Gong developed "Deal Me", a natural language question-answering feature for sales conversations that allows users to query vast amounts of sales interaction data. The system processes thousands of emails and calls per deal, providing quick responses within 5 seconds. After initial deployment, they discovered that 70% of user queries matched existing structured features, leading to a hybrid approach combining direct LLM-based QA with guided navigation to pre-computed insights.
# Implementing LLM-Powered Deal Analysis at Gong
## Company Background
Gong is a platform focused on sales team optimization, initially targeting the challenge that sales representatives spend 80% of their time on non-selling activities. The platform aggregates and analyzes all deal-related information, including calls, emails, and other interactions, providing insights to different organizational levels from sales representatives to top management.
## The Deal Me Feature
### Initial Challenge
- B2B sales deals involve complex data:
### Technical Implementation Journey
- **Data Integration**
- **Context Management**
- **Prompt Engineering**
- **Response Generation**
- **Source Attribution**
## Production Challenges and Solutions
### Cost Optimization
- Faced significant costs due to:
- Implemented cost reduction strategies:
### Quality Control
- Developed comprehensive QA testing framework
- Implemented continuous testing for new model versions
- Created mechanism to validate responses against source data
- Built system to handle model updates and maintain prompt effectiveness
### User Behavior Analysis
- Post-launch learnings:
## System Evolution and Optimization
### Hybrid Approach
- Developed intelligent routing system:
- Benefits:
### Infrastructure Requirements
- Built robust data collection and processing pipeline
- Implemented rapid deployment capabilities
- Created monitoring and feedback systems
- Established testing frameworks for continuous improvement
## Key Learnings
- Strong infrastructure is crucial for AI product success
- Quick deployment and user feedback are essential
- Real user behavior often differs from assumptions
- Hybrid approaches combining structured and AI features can be more effective
- Continuous monitoring and optimization are necessary
- Cost management is crucial at scale
- Model updates require robust testing and adaptation mechanisms
## Results and Impact
- Extremely positive user feedback
- Significant social media engagement and market interest
- Improved user engagement with platform features
- More efficient use of existing platform capabilities
- Better cost management through hybrid approach
- Enhanced user discovery of platform features
Start your new ML Project today with ZenML Pro
Join 1,000s of members already deploying models with ZenML.
