists.
- Embed them within other teams: Data Scientists working in isolation aren’t going to produce their best work or work that others can take advantage of. Instead if you want impact on production systems then you should embed Data Scientists within other teams so they collaborate closely with the people implementing features.
- Break silos: If someone on your team is the single expert on something you've done then you should switch them to a different project. Silos are unhealthy as they can lead to stagnation for the team and the dangers of a single point of failure.