As with any other process, people are crucial for the success of a data science project. After deciding to introduce data science in their companies, the natural step is to hire someone to do the job. The debate usually arises as to what kind of profiles should be recruited for a certain job position. And it’s more complicated because data science job positions are relatively new. Much has been said on the subject of building a good data science team, but it seems like companies are still struggling when it comes to putting together a data team. This is why we are sharing experts’ experiences and tips when building a data science team.
Why and how you build a good data science team
How do companies come to the decision to start looking for data scientists? And what to actually look for in a data scientist. What experience, background, mindset do they need to have? In order to find out more on the subject, we took a look at how one international media group as Schibsted went about this process. They have transformed themselves from a traditional media company to a company powered by technology and data.
Carl Svärd, former engineer manager at Schibsted, brings light to the first moment they decided to start their journey to a digital and data-driven company. Schibsted knew they needed to do something was when they noticed a change in the reading behaviour which resulted in a continuous decline in print. But what was more worrying is that they also noticed that the main source of revenue, which is ads, was under a heavy attack from the largest tech giants on the market – Google and Facebook. The last piece of the puzzle was the emergence of a lot of new players to the media market with totally different operating models. Schibsted were also very aware that the question was not “if” they need to do something, but “how” and how fast they can do it.
Schibsted decided they need to take control over their own data and make the most out of it. They also put technology in the core of their business, and it was no longer just an extension of IT, as it has been previously. The process included shifts from data collection, to insights, analytics and data-driven products.
Companies should not be focused on finding a set of rare unicorns with exclusive skills, they should be building teams with different capabilities.
One of the most important links was, of course, building the data science team. In the process of putting together their current team, they have experimented with several ways of operating their data scientists and teams. Key findings that they discovered along the way are:
1) Operational modes
- Distributed approach – The operating mode needs to be adapted to the desired outcome of the team, and the shape of the organization. What might be an efficient move, according to Schibsted’s experience, is embedding people in other teams and departments. This so-called, distributed model works well for data analytics teams focused on delivering insights and decision support, and it enables close collaboration with stakeholders.
- Centralized approach – This approach is more efficient for data product development, which involves multiple stakeholders across the company. Companies that deliver global components and data sets can make good use of this approach.
2) Data science skills
It takes a lot of recruitment interviews to come to the conclusion that there are no unicorns when it comes to data scientists. Usually, data scientist job descriptions contain a whole array of skills, experiences and qualifications. But the reality is that rarely can one person possess all those traits. But even if a company is so lucky to happen to find that person, there is one factor that comes into play – preferences. A person can be skilled for writing production pipelines for doing machine learning models, but they still want to focus on something totally different.
3) Data science is a team sport
Data science is a really broad spectrum, and it requires a whole team of people. Companies should not be focused on finding a set of rare unicorns with exclusive skills, they should be building teams with different capabilities. A data science team set up for success contains roles like engineers, research or data experimenting people, but also data analysts and data scientists, points out Carl.
4) Type of data scientists – which one do you need?
For his experience working with different data scientists, Carl gives us an insight into two different types of data scientists – Type A and type B.
- Type A – This type of data scientists see their end goal of delivering insights and presentation material. They are the best at doing one-off projects.
- Type B – This type of data scientists want to build the whole product and be part of the product development team.
Also, the roles you give to your data scientists may be conditioned on the operational mode of the team. If you have insights team working in an embedded setting – you need people that want to do analysis. But. If you have centralized team building data product, you need people that want to build a product.
So take into account what you want to achieve with your project and determine which type of data scientists will be best fitted for it.
However, putting aside the type of organization you run or kind of project you are working on remember that sometimes the simplest model or insight can make a lot of difference – you only need the right people, the right data and do the job. You will be surprised that you can come up with more relevant insights than a dedicated team can.
Diversity in your data science team
We are quite aware of the bias in data science. One of the most efficient ways to tackle bias is to include diverse viewpoints in your team. The subject of diversity in a data science team has been much discussed among experts. Sabine Odfjell, a former Principal Consultant – Data Science at Harnham, a principal consultant working with recruitment in data and analytics, talked about diversity and why it’s important to have it in your data science team at the Nordic Women and Data Summit.
A truly diverse workplace does not only mean gender diversity – it implies diversity in age, gender, ethnicity, religion, and so on, explain Sabine. Unfortunately, data science as a community still lacks diversity today and is something we need to work on.
When it comes to RIO, it all boils down to having a diverse team.
But how diversity can help your data science team and company in general? Studies have shown that diverse workforces financially tend to outperform less diverse workforces. More specifically for data science, a diverse data team provides as many knowledge sources as possible, and it contributes to more unbiased and accurate data science insights.
Diversity can help maximize your company ROI. “When it comes to RIO, it all boils down to having a diverse team”, adds Alex Hutchings from Harnham. For example, companies with a female at the top outperform other companies by 26%. Diversity also helps with stability, it increases retention, improves product development, and challenges the status quo. Basically, it works toward the overall improved functioning of a company.
Having diversity in your data science team means stepping out of the box and infusing more creativity into your work. Also, people with different background, age, gender, ethnicity bring different experiences to the table – which helps greatly in combating unconscious bias in data science outcomes.
All things considered, yes having the right technology and quality data does matter, but the one ingredient that makes data science click is having diverse people in your team. Diversity enables you to step out of the boundaries of the imaginary box and achieve new levels of success. Your data science team needs persons that are passionate, curious and not afraid of making mistakes – because data science is all about learning from past mistakes and innovating based on them.