How DonorsChoose.org Grew a Data Scientist (and an Army of Data Masters)

Whenever I talk about the data culture at DonorsChoose.org, the questions I most often receive are about democratizing data. Yes, data is everywhere at DonorsChoose.org. It’s used on every team and every project and there’s a real hunger to use it even more. But how did that happen?

The answer to this questions doesn’t lay in just one place; it has required a huge investment by our leadership team, a robust data infrastructure, and an eagerness among our colleagues to learn. But there was one initiative that, maybe more than any other, made this happen: Data Masters

Goals

Data Masters was a 3-month, 6-part data bootcamp that kicked off in spring of 2016. The program was developed with three goals in mind:

1. Let’s get people across teams thinking like data analysts
2. Let’s get more people coding
3. Let’s give people an in-depth understanding of how our data is structured and how our business intelligence tool Looker works

But the long con? Let’s get rid of the data science bottleneck. Let’s plant a “data master” on each team that would help answer their colleagues’ burning data questions, without having to involve a data scientist in every little request.

Open to anyone across the organization, this initiative was backed by leadership and was viewed as an incredible professional development opportunity by both staff and managers.

When the program was first announced, I was on our operations team. I had a growing love of data and had dabbled with our analytics tools for years, but was frustrated by the roadblocks I’d run into and the long line to get time with our in-house data science team. So, I eagerly submitting my application, along with 14 of my colleagues who would be accepted into the program.

Curriculum

Here’s how the 3-month course broke down, (if you’re interested in more of the specifics, we’ve made the course deck public):

Pre-Work

• Read through Looker documentation on general frontend use (1–2 hour assignment)
• Read through LookML documentation and begin to understand the concepts of data modeling in Looker (1–2 hours)
• Code School SQL Tutorial (1–2 hours)

Month 1 — Asking Questions (2 bootcamps; 6 hours total)

• Intro to databases
• Intro to SQL tools + coding in SQL
• Intro to Git
• Thinking like a data pro: state hypotheses, split complex questions into steps, challenge assumptions

Month 2 — Answering Questions (2 bootcamps; 6 hours total)

• Deeper dive into Looker frontend
• Deep dive into Looker backend, including learning LookML
• Deep dive into Looker Dashboards, documentation, data visualization and calculations
• Continuing to practice SQL and working in Git
• Choosing a final project

Month 3 — Analytical process (2 bootcamps; 6 hours total)

• Forming hypotheses
• Designing experiments
• Sampling techniques
• Statistical methods to analyze results
• Continuing to practice material covered in first two months
• Wrapping up final projects

Final projects & presentations

Results

We now have a group of colleagues across every team in the organization who intimately know our data and how to wield its power. And over a dozen graduates have contributed to our Looker Git repository; folks create new fields in Looker that are helpful for them and their teams. This means the two-person data science team isn’t a bottleneck in running analyses and managing and growing a robust business intelligence tool.

And for me personally, this opportunity was a launching pad. I’ve switched careers and now am a data scientist full time. Ultimately, DonorsChoose.org ended up with homegrown data masters.

Replicating Data Masters at Your Organization

If you’re at the point where a Data Masters-esque program would be a good fit for your organization, here are some things we’d recommend doing:

• Get buy-in from leadership and managers. This is a non-negotiable pre-requisite; these types of programs require a lot of time and effort on the part of everyone involved and it’s just not going to work if your leadership team doesn’t think it’s worth the investment. If you’re running the course, on top of the 18 hours of class time, expect to spend at least 8 hours preparing, another 8 supporting the cohort outside of class.
• Keep the group small. We found 15 to be a good class size; it was small enough that everyone could get one-on-one tutoring during office hours, but large enough for there to be meaningful peer teaching and exchanges.
• Assign pre-work. As a participant, I found this kept me excited and engaged in the lag-time between getting accepted into the program and the actual kick-off. It also got a lot of the basics out of the way so that when we met in person, we really used the most of the time we had together.
• Assign (optional) homework. I was so eager to continue honing the skills we were learning every month, especially since I didn’t have as many opportunities through my day-to-day work to practice. But I also had a full-time job. Making homework available but not required allowed me to get all of the benefits with none of the stress.
• Do capstone projects. This gave me focus and a tangible way to apply what I was learning. It also forced me to learn to talk about data and analytics to a non-technical audience, and gave me confidence in these new skills I had acquired.

Ongoing Work

While I’ve zoomed in on one initiative to democratize data at DonorsChoose.org, the important thing to note is that this work is never done.

We’re continuing to democratize data in three main ways:

1. Fostering a love of data in individuals who show interest. We work one-on-one with folks who are especially eager to learn SQL and LookML, teaching them some basics and talking them through building their first fields in Looker. We encourage Data Masters graduates to keep up their skills and support their teams by tagging them in to answer their colleagues questions and tackle small data requests their teammates submit.
2. Scaling our training efforts. We’ve created a series of video trainings that amount to a self-service onboarding tool to our data and to Looker.
3. Playing a consulting role. We view a large part of our role as empowering others in the organization to make decisions informed by data. We make ourselves available to talk through business problems and find data-driven solutions.

Data democratization isn’t easy, but it pays off in ways large and small. How are you tackling this at your organization? Leave us a note in the comments below!