What does an ML Manager do?
The multifaceted job of managing Machine Learning experts
If you’re building an ML product and hire more than a couple of scientists, you’ll quickly reach a point where they need a dedicated manager. Yet, experienced ML managers may be even harder to hire than ML scientists! This post talks about the responsibilities you can expect an ML manager to cover.
The core job of an ML manager is to keep the ML work on track, building machine learning technology which impacts the company’s products. To do this though, they may be working within different organisational structures. In some companies, an ML team with a dedicated manager owns specific models and pipelines. In others, ML scientists are embedded within cross-functional teams rather than on a single ML team.
No matter how ML scientists are organised into teams, the main ways an ML manager operates are through people management, project or product management, and technical leadership. Alongside these, they’ll keep an eye on the overall health of their team, and act as a bridge for communication with other functions in the company. When defining an ML manager’s role, it’s worth noting that there are a lot of similarities with other technical management roles. So, leaving aside the specific technical details of what the team does, there’s a lot to be learnt from how other management roles are defined.
Influence Company Direction
An ML manager should have their own view of the technology landscape and how the company’s technology should evolve in response. Their view should be informed by their own knowledge, by team members, and by others they work with like advisors and collaborators. This perspective feeds directly into setting the company direction, and from that the ML team goals.
With the exception of a handful of companies, ML is new to many organisations. There’s a strong desire to use ML technology across many sectors, but very few people have the familiarity to know where to begin, how a typical ML project evolves, and what successful implementation looks like. Hence when defining company direction, ML managers find themselves both educating others and advocating for the right choices to create conditions where the ML work is successful.
Once direction is clear, a manager needs to keep up-to-date with what other teams in the company are doing, and represent machine learning interests in company-wide discussions. They ensure that the work of ML scientists fits into a wider strategy, that plans between different parts of the company aren’t diverging, and different parts of the company are aligned on cross-team work.
People Management
People management is a key part of any manager’s role and an ML manager will be line managing a variety of ML experts. There are lots of great resources out there about how to manage people — if you’re a new manager then go and find them! This post isn’t meant to replace those, but rather to highlight the parts of the role which are key to managing ML teams.
ML experts tend to be highly educated and motivated, which can make for a rewarding partnership. Understanding a team, their goals & ambitions, and what motivates them, will be the best way to work together. With this information, a good ML manager can match people on a team to the work that needs doing, and identify both stretch opportunities and places where there’s a gap in skills.
Growing a team relies on being able to give people feedback — both good and bad — usually in 1:1s and performance reviews. Alongside this is coaching team members through tricky situations, navigating uncertainty, and heading off conflicts and other situations as they arise. The rapid expansion of ML as a field has led to a skew towards recently graduated employees, with experienced folk being few and far between. In many organisations, there are no senior ML scientists to look up to. Hence, ML managers have to make sense of the career path and help ML scientists learn to navigate what, for many, is their first job in industry.
Team health
Beyond managing individuals on a team, the health of the team as a whole is the responsibility of the ML manager. This usually means gaining a sense of the various issues that people are facing and proactively moving to fix the ones which are problematic. If multiple people are bringing up an issue, it might be a sign that it’s something worth addressing.
It also means making sure the team is resilient to change. If for example only one person understands how the model training pipeline works, then it becomes a headache if that person decides to move to a new role. Identifying risky situations and mitigating them early is important.
Hiring and Onboarding
Hiring new people to the team and onboarding them is a part of building a healthy team. But it can become a very large part of the ML manager role, especially if the company is growing. ML roles are becoming more common in industry, but they still aren’t crisply defined and can require many different skills. Knowing which skills you’re hiring for and what your team lacks is the starting point for hiring. Add to this tweaking of job descriptions, interviewing, headhunting, information chats with potential candidates, and negotiating offers, all means that hiring can take significant time.
Project or Product Management
Project management and Product management are two different disciplines, but ML Managers have to do a little bit of both. For this kind of work, managers are often thinking weeks (sometimes months!) ahead of individual team members so that work progresses smoothly.
On the project management side, they’ll be responsible for the output of their team — planning the roadmap, balancing competing priorities, identifying risks and surfacing issues before they become a problem. They’ll need to know what people on the team are working on and the obstacles in their way. When unexpected tasks crop up, they’ll need to quickly understand the issue at hand and plan the work to address them.
On the product management side, they’ll need to fully understand how different teams in the organisation are working together to build a product. Together with customer feedback about the company’s product, they have a sense for how ML can and should be used to improve the company’s offering. Whether working with a dedicated product manager, or taking on the role themselves, this is a crucial skill to keep a company on track & using machine learning effectively.
Technical Leadership
The majority of ML Managers have previously been in ML science roles, and so technical leadership is perhaps the most comfortable area of their job. One of the things to adapt to is leading by influence, rather than directly telling the team what to do. As a manager, the job is not to know the right answers, but to bring the right people together to make good decisions.
There is sometimes a fine line to be trodden. Managers are usually experienced ML experts, and often have relevant insight from their past work. They may have seen similar situations, and have a good sense of the right path to take. This experience is crucial in an organisation where there are few senior ML scientists to review work and give technical feedback. Still, knowing how to use their expertise to guide the team rather than to solve all the problems themselves is a skill that’s learnt over time.
Managers may still do technical work like write code, do code reviews, train models etc., but usually only if the team is small. With any team more than a handful of people, management becomes a full-time job. Still, the team’s manager needs to be sure that work is done to a high standard, which means ensuring the right processes are in place and are being followed.
A manager’s tools for doing all of this are 1–1s with team members and with other key people in the company, planning meetings, team standups and retrospectives, company all-hands & other in-person discussions. Outside of meetings, their job can include tasks like reviewing code, experiments and documentation, writing roadmaps & goals, or vision documents.
ML Managers are a key hire in your company to help an ML team perform as the company grows. As with all ML jobs, the reality of the role can vary from place to place and the day-to-day depends very much on who else is part of the team.
Let me know how this ties in with your experience, and if you think there’s anything missing!
Thanks very much to Neal Lathia for the idea behind this post, and for reviewing a draft version.
I work with companies building AI technology. Get in touch to explore how we could work together.
