What is the best way to build a data science team with the right mix of skills?

Having formed multiple Data Science teams, I would like to share a few tips that will be useful in building a Data Science team with a right mix. 1. We will need a mix of Solution Architects, Data Scientists and Machine Learning Engineers while working in a Project. That way the Solution Architect focuses on framing the solution and Data scientist focuses on implementing it and the Machine Learning Engineer works on building the pipeline to making it work. And I personally do not believe in take home assignments while hiring a candidate. I prefer giving live scenarios and see how the candidate solves the problem. That way it is easier to evaluate their problem solving skills and the methods they choose to solve the problem.

A few years ago I build a Data Science Team at DB Cargo AG - my two most important lessons were that on the one hand you have to ensure top management support: you need this support when it comes to having the financial means to build the team, when you want to access data, when you set up the first projects with the departments that will become your internal customers, and when you start to implement measures based on your team's findings. Gaining this support includes a lot of education and gentle expectation management. On the other hand, you need boots on the ground. That is, you need to work with units who are willing and able to discuss your findings and to implement improvements - without them you end up creating pointless analyses.

If I want to be a leader of a Data Science team, I prefer to have these experts: Data Engineer, Data Analyst, Data Scientist, MLOps Engineer, Domain Expert Soft skills is so important: As a team leader, prioritizing soft skills is key to seamless coordination and task execution. Communication, empathy, and adaptability underpin our team's success. I foster an environment where open dialogue, mutual respect, and constructive feedback are the norm. Regular team-building exercises enhance our camaraderie, ensuring that each member feels valued and heard. By clearly defining roles and encouraging collaborative decision-making, our team operates with a synchronized vision, turning individual strengths into our collective power.

I've built several Data Science teams, and the key lesson is to tailor team design to each project's unique needs. For a successful data-heavy project, consider core roles: ML Architect: Translates business problems into data science solutions and oversees technical delivery. Lead Data Scientist: Manages and leads data science aspects of the project for quality outcomes. Data Scientists: Execute day-to-day data science operations and model development. Data Engineers: Ensure data pipeline and infrastructure support. MLOps Engineer: Critical for production and maintenance of machine learning models. Project Manager/Business Analyst: Aligns technical work with business objectives, ensuring project success.

Companies are evolving along with technology. AI platforms are here to ease deployment into production (versioning, data controling, and so on), automating data cleansing and offer feature stores. Gen AI is here to automate even more the process. Nowadays you can create automated ETL jobs (without any data enginerer), get access to your company data on the spot via chatbots (data analysts do not have to create noumerous and many times useless reports). You need to focus on what Gen AI struggles to handle. That is searching for people with soft skills. People who have business acumen, understand your industry easily, have strong statistical background to intreprete the findings, take initiatives, are teamplayers and have good brainstorming.

A breakdown of key roles from my perspective: 1. Data Scientist: The analytical brain driving the quest for insights. - Domain Knowledge, Modeling, and Analysis. 2. Data Engineer: The backbone of all data operations. - Data Access, Pipeline Construction, Database Management. 3. Software Engineer/Developer: Translates analytical solutions into functional tools. - Implementation, Integration. 4. Operations Engineer: Ensures scalability and maintenance. - Deployment, Maintenance. 5. BA or Product Owner/Manager: Serves as a bridge between the technical and business realms. 6. The Citizen Data Scientist: An emerging role without formal training. - Democratizing Data, Using tools to derive insights without deep technical knowledge.

The most important thing to "solve for" is to match your Data Science team to the organisation's current needs for Data Science, and those of the reasonably foreseeable future (eg. 6-12 months ahead). Depending on the types of data available to work with or the types of problems you will be solving , this determines the mix of roles and even inside a group of roles, the particular skillsets you need eg. NLP, image processing etc. Its also very important to consider from the beginning how Data Science products will be supported and maintained in production. ML Ops roles and ML Ops processes will be essential to support ML or many models in Production.

Data science requires a combination of skills, from web scraping to collecting data in various formats, which is referred to as data wrangling. Then, there is a need for skills to integrate and make the data usable, including activities such as data exploration, data cleaning, and presenting it in the required formats for further use. Once the necessary data has been collected, it needs to be prepared for processing. This processing can be done using prepared tools like Power BI or through coding by skilled professionals. This stage of work requires domain experts and individuals with expertise in machine learning algorithms and artificial intelligence.

Begin by understanding what your organization wants to achieve. This will help you figure out the types of people you need. Roles like data scientists and data engineers are common. You want team members who are good at technical stuff like statistics and machine learning, but also those who can solve problems. It's also important that your team works well together. When you hire, consider both experienced pros and recent grads. Use technical interviews to test their skills and make sure they fit with your organization's way of doing things. Give your team the tools and training they need, and keep pushing them to learn. They should work closely with other parts of the company, and be ready to change as the company's needs change.

Problem Identifier & Problem Solver A problem identifier is crucial as it helps to define and understand the problem to be solved. It involves framing the issue, setting clear objectives, and ensuring alignment with the stakeholders' needs. This step is fundamental as it guides the entire data analysis process. Without a well-defined problem identifier, teams might end up working on vague or irrelevant issues, wasting resources and time. A problem solver, on the other hand, is essential as it brings about the expertise, tools, and techniques required to address the identified problem effectively. The problem solver implements strategies to derive actionable outcomes and recommendations based on the problem identifier.

Many capability frameworks and audits will focus on technical skills. However in many situations for a Data Science team to be effective, the team will need significant domain expertise, for example, Retail, Aviation, Automotive etc, so they understand the "context" in which their models and "products" will exist. Equally, the "soft skills" needed for Data Scientists to interact effectively with other teams, especially non-technical teams eg. Product, Finance, the Executive Lead Team etc. will be essential. And again are often under-valued in any assessment. Highly performant Data Science teams have the necessary domain knowledge and can articulate and present their work in a clear and simple way, free of jargon.

Building a robust data science team starts with a capability assessment. Consider a real-world scenario where a company wishes to enhance its predictive analytics. Using frameworks like the Data Science Maturity Model, the company can identify strengths and weaknesses in existing skills and tools. Recognizing gaps, such as insufficient expertise in machine learning, guides targeted recruitment or upskilling, enabling the creation of a balanced, capable team ready to unlock valuable predictive insights.

It's crucial to identify complementary skill sets that align with the specific needs of the projects at hand. Encouraging an open and communicative team culture facilitates knowledge sharing, where members with different expertise can learn from each other.

My method is to rank each person on the skills and characteristics that matter to our mission, and overlay all of those people on a parallel coordinate plot. If I see that we have an important area where we don't have a 5 out of 5 team member, that's the attribute we hire for.

I strongly believe that capability assessments are typically used for the wrong purposes. Yes, of course you need to verify that candidates possess the skills needed for the role. But skills can also be learned more quickly than ever before thanks to ever-improving training and knowledge sharing. Capability assessments, regardless of format, need to provide insight into how a candidate thinks and approaches complex problems. If you find a candidate that shows excellent problem-solving ability but lacks specific skills, they are far more valuable than one who may have a broader toolkit but is inflexible in their approach.

To understand the skills that are required for your data science team, it's important to take into account several key factors. These include the specific goals and objectives of the data science team, the industry or sector in which the organization operates, the team's structure, the data sources that you handle, the types of data you work with, data protection, privacy, and explainability requirements, and many more. Considering all these factors will help you determine the right blend of skills.

The best way to build a data science team is to have a blended skillsets across technical skills revolving around data engineering, machine learning engineering, and pure mathematical data science with a mix of soft skills revolving around business understanding, domain knowledge, communication, presentation and collaborative skills. Technical skills drives innovation & ROI. Soft skills build team spirit, help think out of the box, bridge the gap between business and data.

Para além das habilidades em ferramentas e técnicas, que são essenciais a qualquer análise, é fundamental conhecer o mercado em que se atua. É dessa forma que se torna possível fazer as perguntas certas aos dados e assim encontrar as respostas mais adequadas.

Capability assessment is a crucial step before building a data science team. Utilize frameworks like the Data Science Maturity Model or the Data Science Capability Framework to evaluate skills, tools, and processes. Assess not only technical skills but also domain expertise and soft skills for effective collaboration with non-technical teams. Identify gaps and opportunities for improvement, focusing on critical areas such as data strategy, data quality, data governance, data literacy, data culture, data ethics, and data innovation. A well-rounded team with both technical and non-technical skills is essential for high performance in data science.

When hiring a junior data scientist, look for strong problem-solving and business acumen, as well as clean and well-organized work. When hiring a senior data scientist, look for the ability to ask predictive questions and design effective solutions, as well as important soft skills like communicating and collaborating with cross-functional teams. Additionally, value humility and a willingness to learn in both case.

Gartner's 2022 analysis highlighted a challenge: requiring over two years of experience and three skills could reduce the data scientist candidate pool by 85%. In response, I've shifted my approach from merely filling vacancies to envisioning our team's future. While traditional qualifications remain essential, I also prioritize nontraditional attributes like industry experience and competencies gained on-the-job. Once on board, our upskilling pipeline blends formal education with 6 months of hands-on training, crowned by continuous mentorship. This holistic approach ensures a fusion of expertise and adaptability in our team.

Motivation and willingness to learn together with strong problem solving skills, analytical skills and a pinch of creativity are key. That's what should stand out in a candidate. Additional desirables are business acumen and communication skills. Nowadays a lot of people are skilled in "computing", but not many know what the right thing to "compute" is.

Manage the level of seniority in the team and ensure a healthy balance. I am a huge believer in upskilling and hiring for potential. Tailor the roles to the actual and slightly forward looking needs. Do you need a strong methodology and math background, a lot of business acumen, vast cloud and MLOps tooling expertise, someone who is fast at providing various POC solutions etc.? During the hiring process provide a clear as possible picture to candidates about the role and the company.

Others have highlighted well various tips for recruitment so I’ll offer a slightly different take on this. Outbound, outbound, outbound! Don’t expect an excellent candidate to find you, go get them! Send messages on LinkedIn, forums, wherever and try to establish an authentic connection. Have informal chats, share your company’s vision, show the candidate the goal before doing the “formal stuff.” It takes more time, but you find better people this way. Oh, and don’t underestimate the power of peer-to-peer mentorship if considering a Jr. hire. I’ve never seen my employees develop faster than in a good Jr-Sr pairing.

In my experience, the most crucial capabilities within a Data Science team are not necessarily technical, given the rapid evolution of tools and techniques. Instead, it's more important to find candidates who are highly motivated and have a strong orientation towards problem-solving. To identify such individuals, it is beneficial to consider their contributions to Open Source projects, their participation in conferences, and their personal projects. Moreover, offering remote or hybrid work arrangements can be advantageous, as professionals with these profiles often require a high degree of independence.

When recruiting, prioritize individuals with diverse skills such as statistics, programming, domain expertise, and problem-solving abilities. Look for candidates who not only possess technical proficiency but also demonstrate creativity and a passion for learning. Invest in continuous training and upskilling for your team members. Provide opportunities for learning new techniques, tools, and methodologies through workshops, online courses, and conferences. Encourage collaboration and knowledge sharing within the team, fostering a culture of continuous improvement.

I do want to add a startup’s mindset here along with the others’s mentality. The talent these days in the data world are very agile. And surprisingly, these are more entry-level talent. Yes, experience brings credibility, but early careers brings innovation, agility, faster delivery with lot more agility to learn and deliver. The more versatile your data team is, the more chances are it will run into less troubles in the long run.

The coding challenge is the best. It is the same way students are assessed and for a reason. Only good students get good results when subjected to a challenge, the same goes for employees. If you think you had enough exams in your life already and you don't want to subject yourself to a test that's a huge red flag.

Don't try to build a team from the outset. Delineate a small "low hanging fruit" within your data processing needs, and start with one person. Don't look for a "data scientist", but rather for a data generalist, which have previous experience with : a) data analysis using Python libraries b) implementing data pipelines in a cloud (AWS, etc.) c) backend/frontend prototyping tools (Streamlit, Dash, etc.) d) cost optimization techniques Look for a person who can communicate with the team and convince them to change their current routines. Provide her easy access to the top management and let others know you give a priority to her project. After the first project, aim for larger targets, you'll know when you need more people and their skills

I believe a lot in open science processes and management. Prior to my professional career I have been a contributor to Wikipedia and admin for nearly 20 years. In small teams, I don't believe too much in metrics and KPIs which quickly become objectives in themselves. Collaboration should flow naturally with transparent processes, objectives and shared information. The most important thing is to properly share and document all the key tools and programs and create enough redundancies. Beyond the team, open sharing also create multiple opportunities for external collaboration. I now coordinate the largest pedagogic community on LLMs in French and while it takes some time, it has helped a lot to be kept updated with the newest innovations.

I like to maximize members' chances of doing what they are currently most excited about - turning the "getting to know" members into something extremely valuable and a powerful tool to manage diverse talents and characteristics. I also invite them to mature their talks about the impact, such as by answering: - What is the impact of their current tasks/projects? - What is the impact of doing something differently than what is being done? - How do we measure that? I like to promote a culture of transparency, letting members comfortable talking about what is important to them while also balancing what is important for the company. A company will always benefit from teams being motivated, and people feeling respected is the way to do it.

In managing data science teams, adopting an Agile approach has proven transformative. Here's how we've incorporated its principles: - Value Driven: Our focus consistently remains on delivering maximum value, ensuring every project and task aligns with overarching objectives. - Explorative and Applied: The team is encouraged to explore innovative solutions, but always with an emphasis on real-world applicability. - Iterative and Incremental: Instead of waiting for a 'perfect' solution, we release work in stages, refining as we progress, leveraging feedback at every juncture. - Collaborative: Every member plays a vital role. Through open communication and collective decision-making, we ensure holistic insights and solutions.

Data teams are very agile and hence have the chance to get bogged down with a lot of work. While delivery is the utmost priority, having a systematic production line with clear distinctions and time-to-time learning of the skills in DS, prevents burnout, more interest in learning, and more diverse and versatile data teams.

Effective team management in data science involves more than task assignments; it requires creating a supportive environment for team members to thrive. Cultivate a culture of data-driven decision-making, experimentation, and continuous learning. Encourage open sharing of ideas, feedback, and challenges. Set clear and realistic expectations aligned with business objectives. Facilitate regular communication and collaboration within the team and with other departments. In small teams, emphasize transparent processes and shared information, focusing on collaboration rather than rigid metrics. Open sharing also fosters external collaboration opportunities, enhancing innovation and staying updated with the latest advancements.

Management is getting things done through other people. Many data science practitioners that turned team leaders focus too much on the technical delivery. While it's important, we often neglect the people aspect. When you manage people you have to understand the basic group dynamics. I usually refer to Tuckman's stages of group development. Based on my experience, any team goes through that: Forming, Storming, Norming, Performing, and Adjourning.

A DS manager must live by 3 principles: 1- results first. 2- time is the most valuable thing. 3- divide et impera. It's not about using the latest fanciest tech or the latest neural network model. It's about fitting a solution to a problem within a reasonable time window. Sizing the problem in terms of time is the most difficult for team managers. Another critical issue is the assignment or division of concerns so that the responsibilities of team members do not overlap, when overlap is unavoidable the manager must step in and decide at a finer level who is working on what. So in some cases, micromanaging is necessary to avoid wasting time.

Recruiting and training : recruiting agency and company needs to come up with a new age way of adapting to new received knowledge in this century. sometimes company lose the best skilled person just because he/she does not have the certificate or experience needed. the world need to find a way to monitor, assess, and provide a way to see the potential on the candidate. training for candidates is very important. the world ought to balance and reach for diversity we have as the community. this will make the team balance, technically, soft-skills and visionary.

Creating a successful data science team requires ongoing adaptation and optimization. Follow best practices, including defining the team's vision, involving members in project planning, and empowering them to make decisions. Provide necessary resources for efficient task execution, encourage continuous learning and exploration, and recognize achievements. Balance specialized expertise with versatility, and ensure team members promote data science efforts. Encourage a scientific mindset among higher executives to support research and recommendations for overall team success.

Like the law being the fabric holding a society, best practices are the threads that will make or break a project. The manager should choose the best-practices to follow and the team member or even better a team leader should enforce them.

Through my experience in establishing data science teams for various companies, I've discovered that striking the right balance between specialized experts and versatile professionals is essential. Data scientists should immerse themselves deeply in their specific areas of expertise while staying vigilant about rapid developments in literature, platforms, tools, benchmarking, and more. Additionally, having team members oriented towards sales and marketing is crucial for effectively promoting and positioning your efforts. Lastly, even if you don't have control over your higher executives' selection, it's imperative to evangelize and encourage them to embrace a scientific mindset, support your research, and heed your recommendations.

For startups, best practices are good to have when clearly defined — as long as they don't interfere with the agility of the team and the company. You must take size and objective of your team into account while defining them.

Increase complexity in steps by following the crawl, walk, run approach. Establish a collaborative and transparent culture. For example, promote public channel comms over 1to1 chats. Ensure life balance and help with career development on an individual level.

As I've experienced, making sure your data science team is part of the planning, scoping, and prioritization of projects is crucial. Let your data science team be a part of the discussions early and often if you're looking to derive value from this team. For business professionals it is sometimes difficult to foresee the needs your data science team will have on a project that involves them until you do just that, involve them! Rather than start a project and asking for the data and skills the data science team brings to the table in the midst of an already complex project, make sure they're engaged from the jump to prevent costly and cumbersome work on an already moving train.

Don't expect to be perfect right from the start. IT products evolve continuously - and so can teams. Take the chance to assess improvement potentials on a regular basis and work on them.

documents, documents, documents standardize and automated those template and store as enterprise template. Be willing to hear people and willing to share.

Companies that want to generate sustainable value over time MUST define clearly their data science processes and internal best practices. How to evaluate and prioritize business cases, defining internal coding best practices, define clear roles within the team, a clear data architecture, clear data processes, the tools that are allowed and the ones that aren't (and how to change them)....

Really two words: shared open information and continuous training. It's a fast changing world and with developments like gen AI, data science in the 2020s will look almost nothing like data science in the previous decade. A good dynamic in place for skill acquisition can get you really far.

The significance of agility and quality assurance (QA) within a Data Science team cannot be overstated. Building an effective team that can adapt and deliver reliable results in the dynamic field of data science is paramount. Agility enables quick adaptation to changing needs, promoting method evolution and rapid experimentation for insights. Quality assurance, on the other hand, is the bedrock of trust in the results produced by a data science team. QA helps in identifying and rectifying biases, errors, and inconsistencies that may otherwise lead to faulty conclusions. These principles, when combined, establish a strong foundation for success in our data-driven world.

Navigating the world of data science often reminds me of my own journey — full of twists, turns, and unexpected challenges. Just like in life, our team needs to embrace change, learn from each stumble, and always strive for better. And just as we value trust in our personal relationships, we put ethics at the heart of our work. Every step we take respects privacy and radiates integrity. When I think of building our data science team, I'm reminded that it's about understanding every member's unique strengths, cherishing them, and realizing that, together, we're creating something truly remarkable.

Building a team to work on a Data Science solution is totally dependent on the task at hand. At the onset, its important to: - Understand problem statement, devise & clearly list down the plan to build the solution - Identify level of effort & skill-sets required as part of the solution strategy Once this is done, then depending on the timelines and level of effort : - Identify depending on the complexity of the solution whether it'll entail specialists (like Data Engineers, Business Analysts) or generalists with varied skill-sets who can work on multiple components of the solution on their own Finally, once the team starts with the actual work, its equally important to have a feedback mechanism to monitor status & ensure progress

Consider how the data science team will integrate and support the entire organization. Will they be a resource to matrix out into far flung projects, a center of excellence to provide consultative support, or their own P&L center with their own objectives. Each organization will need their own approach, but I always encourage data science teams to get to the front lines of the business or customer to learn their challenges and opportunities first hand. Don't just take a dataset and try modeling it. Get to know the why and how so that you know you're solving a problem that someone cares about.

Consider the people you're hiring and the opportunities you can provide them short term and long term. Be honest with them. There may not be a good long term opportunity or maybe they are Abit junior but hungry and if they prove themselves the sky is the limit.

It is important to build a team with a good mix of skills such as front-end knowledge (Powerbi, Tableau etc.), data engineering skills, Python, Statistics, and AI/ML skills. More importantly it is important to build a culture of continuous learning in your team to cross skill and upskill your people for their own personal growth as well as to meet the problems of changing technology and business landscape.

Domain knowledge is essential. If your team acts as consultants across many domains, then your only hope is to be good at eliciting relevant domain knowledge from the customers. If you are tied to a specific domain, then recruit or train staff in the domain.

An aspect that is worth considering is the team size vs team management. For instance, what should be the average size of a data science team? One can say that it depends on the size of the project. However, no matter the size of the project, I think that it would make sense to modularise the project and create sub-teams such that no more than 5 people can work on (largely) independent parts of the larger project. In this case, the interface between sub-teams and the chain of communication and reporting should be clearly defined to achieve optimum results. As individual teams grow, management and communication overhead increases. A major example of this overhead is the time spent in meetings compared to the amount of work done.

Leadership is different than management. I like Peter Drucker's definition of leadership "The only definition of a leader is someone who has followers". Don't neglect people and be a role model. And even if it's a tough domain there are some good practices that always work. Throughout my career I've always stick to one simple rule with my colleagues: reward and let people grow.

It's important to understand the business question senario at hand to aid in data team building otherwise in most cases one merely requires a data analyst not a scientist.