In early 2023, Bolt faced a critical scaling challenge: our rapidly growing merchant base threatened to overwhelm our support infrastructure. Rather than linearly scaling our support team, we implemented an AI-powered solution using DevRev, which deflected 60% of the ticket volume while improving resolution times by 30% and increasing customer satisfaction by 25%. This article details our technical implementation, architectural decisions, and the quantitative results achieved.
The challenge: Support at scale
Bolt operates a dual-sided marketplace where we serve two distinct user groups:
Merchants: Our primary customers who integrate Bolt's payment processing into their e-commerce platforms. These clients require white-glove support with rapid response times, technical expertise, and thoughtful solutions.
Shoppers: Consumers who use Bolt accounts for a faster and more secure checkout experience.
The traditional approach of hiring proportionally more support staff would have resulted in:
- Unsustainable cost scaling
- Degraded support quality while training new support team members
- Continued inability to provide personalized support experiences
Solution architecture: AI-Powered support segmentation
Technology stack selection
After evaluating multiple solutions, including Zendesk, Intercom, and Freshdesk, we selected DevRev based on four key differentiators:
- Native AI Integration: Built-in machine learning capabilities for automated ticket routing and resolution
- Audience-Based Customization: Ability to create distinct user personas with tailored experiences
- Flexible Authentication: Support for custom JSON Web Token-based user identification
- API-First Architecture: Seamless integration with our existing tech stack and future integrations with their snap-in technology.
System architecture
Our implementation follows a three-tier architecture:
Tier 1: Authentication & user context
Bolt Frontend → Bolt Server → User Authentication & Context Building
When users access our application, our frontend application makes an authenticated request to our backend services.
The Bolt server:
- Validates user identity using the authentication layer
- Builds user context, including account type
- Packages this data into a standardized payload for DevRev
Tier 2: DevRev integration & audience routing
Bolt Server → DevRev API → JWT Generation with Audience Tags
Our Bolt server communicates with DevRev's auth token API using our internal DevRev client:
- Sends user context payload to DevRev's authentication endpoint
- Receives back a JWT token containing audience classification
- Audience types: merchant, shopper, developer
Tier 3: Dynamic widget configuration & AI processing
DevRev Widget ← JWT Token → Audience-Specific Configuration
The DevRev widget dynamically configures itself based on the JWT audience claims:
- Question Sets: Merchants see technical integration questions; shoppers see account/payment questions
- Escalation Paths: Merchant issues can trigger immediate engineer paging; shopper issues follow standard queues
- Knowledge Base Access: Filtered articles relevant to user type
AI integration & knowledge management
Smart content delivery
DevRev's AI engine analyzes user queries in real-time and:
- Semantic Matching: Uses natural language processing to understand intent beyond keyword matching
- Audience Filtering: Only surfaces knowledge base articles tagged for the user's audience
- Dynamic Resolution: Provides step-by-step solutions rather than generic article links
Knowledge base organization
We restructured our knowledge base with audience-specific tagging:
Merchant-Focused content:
- "How to configure webhook endpoints for transaction callbacks"
- "Troubleshooting API authentication errors"
- "Setting up test vs. production environments"
Shopper-Focused content:
- "How to request a refund from a merchant"
- "Updating payment methods in your Bolt account"
- "Understanding Bolt's buyer protection policies"
Example: Refund query processing
When a user asks about refunds:
Merchant query: "How do I process a refund?"
- AI Response: Step-by-step instructions for using the Bolt merchant dashboard
- Includes how to perform refund operations from the dashboard or Bolt API
- Links to webhook configuration for refund notifications
Shopper query: "How do I get a refund?"
- AI Response: Explains the merchant-initiated refund process
- Provides template language for contacting merchants
- Links to dispute resolution if the merchant is unresponsive
Qualitative improvements
- Merchant feedback: Support responses are now personalized to the merchant’s implementation, technical in nature, and relevant to the query, not generic
- Engineering team: Reduction in support escalations requiring account management and engineer involvement
- Support team: Higher employee satisfaction resulting in the handling of more complex, meaningful work
Future enhancements
- Payload generation: Expansion to allow DevRev to parse the Bolt OpenAPI spec to generate merchant specific code samples.
- DevRev workflows for API actions: It is the goals that merchants can ask questions and we can modify merchant settings directly from the merchant widget
- Support order questions: Shoppers will be able to cancel, manage refunds through conversational AI
Conclusion
The integration of DevRev's AI-powered support platform transformed Bolt's customer service capabilities while avoiding unsustainable cost scaling. By implementing audience-based routing and intelligent automation, we achieved significant improvements in response times, resolution rates, and customer satisfaction.
Overall, we were able to create a highly reliable system with a near 100% success rate in detecting our support audience. The key to our success was recognizing that different user types require fundamentally different support experiences. Rather than applying a one-size-fits-all approach, we built a flexible system that adapts to user needs and context while maintaining the efficiency benefits of automation.
For organizations facing similar scaling challenges, we recommend:
- Prioritize User Segmentation: Identify your distinct user personas and their unique support needs
- Invest in Integration: Budget adequate time and resources for seamless system integration
- Measure Everything: Establish baseline metrics before implementation to quantify success
- Plan for Change: Prepare your team for new workflows and provide adequate training
The future of customer support lies in intelligent systems that combine the efficiency of automation with the personalization of human service. Our experience with DevRev demonstrates that this future is not only possible but immediately achievable with the right technology partnerships and implementation strategy.
Authored by a member of Bolt’s support team, based on their firsthand experience.
