Webinar discussion: Collaboration Across an Organization - What contributes to data science teams that collaborate well with non-data stakeholders?

Continuing the discussion from the Building Effective Data Science Teams Webinar:

Collaboration Across an Organization

What contributes to data science teams that collaborate well with non-data stakeholders?

Our panelists for this webinar were:

  • Kobi Abayomi, Senior VP of Data Science at Warner Music Group
  • Gregory Berg, VP of Data Science at Caliber Home Loans
  • Elaine McVey, VP of Data Science at The Looma Project
  • Jacqueline Nolis, Head of Data Science at Saturn Cloud
  • Nasir Uddin, Director of Strategy & Inspirational Analytics at T-Mobile
  • Moderated by Julia Silge, Software Engineer at RStudio, PBC

Some discussion in the webinar:

Greg : I think part of that is just the willingness to communicate. Sometimes, people want to work in a box or in isolation, so breaking that mindset and saying no, we need to collaborate with each other. We’re all on the same team. We want to provide value for the company. We want the company to be successful.

Shift the mindset to all being on the same team, working together for this as opposed to “let me go do something and bring it back and see if it works.” The latter comes across as something that’s being done to the business rather than something they’re a part of so the adoption and use would suffer. This goes into the value as well. If they’re not using it, then there’s obviously no value.

Again, it’s also about being open to change. Just because we’ve always done it this way doesn’t mean we have to. So some of those same topics are relevant here as well.

Kobi : I’ve been at media companies not tech companies for my last couple of jobs. Collaboration is a huge piece of being able to understand the way that the business measures success. I’m listening to all the things that everybody else said on the panel. When you first asked that question, Greg said it exactly, communication.

Also, product management. When I first started, coming from academia, I was less of a devotee to product management but I found it extremely useful to have someone whose job is to figure out how this thing looks, how people will react with it, push that back onto the data science team, all while having having that role separate from the people who are going to use it. People will get frustrated if they don’t feel that they have the same language that you do. I think what I found more often than not, is people in frustration will just get quiet rather than push back so that you can refine and change it. Carving out a role on the team for somebody whose job is just to do that, the ingress and egress to the client side, is super important.

Julia : That’s an interesting point because I think a lot of us have probably heard about self-service data and democratizing data in our organizations to make it so everyone can query the database and everyone can get their own stuff, but then have also experienced frustration as you just said. Like, “oh no, people are finding some kind of trend that I don’t think is real.” Maybe you have experienced frustrations with giving people mass access to data or the democratization of data. I’m wondering if part of it is the lack of applying the lessons of product management to those self-service data internally and not thinking like a product manager as you make these things available internally?

Kobi : Let me give you an exact example without divulging any intellectual property. Optimization. The linear programmers and operations research want to solve for the most optimal outcome. At one job, one of the things we were optimizing over was the number of views of something over a certain period of time and arranging things to make that most optimal. Turns out in some settings, like ad sales, that’s not actually what you want. You’re dealing with inventory that has different costs, different rates of return, and it’s a nuanced thing. There’s a bit of an art to it. The first go round, I’m telling a story in developing a model where we know we’re hitting this at 100% and causing a bunch of frustration elsewhere in the organization because it made it more difficult for them to do other things that they needed to do.

It’s those sorts of conversations that you need to have - again going back to communication, communication, and communication. This is a relatively new field, we do something which is relatively technical, and society in general is relatively innumerate. The lift here is important in being able to converse with what’s going on and the people who are making the money. It’s super important because the downside or the negative outcome is the silence that can happen between a data science team and the rest of the organization. I worry about that at any company, in any data science team, and as I think about data science as a field.

I remember when places were getting rid of their statistics departments, when you would see an econometrics department and a psychometrics department, but there wouldn’t be a stats department. I remember trying to think of different names, being at society meetings and thinking, “Should we call it analytics? Should we call it data science?” We enjoy a time of popularity in the zeitgeist right now and I think it’s important for us, as we’re instantiating ourselves in this business, to do it in the best way. Communication is super important and to have that language back and forth about the technical things we create and the things that are meaningful, especially at places that aren’t tech companies.

Jacqueline : I think one thing that came up was this idea that people are silent when they should be talking. This is something I’ve noticed leading a data science team. I have tried to teach my team about going and telling people stuff, even if it’s kind of scary.

To another point that was made, if someone says something you don’t agree with even if they’re more senior than you, you have the right to push back on it. If you don’t feel comfortable pushing back, that is the job of your manager or the data science leader. I think it’s certainly the case that communication is key. But to the people who are not yet managers or directors, communication can be really difficult. I think a lot of our job is to figure out how to actually get people to do the communication that you know the team needs, but is not necessarily obvious. I really want to stress that mentoring on communication is so much of the job of a data science leader and I feel very passionate about that.

Elaine : To emphasize that, I remember someone on my team who had done this great work and it had kind of gotten brushed aside because people didn’t really understand how it fit. I told him, if you spend three times as much time continuing to communicate this until people understand as you did doing the work, it will be worth it. I don’t think he was excited to hear that, but we need to shift our thinking to making sure that people understand the work that we do. It really doesn’t matter how great it is, the communication is even more a legitimate part of the job. People need to think of that as a valuable way to spend their time, especially when they’re coming out of school and new to a team.