How Pinterest Analyzes 30,000 Employee Comments Securely with Posit Workbench & Snowflake

Employee engagement surveys provide a structured way to gather employee feedback on job satisfaction, culture, and management, which helps organizations improve performance, boost morale, and reduce turnover. 

Although employees commonly use the five-point Likert scale to indicate their level of agreement with statements in surveys, these often include comment sections or open-ended questions. This specific employee feedback, however, can pose a challenge for people analytics teams.

Trevor Fry, Lead Data Analyst, People Insights and Analytics Team at Pinterest highlighted that before adopting a centralized data science environment, they faced significant operational hurdles when accessing sensitive data housed in Snowflake.

• Slow, Multi-Step Data Access: Sourcing and preparing data required a “four or five-step process with multiple chunks of code”. This manual approach consumed time and made collaborating and sharing work with peers difficult.
• Need for Standardization: Analysts used a combination of basic tools like Tableau and Google Sheets. This resulted in outputs that “kind of all looked and felt a little different,” motivating the team to unify their approach and work by “marching to the same beat”.

Pinterest addressed these challenges by deploying the Posit Workbench Native App directly within their Snowflake environment.

• Seamless Snowflake connection: The Posit Workbench Native App eliminates connection friction by automatically managing Snowflake credentials. Users instantly inherit their assigned roles and permissions without writing boilerplate code or exposing credentials in code, making database access both seamless and secure.
• Familiar user experience: For Trevor and other analysts well-versed in R, the change was an “easy transition” because Posit Workbench provides the familiar RStudio interface “just in my browser”. The team was able to “hit the ground running” immediately.
• Multi-language flexibility: Posit Workbench offers Pinterest’s data scientists flexibility in their development environment, allowing them to use R or Python within their preferred tool, including RStudio, VS Code, JupyterLab, or Positron, depending on their specific needs.

Implementing Posit Workbench enhanced the daily work experience and effectiveness of the People Analytics team by significantly increasing productivity and enabling complex statistical modeling. With Posit Workbench integrated into Snowflake’s AI Data Cloud, Pinterest’s analysts now work in a secure, fully governed environment that accelerates exploration and advanced modeling.

Accelerated productivity and risk reduction

• Simplified Workflows: The direct connection reduced the multi-step process for data access and preparation down to “one code chunk”. This single chunk combines the initial SQL query (for aggregations) and the subsequent R analysis (using filters or tidyverse mutations). This speeds up work by reducing the necessary steps to source and pre-process data.
• Improved Collaboration: Having the entire process in one code chunk makes sharing analysis with peers or counterparts “more fluid”. The single environment also helped analysts who were previously relying on other tools to “adopt [R] a lot more”.

Enabling advanced statistical analysis

Posit Workbench is essential for conducting in-depth, regression-based analyses that cannot be performed efficiently using SQL alone.

• Driver Analysis: Trevor regularly uses R for regression-based analyses, including the Relative Weights Analysis (RWA) package, which efficiently identifies the key “drivers” of employee sentiment and engagement (e.g., commitment) from survey data. Identifying these root causes helps the business focus on specific intervention points.
• Sophisticated Modeling: The team uses Structural Equation Modeling (SEM) to look at how different factors relate, such as using mediated regression to understand how sentiment about management correlates with trends in sentiment about strategy. This allows them to tell a sophisticated story beyond just reporting the numbers.

Processing unstructured data

• Comment Analysis: The team handles a high volume of unstructured data, receiving 10,000 to 15,000 rich comments per employee survey. Posit Workbench is necessary because R scripts leverage Natural Language Processing (NLP) techniques to clean the data (e.g., handling spelling errors) and sort/subset comments before running thematic analysis with AI tools. Without this integrated connection, processing this volume of data would be a “roadblock for sure”.

The analyses generated by People Insights and Analytics enable leadership teams at Pinterest to make data-driven decisions about focus areas for the company and subsequent action plans for each of those focus areas.

No Ops headache: A “set it and forget it” experience for IT

For the Enterprise Data Platform team at Pinterest, enabling the People Analytics team couldn’t come at the cost of increased operational complexity. Traditionally, hosting data science environments involves provisioning Linux servers, managing complex networking for secure data access, and scheduling downtime for manual patches and upgrades.

By deploying the Posit Workbench Native App on Snowpark Container Services (SPCS), Pinterest eliminated this maintenance burden entirely.

Because the application runs natively within Snowflake’s infrastructure, the platform team realized immediate operational benefits:

• Zero Infrastructure Management: There were no external servers to configure or secure. The compute runs directly within the Snowflake security boundary, adhering to Pinterest’s strict governance standards automatically.
• Automatic Upgrades: The Native App model ensures that the data science team always has access to the latest features and security patches. Upgrades happen automatically in the background, removing the need for the admin team to plan and execute manual update cycles.

This “hands-off” administration allowed the platform team to focus on strategic data initiatives rather than routine maintenance, while simultaneously delivering a state-of-the-art environment to their end users.