What does a data science team need to thrive?

(upbeat music) - [Interviewer] So what does a data science team need to thrive? Let's say I put a team into place. What tools or resources will they need? - Well, the first thing that a data science team needs is actually they need to be part of the team. The broader team. Most people try to say, "Hey look, I've got a "data science team," and they just put them in. And then that team gets what you might call is organ rejection, where the antibodies attack it.

And so what you need them to have is you have to create pathways for them to actually be part of every team, and be able to say, "Hey, how can I add value here?" "What's another way we might be able to work?" A data science team without access to data is not much of a team. (DJ laughing) And it's surprising how often a data science team is brought into a company, or people take incredible efforts to build a team, and they don't have access to any data. So once that has happened, and the team has access to the data, they have to be able to clean the data.

It's still, to this day, that it's about 80% of the work that a data scientist is doing, just cleaning data. Then, once they've cleaned the data, how can they actually have context around the problem to work with that data? What other tools, do they need more of a Hadoop style tool or can they do it with smaller data? Or maybe it's they need to use technologies like Kafka to stream and move around the data very quickly. There's so much technology out there. That's part of the beauty of the, it's less about the questions of what technology they have, but rather the problem that I often refer to as the toolbox problem.

And what happens is a company or an organization typically says, well, you get to have one tool. When you call your plumber up, your plumber doesn't say, "Well, I'm going to bring one tool over." They bring a whole toolbox. The same way the data scientists and technology teams need a toolbox, a suite of tools. And so how do you allow them to have that suite of tools to work, and operate, and build and try and test things? And then, how do you, once you test it, how do you learn collectively? And so, what I'm arguing for is a much more ambiguous data science team approach where the team is actually bled into all portions of the operational structure.

And seeing where, you know, good things can happen. And that basis of that first starts with democratization of data, good tooling and technology, deep curiosity to actually figure out what are the insights, and then creativity to turn that into new ways of operating within the organization. (upbeat music)