Mike Zhou is a recognized leader in artificial intelligence and machine learning, known for his expertise in developing intelligent systems that drive business success. With a strong foundation in data science, engineering, and mathematics, Mike has led transformative AI initiatives, including creating the most advanced customer-facing AI solution in the accounts receivables management industry. His leadership philosophy centers on assembling high-caliber data science teams, fostering collaboration between technical and business stakeholders, and ensuring that AI-driven insights translate into real-world impact.
In this TechBullion interview, Mike shares his approach to building and scaling data science teams, balancing technical excellence with business acumen, and creating a culture of experimentation and innovation. From the key qualities he looks for in top talent to strategies for aligning AI projects with broader company goals, Mike offers valuable insights for leaders navigating the evolving landscape of data-driven decision-making.
What key qualities do you look for when assembling a data science team, and how do you balance technical expertise with business acumen?
Great question to kick this off! In addition to fulfilling the criteria for job descriptions, I try to recruit team members who have as many of the following attributes as possible:
Intellectual Horsepower:
- With the introduction of AI solutions, it’s now even more critical to have team members who can think through and solve problems intellectually.
- It’s pretty easy for someone to be educated with data science knowledge. Still, it’s difficult to train to be a high performer and solve a problem in an unfamiliar space with elegance.
Relentless Curiosity:
- I look for candidates with an innate curiosity about their domain, discipline, and adjacency. In my experience, I’ve found a high correlation between this attribute and the outcome of producing outstanding results through a novel method or approach. This is particularly relevant for data science, given the research-focused nature of the discipline.
Story Telling Ability:
- This capability does not necessarily need to be present in all team members, but it should undoubtedly be present amongst various team members and line managers. The ability to sell data science and explain complex concepts to an external audience is a highly valuable skill set. In addition to greater understanding and recognition of the work being done, it also improves collaboration and results.
How do you foster collaboration between data scientists, engineers, and business stakeholders to ensure data-driven insights lead to tangible impact?
In different roles I’ve had throughout my career, I’ve operated in a manner where data scientists operated independently and alongside other disciplines in a squad or tribe. The best structure is very context-dependent.
The mutual understanding of objectives should be aligned at the leadership and management levels as well as the individual contributor level. I have also seen how effective the alignment of non-monetary and monetary incentives can be in ensuring that commercial and technical staff are aligned in striving for the best outcome for their organization. Depending on the extent to which technical staff can affect the commercial outcomes, it makes sense to consider ensuring their incentive structures are similar to commercial stakeholders.
In addition, it’s essential to encourage cross-functional engagement, and team members from different disciplines consider themselves as one team with a common objective.
What strategies have you found most effective in managing and scaling data science teams while maintaining high performance?
Processes can often be overlooked, and it’s a more familiar sight for technical development teams to forego process setup and documentation while deep in build mode. However, ensuring the right processes are set up is essential before hiring more team members rapidly. I have seen small teams operate well but become less efficient as they grow.
It’s also essential to be balanced with how much time is invested into setting up infrastructure and processes. However, it generally makes sense to onboard new team members after relevant processes have been set up so they can operate productively without needing the esoteric knowledge of a few initial team members. This resultant knowledge gap and differences in working styles often hold back scaled teams from reaching their full potential. Instituting processes that new team members are requested to follow will help ease them into the operating rhythm of the team. The leadership structure is also key.
This structure is critical to ensuring the successful scaling of a high-performing team. I opt to hire line managers who have deep expertise in the craft. Having managers who act more as professional project coordinators who do not possess the ability or experience to help the team execute is likely to be less effective than technical individuals experienced in innovating. Oftentimes, I groom internal candidates for management positions wherever possible.
How do you ensure your data science team remains agile and adaptable in response to evolving business needs and technological advancements?
The data science team needs to be capable of rapid deployment of models and features without assistance from non-dedicated resources outside of the team. Engineering teams often have competing priorities. This will increase the velocity of data science models and features being shipped. In many cases, there either needs to be a platform that allows for rapid deployment of data science models and processes or individuals in the team with the capability dedicated to implementation.
Due to the nature of the work in my current place of employment, I have intentionally built a team that possesses both model development and ML Ops skillsets. Some team members are also experienced in software application development. It may take longer to recruit for this profile, but it certainly helps ensure that the team remains agile.
In the context of data science projects, teams can better service business needs if they are more effective at planning. The reason for this is that with every project, we are faced with a decision on how far we progress on the theoretical curve of diminishing marginal returns from time spent. For instance, spending 200 hours on a new problem space within an organization will likely yield a more tangible result than the same time spent on optimizing for a metric that has already undergone multiple modeling attempts.
This is unique to research-oriented pursuits. A well-planned roadmap considering ROI on time spent will be much more effective in executing on a balanced approach to achieving and meeting evolving business needs.
Culture of Experimentation:
- Some of the best data science processes and models that I have developed or overseen have resulted from a particular experiment that occurred from a thinking process that originated from a spark of curiosity, which oftentimes may not be blatantly obvious.
- Developing a culture of experimentation and testing within the data science team is essential. Leaders must ensure that team members are encouraged and comfortable to experiment. This approach has a cost, but there is also a substantial reward when done right.
What role does mentorship and professional development play in your leadership approach, and how do you help team members grow?
Effective professional development requires support from leaders and commitment from team members. There are two focuses I always aim to institute.
Mentorship at All Levels: In the teams that I have built, I have intentionally instituted and encouraged mentoring to occur not just from direct line managers but also from tenured colleagues. This ensures that team members are learning about their current roles and their future careers.
Structured Learning: Within the data science discipline, team members need to spend time experimenting and understanding their craft. Team members need to be able to spend time on self-learning. This may include self-paced online courses or a project that allows the team member to experiment with a new technology. Furthermore, knowledge-sharing sessions on how a method or technology has been applied for a particular internal project is essential to fostering a culture of synchronous peer-to-peer learning.
How do you align data science projects with broader company goals to ensure they contribute measurable value?
In a commercial context where delivering shareholder value is key, I always try to measure ROI on time spent. Even though a material component of the work done in data science teams can be classified as research, and thus ROI may be difficult to quantify, it’s essential to measure outcomes and contextualize those outcomes with the inherent variability of the craft. This can be tailored for different audiences within the organization.
Doing this well will build the reputation of the data science team. It will help leadership groups make objective decisions regarding investment in data science concerning company strategy. This circles back to why I have the contrarian opinion that storytelling is an important skill to possess in a data science team.
Can you share a specific example of a project where your leadership helped drive a major success in data science implementation?
At my current company, I was hired to build out the data science team and core infrastructure. The team started with me. I eventually grew the team to the point where there were 10 data scientists and another 25 data analytics and data engineering professionals, following the expansion of the remit beyond data science.
From the outset, I have been very deliberate in recruiting for the key qualities I look for: intellectual horsepower, relentless curiosity and storytelling ability.I have seen these qualities play out to drive success in multiple ways. Firstly, with the culture and guidance I’ve put in place within the team, I’ve noticed that like-minded individuals with similar qualities tend to over-exhibit their qualities, resulting in an impact that compounds whereby the team members bouncing off and inspiring each other. Secondly, attracting similar talent to join us on our journey became more effortless. During the interview process, I often had the candidate meet multiple existing team members; many commented to me at a later stage that the team was the main reason they accepted our offer over others.
As a result, we had incredibly long tenures with several team members and delivered several outstanding data science features and products. We were one of the first teams in our industry to develop and ship an AI-native solution, and we received praise from several of our clients.
What advice would you give to other leaders looking to build and lead a high-performing data science team?
Key tenets of advice that I have found incredibly useful throughout my career:
- Hire leaders in your organization that reflect the team that you want to build.
- Empower your team to experiment, to learn, and to collaborate.
- Ensure that your team (throughout all levels) understands the impact that they are making. Progressively increase their ownership of the outcome, where warranted.
- High performers are hard to keep, and underperformers are hard to move on. Ensure the right non-monetary and monetary incentives are in place for the right people to be motivated to be part of the journey.
