The idea of autonomous cars navigating the roadway may have seemed far-fetched just a few years ago, yet right now, engineers at Google are using machine learning to get past limitations in self-driving technology. What’s more, as they open up about their research and share their findings with the broader community, they’re attracting top talent in the field.
The truth is, most organizations today are talent constrained. We all know we could do much more if we could just find the right people. But where does the talent come from, and, more importantly, how do we assemble a team that’s capable of working on tomorrow’s problems today?
We’re in the middle of a revolution, where software can be trained, not merely coded. If your team spends all of its time on traditional engineering, but neglects to use the data you have to benefit your customers, then you’re failing. It is not just an opportunity but an obligation for most tech organizations: build a high-performing machine-learning team.
Assembling the right team with the skills, knowledge and courage to tackle the impossible problems in your business can be challenging, but it’s also a whole lot of fun. Where to begin?
Find practitioners working in the field or look to academia.
Finding machine-learning experts who have practical experience working in the field might seem hard — and it is. But it’s something you just have to do. One obvious starting point is looking in established organizations that already have machine-learning teams effectively competing with one another (there are many excellent organizations out there, e.g., Facebook, Salesforce, Google, Apple). But that’s only part of it. If you find them, they still need to want to join your team.
There are also people in academia or those who are transitioning in their careers — mathematicians, physicists and data scientists. While it may be easier to find talent in this group, people in transition or freshly minted Ph.D.s often don’t have practical experience.
In Adobe Document Cloud, we like candidates who have published papers, not just because it gives us something concrete to prepare for in the interview (editor’s note: the author works at Adobe Document Cloud). It also helps us understand how someone thinks and the rigor they bring to specific problems. Nevertheless, the best candidates coming out of academia are those with practical work experience and a publishing record. There’s a slender pipeline of people like this — it is an accurate cliché to call them unicorns. Unicorns that do exist.
Nurture your existing team.
Engineers are not stupid. They know that machine learning is a hot area, commanding fantastic compensation and working on fascinating problems. Within your organization, some of these engineers are taking the time to learn these skills on their own or with their managers’ encouragement. Find these people and reward them!
One benefit of looking to your own tech staff for machine-learning talent is that these people are self-selected. In many cases, they’re taking classes at Stanford, Harvard or Coursera — or they’re teaching themselves. Most of these people are not yet practitioners, and they don’t have published papers. But they do have passion, and they know your business. That’s critical.
If you can add one or two unicorns to an existing team of people with passion who are developing their machine-learning skills, that’s the beginning of a decent machine-learning team. The composition of your group may not be 100 percent machine learning — it’s more likely to be 30–70, science versus engineering. But, much of the work the team actually does involves getting things to work, to line up and exist in a scalable way.
Build the culture for a machine-learning team by thinking in terms of problems that can be solved with data.
Machine-learning expertise solves complicated, seemingly intractable problems. But don’t miss the opportunity to solve problems with the talent you already have. There are plenty of machine-learning problems that entry-level people can solve. There’s no such thing as a machine-learning product manager. Instead, every product manager and every executive needs to ask their team, “As you look at our customers, how can you use data to improve the product?”
An innovative, open culture needs to be in place for a high-performance, machine-learning team to thrive. In fact, if you have the right team, but you’re not working on imaginative problems, your people will quit. They need to be able to talk about their work a year from now and publish at top-tier conferences. At Adobe, we encourage publishing our results. We don’t view the work we do at that level as proprietary, and we’re happy to contribute it. Publishing papers is how you build a reputation. If we publish, we become an attractive place to work and a magnet for talent.
Train others and put skills to use. Learn, practice, test and repeat.
When you find team members on the journey, even those who didn’t self-select, they need to learn machine-learning skills, too. Broaden the base by green-lighting training for anyone and everyone. In many ways, this is a change for managers and executives.
The challenge is that once your team learns machine-learning tactics, they often can’t apply them because there is no outlet in their product road map. People come to me all the time and say, “I’m ready to work on this, I took the course.” Unfortunately, it’s not always possible, because that’s not what the road map is asking for.
People today really want to work in this space. And it’s critical that the machine-learning capability of the organization is decentralized and distributed throughout the entire organization.
The process of building a high-performance, machine-learning team can be difficult, but it’s also really fun. It’s a holistic process, with designers, product managers and executives all playing a role. Marketers, managers and others need to think about the problems to be solved. Take your pre-existing staff who need to learn machine learning, offer them courses or sessions inside the company where they can learn new skills, and make it easy to put them into practice.
David Parmenter is director of data and engineering at Adobe Document Cloud. To comment, email firstname.lastname@example.org.