Photo by charlesdeluvio on Unsplash

The Art of Hiring Great Data Scientists: Best Practices for Building a Strong Data Team

Note this article was co-authored by

Karen Church

& I

As leaders in a fast-growing scale-up, we know that hiring great data scientists is crucial to driving evidence-based decisions. So how do you identify the right talent? In this post, we will explore some best practices for building a strong data team based on our experiences of hiring for the RAD function at Intercom.

1 Define your needs and the role

Before you start the hiring process, it’s important to reflect on the type of candidate you would like to hire today, and be open to the possibility that they may have different skills and experiences compared to your past hires.

This might be because you’ve learned through experience about what type of candidates make the best hires at your specific company — we’ll talk more about this in later sections. It might be because your team structure has evolved. Or it might be because the business needs have changed. Either way, if it’s been a while since you last advertised an open role on your team, take a moment before re-posting the same job description and ask yourself whether it best represents what the team needs now.

For example, in our most recent round of hiring, we updated our job description to:

• Call out that RAD data scientists, while part of the Product org, partner with go-to-market teams like Finance, Sales and Marketing — we have found there is huge value in bringing product analytics insights to these teams.
• Remove skills like Python/R from our “What skills do you need” section for Mid-level and Senior data scientists — we’ve found that these skills are not a must-have for someone to be successful here.

2 Look for more than technical skills

Each round of hiring we’ve done has strengthened our conviction that technical skills are not the most important thing when it comes to hiring great data scientists for a team like RAD. What some folks call ‘soft skills’ — curiosity, a passion for the business, communication and collaboration — are the most critical for data scientists to drive impact.

Make sure your job description and your interview process can identify candidates who are:

• Curious by nature: folks who thoroughly enjoy exploring data, asking questions, and seeking out insights. Look for candidates who are eager to learn and experiment, who have a track record of pursuing interesting projects, and who are not afraid to challenge assumptions.
• Passionate about understanding and improving the business: folks who have as much passion for the business as they do for data science. This can come through as a strong interest in the product, the market and your competitors.
• Excellent communicators: folks who can articulate complex ideas in simple ways, folks who can communicate clearly and concisely, folks who can communicate the why and the impact of the work in compelling ways. Data science is designed to help drive effective decisions. Data scientists need to be able to influence those decisions which requires effective communication.
• Proactive collaborators: in order to drive actual impact, data scientists will need to work closely with partners in functions beyond their immediate team — product managers, designers, engineering, sales teams, marketing teams, customer support teams, finance teams — and so strong collaboration skills are a must. Those partners love working with data scientists who bring their own opinions and ideas to these collaborations, rather than just answering the questions they are asked.

3 Assess Skills, Behaviours and Culture Fit

While the ‘soft skills’ we listed above are critical, they can be harder to test for in an interview than technical skills like the ability to program or write SQL. At our onsite interviews (the final stage in the process), we ask candidates to participate in 4 sessions over 2.5 hours, which are now heavily weighted towards those critical (soft) skills:

1. A 1 hour panel presentation where we ask candidates to present an impactful past project they worked on. The audience for the presentation includes people from Data Science, Research and Product Management. When preparing candidates for the onsite, we ask them to spend most of the presentation on explaining the problem, what the impact of their work was on solving it and how they collaborated with others. We ask them not to get into the weeds of the methods they used as our aim is to get a strong read on their storytelling skills and their focus on impact.
2. A 30 minute behavioural interview with a Product Manager, to test product thinking, collaboration and problem-solving skills. The aim is that this runs like a real working session, where the interviewer will introduce a situation based on a recent, real-life event, e.g. “we made a product change and in the period immediately following the change, customer activation rates dropped”, and the candidate then works through this problem with them, asking clarifying questions, coming up with hypotheses and proposing actions to resolve it.
3. A 30 minute behavioural interview with a senior RAD team member, this time focussed on Intercom’s company values. For example, Intercom has a value of “impatience”, so we will ask the candidate for an example of a process in a previous role that slowed them down and what they did to overcome it.
4. A 30 minute paired SQL session with a data scientist from the team. The interviewer will give the candidate access to a database and then ask a series of questions designed to test basic SQL skills: filtering, ranking and joins. We know that this basic level is needed for someone to be able to hit the ground running when they join the team and have strong impact.

This approach has evolved considerably from what we used to do. In the early years of RAD we over-indexed on technical skills and did a 3 hour technical onsite to assess a candidate’s ability to produce an analysis based on an Intercom dataset. We realised this was making big demands of the candidate — the dataset was unfamiliar to the candidate so they had to spend significant time figuring out what it represented — while not giving us great signals on the skills we really cared about like how they tell stories with data. It’s unrealistic to expect a candidate to form a strong opinion and put together a coherent story based on unfamiliar data about a product they only have a surface knowledge of.

We’re confident our new process, while still challenging, strikes a better balance. And we’ll continue to reflect and iterate as needed as we continue to grow and scale the team for impact.

4 Embrace and Foster Diversity and Inclusion

Building a diverse and inclusive data science team is not only the right thing to do, it’s great for business. Research has shown time and time again that diverse teams are more innovative, they are smarter and they ultimately lead to better performance. Yet most studies estimate that just 15%-25% of data scientists are women [here, here]. Make sure your hiring process is designed to attract a diverse pool of candidates, and be intentional about creating a culture of inclusivity within your team.

Things we have applied over the years to help foster diversity and inclusion within RAD are:

• Investing time upfront to build more diverse hiring pipelines, e.g. reaching our directly to diverse candidates about our open roles to ensure they are aware of open positions.
• Reviewing our job descriptions to remove any bias
• If/when possible, ensuring our onsite panel is diverse.
• Showcasing the people that make-up RAD so that folks get a sense of the diversity of the team they are joining.
• Establishing a clear mission, vision, and set of values to help team members works towards a shared vision.
• Emphasising open, transparent communication across the team and celebrating the wonderful folks who make up the team

While there are many more facets of diversity than gender, we’re proud to highlight that 50% of the Data Scientists on RAD identify as female.

5 Reflect, Learn and Iterate

A successful, high-performing data science team is not built overnight. It requires a thoughtful and iterative approach where you intentionally reflect on, learn and improve your hiring process and practices over time.

When we review our hiring process we reflect on the data scientists we have hired and we look forwards at company and RAD org needs.

For the data scientists, we ask ourselves: have they been successful? Have they added value to our team and company? Are the skills we assessed at the time of hiring showing through in their daily practices? If the feedback here is positive, we’ll work with our recruiting team to figure out how we can source more candidates like this. If the feedback is mixed, we’ll run a more detailed retrospective on our job descriptions and interview process, to see if there are opportunities to refine further.

For the company needs, we reflect on where the company is headed, any refinements to company strategy and what types of skills we might need to amplify on the team as a result. For example, Intercom’s positioning is now very focussed on Customer Service. This means that we are more open to candidates with a background working in Customer Service operations.

For RAD org needs, we reflect on the future of RAD, our team vision and key programs we drive. For example, we’re bullish on the combination of quantitative and qualitative insights and are keen to see the lines blur more across researchers and data scientists. This means we’re especially interested in data science candidates who have experience partnering closely with research and/or who have ideas on how we can shape the future of an org like RAD.

Conclusion

Building a strong analytics and data science team starts with hiring the right data scientists. By looking critically at your hiring process, from job descriptions through to the interview process, and continually reflecting and iterating, you can build a happy, healthy, high performing team that has a strong impact across your company.