TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

Collaboration and Freedom Key to Data Science Solutions

We continue our interview with the head of an unusual and highly successful data scientist team at Cisco, Robert J. Lake. His approach to managing his team focuses on allowing people to learn and take risks.

• By Linda L. Briggs
• April 27, 2016

"A lot of people focus on technical skills. I'm really much more interested in soft skills," says Cisco's Robert J. Lake, who uses some unusual approaches in managing his diverse and highly successful team of data scientists. In the first part of this two-part interview, he described the three key skills he looks for when hiring -- a passion to learn, a willingness to take risks, and an insatiable desire to solve problems.

Also, in an industry often dominated by men, Lake's team is half female. He doesn't purposefully hire that way, he says, but "when you look for the right people, it's amazing how diverse they happen to be."

Upside: It's interesting that when you listed the skills requirements for your data scientist team, you didn't include knowledge of software coding. That's not a required skill?

Robert J. Lake: No. To be honest, I've changed how I train people on the team. One of the things we do is every fourth Friday we have what we call a hackathon. It's a whole day without exception; no one is allowed to miss it because of meetings. This is a day when I give the group a problem to solve as a team. They can work in small teams, they can work in big teams, or they can work individually. They just have to come up with a solution.

What I've found is that the more visual tools we give them to play with, the faster they are able to solve the problem. We got them away from coding, we got them away from writing pure SQL, we got them using tools such as Statistica and Pentaho where you drag and drop nodes together. The tool does the initial work for you and sets you up.

The more we did this, the more I realized that we can move faster if we don't go the traditional route. There's nothing wrong with writing Python code when it needs to be done, but a lot of stuff now is literally drag, drop, and go.

If you train people that way, it gives them the foundation and it gives them the confidence to be able to do these things. So we teach people how to do data mining using these tools, how to bring things together in a scientific way. Once they have some practice, we start teaching them the theory behind some of those pieces.

I can take someone with no technical background and train them up. Some of these folks will be really good and some may not be as good. It depends on where their passion is. That's the second ingredient. They have to be passionate about it.

For lots of people, the appeal of the job is not about the coding. It's that they want to solve problems and do something that's meaningful to other people. That's what we do with data mining. We teach them that it could be applied to many things -- it could be cancer, it could be diabetes, it could be anything. It's whatever ignites their imagination.

They realize they're doing something they didn't think they could do, and they grow in confidence and become stronger people. I love data mining, it's my passion, but if you don't, if there's something else you want to do, I won't be insulted.

That's what I tell my interns: "Look, you're going to work really hard. I'm going to teach you a lot about data mining, but there are two things I want you to keep in mind. One, I want you to enjoy your time here. Two, I want you to learn if this is the discipline you want to work in. Part of my job is to help you understand whether this is the right career for you."

That's an interesting approach. It really redefines who can be a data scientist and what skills they need.

Absolutely. I'm not saying that you don't need team members with master's degrees, or with a Ph.D. I have a master's and that's part of my background. You still have to have strong people in lots of areas.

If we look at the past, all employees used to have to have two major skills. One was subject matter expertise -- they were a machinist, or a typist, or a vehicle mechanic, or whatever. The second skill everyone needed was some sort of communication skill so that they could connect with other people. That's been the tradition in the workplace for a very long time.

We're now in a disruptive period. Everyone is going to need a third skill, and that skill is the ability to absorb and understand some form of data or information, and know how to use that data to make a decision. There is so much going on, I really do believe that every person in the world needs to understand data, and be able to make decisions from data. Not to the level that I understand it, or some people understand it, but enough to be able to make decisions from it.

We make decisions anyway all day long. We need to build that skill set up, because the constant presence of data today is faster, and more ferocious, and more intense than it's ever been before.

So that will have to be the third field that all employees have. I often think: how do we help people be more passionate about this in the real world? That's going to be a way to improve the workforce of the future.

How do you set up your team of data scientists? What's the structure? Do you typically have a manager?

It's all about collaboration. I have a technical lead who is my most senior data scientist, who is going to do the technical review. I have my senior team members organize and assign a project. Even some of the juniors can assign a project, depending on the level of skill.

They are then responsible for project management. What happens is that the rest of the team can opt in and opt out to help out with the project. Because we use the agile approach of project management, they write small stories of what needs to be done, and then people pick the ones they want to work on. I do that because I want people to volunteer.

One, I'm watching to see how people work. We don't want people always doing the same thing; we want people trying new things out and taking a risk. Sometimes this means that someone says, "I'm going to take this story", and someone else says, "Oh, I want to understand how to do that. Can I shadow you?" So it promotes self-learning.

Also, having the team members select their own work means that I'm not picking and choosing and showing favor -- all the things you get when you assign people to projects. We avoid all of those dynamics.

From a manager's point of view, I get a lot of information about how my team is operating with this structure. I can make adjustments and give coaching and guidance to all the team members, encouraging them to take risks and try different things.

People say, "That sounds great, but when it gets chaotic, what happens?" Well, it actually rights itself. It's amazing how people actually hold each other accountable and how they work together. We increase collaboration across everything. I haven't seen yet -- we've been doing this for over a year now -- the same project leaders having the same people working with them. I see people mixing it up because they can, and they want to.

What does the team size tend to be?

It can be anywhere from one person up to the whole team, which is currently 14 people. In fact, we technically can go beyond that because we have about 12 people who are doing construction projects with us right now. They are [from elsewhere in the organization] and are being mentored by us. They are working with us on our projects to get some side experience.

We can have very large teams or very small teams. What's interesting is that if you watch them, the team grows and it shrinks, again and again, according to the work. It's not as if I assign people and have them sitting around saying, "OK, what's next?" They just naturally handle that themselves.

You call your team the predictive analytics decision solution team. Why not use the term data scientist?

A lot of people don't really understand what a data scientist is. It's really a decision scientist. At the end of the day, whether you're doing advanced analytics or data science, all we're trying to do is help someone make a decision. We included predictive analytics because we are working in the predictive area.

I wanted a team name that really drove and anchored people to what we do, not "Oh, that's just a hot, new, sexy term and you won't be around for long." No, I'm anchoring this on decision solutions.

It doesn't matter whether I use data science, basic decision-making, or some simple logic and add a little bit of common sense to help someone make a decision. Sometimes we simply talk through things with someone, and from doing that, they decide, "You know what? I don't really need anything more. I'm going to do this because I just realized it makes sense."

We go from that all the way up to full-blown analytics. That's our job. The name of the team is a reminder to stakeholders, but mostly to ourselves, that our job isn't just to run through a bunch of algorithms -- it's to help people make decisions.