AnnouncementPinecone serverless on AWS is now generally availableLearn more

Gong selected Pinecone as the core database infrastructure for Gong's innovative AI system, Smart Trackers. It plays a crucial role in storing and processing vector embeddings, enabling efficient tracking and classification of concepts within user conversations. Through Pinecone, Gong achieves efficient vector searches, empowering Smart Trackers to offer users precise and relevant examples for concept tracking in conversations.



Use Cases





vectors representing sentences


cost savings with Pinecone serverless

Gong is a revenue intelligence platform that transforms organizations by harnessing customer interactions to increase business efficiency, improve decision-making, and accelerate revenue growth. Gong's commitment to innovation led to developing proprietary artificial intelligence technology to enable teams to capture, understand, and act on all customer interactions in a single, integrated platform.

In early 2020, Gong was a trailblazer in adopting a vector database for semantic search. Spearheading this endeavor was Jacob Eckel, VP, R&D Division Manager, who has played a pivotal role in Gong's journey from its early days. From being the first Gong hire to now leading Gong's AI Platform R&D division, Jacob’s team is dedicated to supporting Gong's platform through cutting-edge AI technology.


Revolutionizing Conversational Analysis with AI

To transform revenue intelligence, Gong relies on effectively processing and analyzing the wealth of data collected from customer conversations. To perform a comprehensive analysis for these conversations, it is necessary to track and extract relevant concepts from sentences and capture the nuanced expressions in order to provide valuable insights and actionable intelligence for users. Gong’s market-leading insights and intelligence, based on billions of customer interactions, enable organizations to develop high-performing revenue teams that are more productive and effective through advanced coaching and deal execution.

Gong developed an innovative "active learning system" called Smart Trackers, a patented technology that leverages AI allowing it to detect and track complex concepts in conversations. Smart Trackers leverages AI algorithms and semantic search to dynamically analyze conversations. With Smart Trackers, Gong users contribute to the model creation process without the need for technical proficiency and specialized skills for algorithm building and data management. In this system:

  • Users provide examples of sentences that express a specific concept and classify them as relevant or not.
  • Each model represents a concept in the user conversation. For example, a sales manager wants to track when a competitor is mentioned during a sales call — now, the competitor becomes a concept.
  • The system retrieves many more examples from real conversations based on the user-provided examples.

Smart Trackers depend on real-world examples from conversations for continuous learning and adaptation from users' feedback. "Users want to track different concepts that occur in conversations, and simple keywords do not work,” shared Jacob. Using traditional keyword-based methods wouldn't be effective with this system because conversations are dynamic and rich with nuances, contextual variations, and subtle expressions.

Because of keyword search limitations, Gong needed a vector database capable of supporting the tracking and classification of concepts without being computationally heavy, delivering prompt results to users.


Leveraging Pinecone to Fine-Tune Smart Trackers

As Gong delved into the complex world of concept detection, Gong’s ML team recognized the need for a vector database solution that would support searches across a vast corpus of conversational data. While other options were available, including open-source alternatives, Gong selected Pinecone as their vector database partner to support the development of Smart Trackers.

The decision wasn't merely about addressing immediate needs; it was a forward-looking move. Gong saw Pinecone as a partner with the capability to lead in the market, aligning with their own aspirations for growth and technological advancement. Early in Pinecone's development, Gong recognized the potential for a lasting partnership that would go beyond a transactional vendor relationship. “We decided to be both a customer and design partner of Pinecone," said Jacob. The partnership with Pinecone not only provided the necessary infrastructure but also a trusted ally that understood Gong's use case.

Pinecone serves as the core database infrastructure for Gong’s technology underlying Smart Trackers, playing a crucial role in storing and processing vector embeddings for tracking and classifying concepts within user conversations. At the end of each user conversation, Gong processes each conversation and converts them into sentences. Gong anonymizes the data by embedding each sentence in a 768-size vector, along with metadata. The resulting information is encrypted and then stored in Pinecone. Using encryption and namespaces ensures a secure and efficient storage mechanism for the vast array of conversations processed by Gong.

Once sentences are embedded, Gong uses Pinecone for vector searches to identify sentences similar to the provided examples. Here is how the workflow works:

  • Smart Trackers Processing: Smart Trackers process the search results, analyzing and classifying sentences based on their relevance to the tracked concept.
  • User Presentation: The processed results are then presented to the user. This step allows users to track and monitor the identified instances of the specified concept in conversations.
  • Accurate Labeling: Using a vectorized representation of the concept enables the system to efficiently capture the nuances and contextual variations present in user conversations, facilitating the accurate labeling of each concept.
  • Fine-tuning: Pinecone helps fine-tune the system by identifying specific concepts that can be applied to all conversations or subsets of conversations chosen by the user.
High level Diagram: Classification Use Case

By integrating Pinecone into this process, Gong achieves efficient vector searches, enabling Gong’s Smart Trackers to provide users with precise and relevant examples for concept tracking in user conversations.

As the partnership with Pinecone evolved, the requirements to support Gong's needs also evolved. Gong already knew that Pinecone was the leading vector database in the market and wanted to continue exploring an alternative to its existing Pinecone architecture for enhanced efficiency. Even after comparing it to other vendors, they trusted Pinecone to address their evolving requirements. Gong's use case is one of the primary inspirations for the recently available Pinecone serverless, enabling Gong to seamlessly leverage vector storage at any scale while achieving substantial cost reductions.

“Our choice to work with Pinecone wasn’t just based on technology; it was rooted in their commitment to our success. They listened, understood, and delivered beyond our expectations.” - Jacob Eckel, VP, R&D Division Manager at Gong

Transition to Pinecone serverless results in 10x cost reduction while maintaining peak performance

Gong tested the recently available Pinecone serverless as a design partner during the development period. The transition to serverless wasn't just about cutting costs; it was a deliberate step towards achieving optimal performance and scalability. Pinecone’s approach to Gong's specific requirements prioritizes storage efficiency, allowing Smart Trackers to handle large-scale vector searches while maintaining adaptability to latency needs. With a focus on delivering optimal storage capabilities, Pinecone ensures that Gong can efficiently store and process the vast amount of vectors in their Smart Trackers without compromising performance. With Pinecone serverless, Gong has been able to:

  • Substantially reduce costs: The newly architected design of Pinecone serverless empowers Gong to leverage limitless storage and efficient computing capabilities specifically tailored to their workload, resulting in significant cost reduction. Since the transition, Gong's team has experienced a remarkable 10x reduction in costs.
  • Build at any scale: With Pinecone serverless, Gong can now store billions of vectors, representing a wealth of information extracted from diverse customer conversations. With this new architecture, Gong can substantially reduce cost at scale enabling precise searches with rich metadata.
  • Access easiest-to-use technology: The transition to a serverless architecture empowers Gong to focus on crafting cutting-edge tools and models to reshape their revenue intelligence strategy without the burdens of provisioning, sharding, and rebuilding indexes.
"Pinecone development of serverless showcases the power of a true strategic design partnership" - Jacob Eckel, VP, R&D Division Manager at Gong

A Vision for the Future

Gong has experienced substantial enhancements in their concept detection process since adopting Pinecone serverless:

  • Scale: Billions of vectors representing sentences in Pinecone
  • 10x cost savings with Pinecone serverless

The strategic partnership with Pinecone has laid the foundation for future innovation, and Gong is exploring new use cases. The team is also planning to increase the number of vectors and incorporate additional functionalities, ensuring Gong stays at the forefront of technological advancements in revenue intelligence. Jacob shares, "Pinecone has proven to be a valuable partner in advancing our technology. We are excited about future functionalities."

“Pinecone serverless isn't just a cost-cutting move for us; it is a strategic shift towards a more efficient, scalable, and resource-effective solution.” - Jacob Eckel, VP, R&D Division Manager at Gong

What will you build?

Upgrade your search or chatbots applications with just a few lines of code.