As a medium-sized company, Twitter only has one science job family: Data Scientist. As a comparison, Amazon has several science job families: data scientist, research scientist, applied scientist, economist, and etc. As a result, the DS role in Twitter is very broad, including responsibilities in the following three big areas.
To help the Data Scientists grow, teams usually take a portfolio approach during roadmapping, to ensure the planned work is balanced across all three areas.
(Some additional info about Twitter DS org. Twitter is organized in a Matrix structure, similar to many high tech companies, different from companies like Amazon. A Data Scientist will report to a DS manager and sit within a DS org. At the same time, the DS will be imbedded within a product area, i.e. search. They will work closely with PMs or Engineers in that particular area, and will be evaluated by them too. In short, the DS org provides management support, while the product area forms a stable virtual team which allow all functions to work together. This structure has its pros and cons, which is beyond this post.)
Metrics & Measurement
This is the foundational data work that helps us understand how a product is performing. Typical projects include core dashboards, measurement plans, refining measurement strategy/methodology, or forecasting future growth. In big companies, DS work might be heavily focused on this area.
Typically required skill sets include data manipulation and visualization, experimentation, business analytics.
Model Improvement
This type of work is generally required when supporting products with ML models in production, i.e. search. Typical projects include analyzing model performance, diagnosing issues/gaps, or proposing and testing new features both offline and online. In some companies, Applied Scientist or ML scientist might work in this area.
Required skillsets include: a deep understanding of ML products, and familiarity with the production environment, etc.
Research & insights
This is the most well-known and most diverse area for data scientists. It includes answering research questions related to customer behavior or business development, generating insights that drive business strategies and decisions, and producing reusable frameworks to solve business problems at scale.
Required kill sets here are very broad, some examples are: using causal analysis to understand business impact, using supervised/unsupervised learning to detect Customer behavior patterns, or using simulation methods to optimize business outcomes.
