Gong Utilizes Mona to Monitor Their AI System for Performance Optimization

Gong unlocks reality to help people and companies reach their full potential. The Reality Platform™ autonomously empowers customer-facing teams to take advantage of their most valuable assets – customer interactions, which the Gong platform captures and analyzes. Gong then delivers insights at scale, enabling revenue and go-to-market teams to determine the best actions for repeatable winning outcomes. Since late 2019, Gong has been utilizing Mona’s intelligent monitoring solution to monitor its platform for performance optimization.

In 2015, Gong was founded to bring visibility to sales and go-to-market processes and improve them over time. Using AI to help solve this problem, Gong automatically records, transcribes, and analyzes customer calls in order to provide teams with insights and recommendations.

Gong is a customer-centric company and its mission is to unlock reality to help people and companies reach their full potential. Using AI within their processes, Gong has multiple teams of dedicated professionals to manage the ML pipeline. Noam Lotner is the Research Operations Team Lead where he manages a team that handles the operational layer of creating and maintaining Gong’s AI models. Specifically, Gong is based on the automatic analysis of conversations – and other points of customer interaction – which requires processing using ASR (automatic speech recognition) and NLP (natural language processing) models.

After a call is processed, several AI models are running simultaneously which need to be monitored for accurate results. Working closely with the research team, Yaniv Levi, Engineering Team Lead at Gong, is responsible for managing the processor, running the AI models, and monitoring the entire process.

Prior to implementing a monitoring solution, the team at Gong relied on manual processes and customer feedback to know if models were underperforming. Each AI model would undergo a QA process but this was not feasible at scale. Nor was relying on customer support tickets.

Noam and his team are responsible for the full life cycle of models ranging from voice identification, language detection, transcription, NLP, and more. When models are in production, the team has to ensure that the models are continuously providing adequate results and detect when issues arise given changes within the data, such as new customers in separate regions, changes in communication channels, platform shifts and many others. Not only are there changes once new data is ingested, there are also external factors that can impact the data and performance of the models.

The team at Gong needed a monitoring solution to accommodate two distinct use cases. They decided to move forward with Mona’s monitoring platform due to its flexibility and customization. Gong’s main use case for monitoring is conversation intelligence, running on either voice or video inputs, in which they have various models running at the same time for language detection, speech recognition, automatic transcription and more. Mona is able to monitor all of these models within the same context, allowing Gong’s team to identify correlations among the behaviors of these models.

When it comes to Gong’s AI models, it is really important to maintain the performance of the model in order to accurately classify the parts of the conversation to the different speakers. Specifically, the model is able to securely identify the sales representative and the prospect on a call so that the team can use the data to formulate insights into the conversation. Once there is a low confidence score for a model, Mona proactively alerts Gong’s relevant team members and also provides possible correlating factors, which helps to quickly find the cause for the low prediction accuracy.

The research team uses Mona to monitor each of their model behaviors for the purpose of continuously learning from production data. They use Mona to extract segmented data, such as specific language, customers, and platforms, to understand how the model performs. Noam, the Research Operations Team Lead, monitors model behavior and resource usage across the different data segments and time periods in order to compare them with experimental models. By having complete visibility into the models, the research team is able to continuously optimize on current models and create new ones with a better understanding of how they will actually perform in production.

Yaniv is an Engineering Team Lead at Gong. On a daily basis, Yaniv is looking at Mona’s dashboard to ensure their AI models are performing as they should. Mona’s dashboard provides an overview of Gong’s entire AI-driven product, automatically displaying any anomalies and insights that need to be addressed. Once Mona sends an alert about an anomaly, Yaniv will investigate the anomaly in Mona’s platform and is provided with possible explanations for resolving the issue.

Mona’s flexibility is crucial to Gong’s research team. Each team member is responsible for different AI models, and needs to track different metrics, the ones that matter most to her/his models. Being able to add custom fields and automatically calculate custom metrics is a key aspect of the flexibility that the Gong team enjoys. Noam Lotner says that the metrics he tracks with Mona include prediction scores, success rates, score deltas, and confidence intervals. He gets instantly alerted if there are any changes in those metrics over time in any given segment of the data. Additionally, Noam is able to compare metric values and track changes across time periods. Mona provides full visibility into the performance of the models for each of Gong’s customers.

As he became more familiar with Mona’s capabilities, Yaniv was able to improve his model release process. Yaniv creates “shadow deployments” by running all of the call processing in experimental mode in order to evaluate new tools or model versions using the same datasets that are currently running in production. In other words, Mona enables Gong’s ML engineering team to test new infrastructure and new model versions to get granular reports on specific segments where they underperform, compared to the version currently in production. By comparing the experimental execution data with the production data, Yaniv gets a clear picture of all the metrics that are being sent to Mona, how they behave in the experimental tests that are being conducted, and how they will behave in production. This provides Yaniv with a granular understanding of the model performance before it is released into production, which would otherwise be impossible to do without Mona.

Now that Mona has been implemented at Gong, the team is able to optimize performance of their data and models. As external factors can impact the performance of Gong’s AI-driven platform, the team is confident that they will be able to minimize performance degradation by using Mona’s automatic insights and alerts. By being able to monitor all of their models within the same platform, the team at Gong is confident that their experimental models will perform as expected once they are released into production. Mona provides the team with peace of mind that their models are working properly and the performance is not degrading over time.