The world’s leading publication for data science, AI, and ML professionals.

How I built a data science lab and went on sabbatical

Learn how to effectively develop data science unit and avoid common pitfalls

Undoubtedly, there is a huge demand for data analytics and most every company wants to be perceived as data driven and capable of extracting useful insights which may ultimately position it ahead of competition. With the goal of Artificial Intelligence and machine learning as enablers, data science units are and will continue to be spreading within organizations regardless of size and industry.

Nevertheless, establishing such a unit is definitely a challenge due to its research & innovation nature with all its consequences and needed capital investment. The skillsets required are currently hard to find on the market, but by getting the backing, assembling and organizing the right team you can be guaranteed success and profitability.

JLL, as a corporate real estate company (Fortune 500), manages a large portfolio of buildings for our clients and as a result generates a great amount of data. In the past, this data was primarily used for visualizations and reporting, but the question was raised, can we do something more advanced by applying Data Science and extract more meaningful useful information for the benefit of our clients? With this goal in mind we started our journey that led us to development of a Data Science Lab, up till now team of six data scientists and engineers who are delivering innovative advanced analytics solutions in area of workplace, energy & sustainability and facilities management.

Many such units fail or experience great difficulties in the early stages and our start was no different. It was a bumpy road with circling around but within a few months our team managed to establish a smoothly operating lab. Below you will find some of my most crucial non-technical tips that you may wish to follow to save time and money, cut corners and steer around avoidable mistakes. They present the most common problems new teams encounter and solutions that can be a help to any data science team to become more successful.

1. Get support from those above and next to you

Without proper backing and support your project will go nowhere. First of all, your sponsors need to understand that putting together a data science unit is much like financing a start-up, it requires investment and time to obtain profitability. The time needed depends on many variables and constraints but should be clearly defined in your project plan as it usually takes 9–18 months to begin showing positive results.

Although highly important, profitability should not be a unit’s only aim as setting up the right team brings with it other tangible benefits, including providing a data driven & innovative image for your company which may generate revenue in other branches of the business and help position the company ahead of competition. Not every client is ready for a full blown data science solutions, but there is always room for simpler products like data management, reports or visualizations.

Developing a data science unit can be much like riding the same rollercoaster most start-ups experience. Therefore, to achieve success requires having your sponsors’ trust and freedom and the flexibility to make your own decisions as you progress and adapt to the needs of a changing environment.

Additionally, the unit should be positioned within the organization as closely as possible to the CIO/CDO to ensure sufficient funding, influence and support of other units. Building a data science team is not one-man job but requires the active involvement of stakeholders, especially those that will directly or indirectly benefit from your unit’s existence.

2. Find talented people and organize them well

Building a team can be tricky and often requires many trade-offs such as finding the right balance between: budget and salaries, hard and soft skills, specialization vs generalization, the right mixture of personalities or data competencies and positions. It is good to start fresh with a new team that will be dedicated to the unit and your goals. Before recruiting new personnel from the market pool look for available talent from within your organization and then find the best ones from outside that will bring outside the box thinking and are not biased or intimidated by organization culture.

Due to the high demand, recruiting good data scientists can be quite a challenge. To secure outside help, do not hesitate to use your own and other team members’ professional networks, data science meetups and conferences for that purpose since these may be the most efficient means of recruitment.

How many data scientist and engineers are required to run a unit? Start small and grow as needed. Depending on the size of your organization, usually you can begin with 3–4 data scientists and expand from there as new business appears on the horizon.

A downside of the AI and data science hype is the shortage of skilled data scientists available. Don’t expect to find an exact match for your specific job specifications. There are wide differences in tech skills and companies perceive necessary data scientist competencies differently and train them according to their own needs, but there are always capable candidates who are flexible, willing to learn and ready for new adventures.

Hard skills are highly important but remember, if a potential candidate is lacking in certain areas they can always be trained to meet your expectations. When recruiting, focus on the most essential skills, such as machine learning, coding in R/Python and creating visualizations in Shiny or Tableau. However, your focal point should be soft skills, personality and candidates’ ability to align with your team. Preferably, put together a mixed team of personalities whom will balance the mood of the team – and from my personal experience I suggest you also include a rebel! They often help stimulate required changes, are not afraid of risks and are willing to work hard when needed.

Regardless of your organization’s culture, creativity is a fundamental element for any innovative unit. It is important to promote a casual start-up setup of specialists who have the common aim of developing the unit together. This will help support team agility, out of the box thinking, collaboration, culture sharing , openness and integrity. Don’t forget to develop your team using the individual training approach which should include online courses and attending conferences.

3. Focus on generating tangible benefit

A significant number of data science, general research & innovation units fail at a certain point due to over-engineering delivered solutions. Always bear in mind the necessary development effort, time and cost as in the majority of cases, the solutions even those based on rule-based engine often meet client expectations and there may be no need to deploy deep learning everywhere. Aim for the simplest ones that can be deployed quickly and are easy to maintain.

As data science is a relatively new concept (with old roots), many clients may not have specific needs and simply rely on you to deliver useful solutions. Do your academic and industrial research, brainstorm use cases with your team, consult with experts and your client, and for proof of concept choose only those that bring tangible benefits to your client (quick wins). Of course, by that I mean…money! So, focus on solutions that generate cost savings, foster operational performance, increase revenue or lead to competitive advantage.

When running a data science initiative, regardless if your starting point is research, PoC or a pilot, always utilize project management and software development approaches. To accomplish the research nature of such projects, adapt and tailor agile methodologies, including scrum, rapid prototyping and continuous delivery which will ensure successful completion.

Treat your unit’s development from a project perspective without the constraint of an end date. Your final product should be delivered in interim steps that you defined in a roadmap. This should serve as a masterplan with horizon for at least 2–3 years and together with your mission and vision (or equivalent) should be defined as early as possible (alter when necessary as you progress).

4. Foster research and creativity

This is easily said, but how does one realistically spark curiosity, learning and creativity within a team? These are fundamental factors of success for every data science unit. You will often find yourself navigating unchartered waters, researching new methods and applying them to topics for the first time. Therefore, you need a team that is able to work together, share ideas, brainstorm and come up with solutions – but how is this done?

It is important to not overload your team with projects and leave them some room for research and self-development. You can even sanction special times such as dedicating every Friday for creativity sessions, learning new technologies, researching chosen topics etc.

Diversity is another key element in relation to a team and their projects/responsibilities. The combination of different personality traits within a unit can definitely initiate constructive discussions.

Reinforce communication by promoting an open, informal culture, knowledge sharing and introducing creativity sessions or community of practice may broaden boundaries of each member and the team. Additionally, by mixing project teams, shifting areas of focus and responsibilities, teams will acquire a broader perspective of the business and will be able to perceive topics from different angles.

The role of the leader is also to provide vision and define the purpose of initiatives undertaken. Giving team members a broader perspective in reference to organizational goals, instils in the team that they are an integral part of the bigger project and helps them understand how their personal work fits into the overall strategy.

Team members should also have autonomy and trust to conduct projects and make their own decisions. Leave room for failure and provide coaching when necessary. As a result, you should have self-driven and creative data scientists who can research a topic, define the needed methodology for PoC and conduct the project end to end.

5. Marketing and UX

In real estate we love to say – location, location, location. In the case of ensuring the successful start of a budding data science unit I believe the key words are – marketing, marketing, marketing. Even the most sophisticated solution that can generate enormous savings would struggle with adaptation without proper marketing. Therefore, invest heavily in packaging your solutions as products with marketing materials, including a website, using social media and other available channels. Since you cannot be an expert in every field, seek support from your company’s marketing and sales people.

Nevertheless, the key factor of success for any data science solution is how insights are delivered to the end user and how he can interact with the solution. Therefore, do not limit yourself only to data visualizations in Tableau, Shiny or other similar products but rather invest in proper UI and java script web applications development.

Hire specialists or consultants ** and train the rest of your data scientists in modern UI/UX design because it is your frontend that reflects clients’ needs, delivers usability and provides an enjoyable experience which will increase your clients’ appetite for more, and significantly boost your sales. After all, visual perceptio**n is the key.


And what about that sabbatical?

As I am writing this article I am in Charlottesville, the green homeland of many US presidents and place of many contradictions. During the last few months our family expanded, my wife received a prestigious research grant at the University of Virginia and we decided for the three of us to spend 4 months in the US. So, I am now on sabbatical/parental leave focusing on family and research that I neglected recently – writing an academic paper, meeting with researchers and broadening my horizons.

It was definitely not an easy decision to leave my team at the scaling-up stage but I spent weeks preparing them for that occasion. Most of my team are experienced, self-driven data scientists who can overlook their projects on their own and continue to generate new business. Nevertheless, I took into consideration that especially the creativity aspect of the team may suffer due to a lack of clear Leadership and research guidance. Therefore, the unit is currently focusing on propagating existing products to new clients rather than defining new solutions. However, in the short term, the effect of my being absent should have limited impact and rather serve to help mature the team.

As a final advice – be persistent in building your data science unit. You will encounter many roadblocks but remember – if you don’t make it happen nobody will. Like most of us, you may also not be an expert in every topic, so although it may be difficult try to surround yourself with people who will fill your gaps in knowledge or expertise – preferably people more accomplished than yourself in certain areas.

Persevere and you will succeed – wishing you the best of luck!


Certainly the article just scratches the surface. So please feel free leave a comment with your own experience or view on the topic or contact on LinkedIn.


Related Articles