Rippling rebuilt its GTM data platform on Databricks using a medallion architecture, Lakeflow Spark Declarative Pipelines, Delta Lake and machine learning pipelines that continuously process first-party and third-party data.
At the center of the experience is GrowthOS, Rippling’s internal AI application that gives more than 2,800 GTM employees a single place to interact with trusted business data. Rather than navigating dashboards or writing SQL, employees can ask questions in natural language while AI agents access the same governed data through the Genie API.
Creating a trusted foundation for people and AI agents
Before those insights reach users, Rippling applies machine learning-based entity resolution to reconcile customer and prospect records across hundreds of millions of records from internal systems and third-party data providers. By resolving duplicate identities into a single trusted record with confidence scoring, the company gives both employees and AI agents a consistent foundation for analytics and automation.
“Genie became the conversational layer for our GTM AI platform, giving teams and AI agents access to trusted data through a single, governed interface,” said John.
Genie Agents now power natural-language analytics across a wide range of use cases within GrowthOS. Employees can use plain language to:
-
Analyze campaign performance
-
Understand customer sentiment
-
Investigate sales objections
-
Retrieve account intelligence
AI agents use the same governed interface to support automated workflows across the customer lifecycle.
Because Genie sits directly on top of governed data, Rippling can continuously improve prompts, semantic definitions and business logic through Genie while maintaining centralized governance across the organization.
To support those experiences, Rippling also built semantic search using Databricks AI Search. Rather than recomputing embeddings every time someone asks a question, the team created an AI-ready retrieval layer that enriches conversations during ingestion and serves grounded results through hybrid search. Their internal principle became DRY(E): Don’t Repeat Your Embeddings.
The result is faster retrieval, lower inference costs and more consistent answers for both employees and AI agents.