Data is at the very heart of how we understand and improve the Discord experience of our users all across the world! In this blog post, we’ll discuss how Data Science contributes to Discord’s mission: how we function, what makes us unique, and what’s coming next! (Plus, we’re hiring!)
The mission of the Data Science team is simple: we want to improve decision-making and inform strategy through the analysis, modeling, visualization, and distribution of information.
Easy enough, right? So how do we do this?
Data Science is an overloaded term in tech. Here at Discord, Data Science combines the full spectrum of analytics, machine learning, and causal inference into one team. This cross-disciplinary structure lets us leverage the right combination of skills and knowledge to overcome the challenges that we and our partners (e.g. Product) face — and learn from one another in the process. Regardless of what the solution is, we can solve it — and our partners trust us to do so.
We have three general Data Scientist archetypes: Analytics, Machine Learning, and Causal Inference. The responsibilities between these archetypes have some ambiguity and overlap — and that’s ok!
At the end of the day, we’re all problem solvers. In general:
- Analytics Data Scientists leverage their deep analytical skills and product intuition to solve a wide array of complex problems — from strategic analysis to dashboard creation and experimentation design and analysis. Recent projects include building experimentation metrics pipelines and dashboards (an incredible challenge, given the scale and complexity of our product!) and conducting deep-dive analysis into the impact of holiday events.
- Machine Learning Data Scientists apply their deep expertise in supervised and unsupervised learning to develop predictive models for internal usage. These models leverage state-of-the-art technology to balance performance and interpretability, providing insights, guiding strategy, and powering downstream analytics and modeling. Feature engineering often gets complicated due to the nonlinear (and long-tail) behaviors of our users! Recent ML projects include constructing a suite of calibrated models to understand the long-term value of subscribers as well as forecasting Discord’s growth for years to come.
- Causal Inference Data Scientists develop, use, and evangelize novel methodologies to understand the causal impact of product features and consumer offers through A/B testing and quasi-experimentation. One recent CI project used Bayesian Structural Time Series to understand the impact of a marketing campaign based on a recent geo test.
Team Structure & Partnership
Modern Data Science organizations are typically structured as:
- Centralized: data scientists operate as consultants throughout the company
- Embedded: data scientists directly report to business units
- Some hybrid approach
At Discord, we’ve decided to take a hybrid approach - combining the benefits of a centralized function with the strengths of individual sub-teams that are aligned 1:1 with major company pillars. In this way, we can progress towards a broad ambitious data vision for the company — including developing the right foundations, establishing best practices, and building a data community — while still providing deep direct support needed to drive the success of specific business and product initiatives.
The key areas we support include:
- Growth: Optimize the user experience, covering everything from streamlining registration flows to helping users find new servers and friends
- Revenue: Grow our premium subscription services, Nitro & Boosting, through development and continued improvement of high-value product features
- Marketing: Improve awareness of product features and consumer offers
- Communities, Channel Activities, and Bots/APIs: Build and maintain capability for communities on Discord to grow and flourish
- Safety: Help make Discord a safer place for all of our users
- Messaging & Experience: Improve the core A/V and messaging functionality that makes Discord an amazing platform for its users to find belonging
Across DS teams, we pride ourselves on hyper-collaborative partnerships. Oftentimes, Data Science teams operate within a company as service teams, brought in late in the product development lifecycle to conduct an analysis or build a dashboard or write a query. Data Science at Discord is a partner team, not a service team. We focus on proactive and strategic relationships with product managers, marketers, and engineering managers, partnering with these teams from ideation all the way to post-deployment. As part of that process, we focus on scaling (ourselves and our partners), building self-service tools, and strategic impact.
Data Science also couldn’t bring data to bear on solving problems without a close partnership with the Data Platform team. We work hand-in-hand with our engineering partners to identify and build infrastructure needs, novel capabilities, and high-leverage solutions to problems ranging from democratizing experimentation at scale to automated data quality checks.
Skills and structure provide a framework for success but a world-class Data Science organization also requires the right culture. All of our data scientist roles are deeply technical positions that interface directly with business and product staff. This powerful combination of technical and business acumen means our data scientists are in high demand and drive outsized impact. Teams want to work with us, and data scientists love what they do. Jason has written about our company culture here and these cultural tenets form a foundation for how we operate.
Data scientists at Discord seek to be hyper collaborative so that other teams proactively seek us out, knowing that we are invested in solutions. Everyone on the team is a leader — data scientists are empowered with autonomy and agency to proactively figure out solutions quickly and efficiently. Excellent judgement and strong communication skills are a must, and data scientists should know how to share ideas and progress with a variety of audiences, know when to collect feedback and know when to charge ahead.
We also recognize that none of us are perfect and value modesty and vulnerability. We want to grow as data scientists, professionals, and people and therefore invest heavily in direct, candid, and empathetic feedback — even in our interview process!
Data Science at Discord is quickly growing and evolving. In fact, we’re building out a sister Data Engineering organization starting in 2022! If you enjoy solving challenging problems using massive data sets and state-of-the-art technology, working with incredible partners at an amazing and unique company, and like what you read above — please consider joining us on our journey!
For more information on available roles in Data at Discord, be sure to check out our current job listings for the Data Science team here: https://discord.com/jobs?team=data-science.