AI Support at Vercel: Cutting Manual Tickets by 31%
McKinsey's latest AI survey reports that 65% of organizations now regularly use AI—nearly double the rate from just ten months ago—with many applying the technology to customer support. Vercel's own AI support agent has reduced human-handled tickets by 31%, a result achieved while maintaining support standards across a growing customer base.
The Support Bottleneck
Support teams contend with a steady stream of repeatable inquiries: guiding customers to documentation, helping with domain configuration, and answering common setup questions. At the same time, customer expectations for instant, 24/7 assistance continue to rise, pushing organizations toward AI-driven chat interfaces. Gartner predicts that by 2027, chatbots will be the primary customer service channel for roughly a quarter of organizations.
The opportunity lies in letting AI handle the high-volume, repetitive work while human support engineers focus on complex, context-rich cases that require genuine expertise. Done thoughtfully, AI complements rather than replaces human judgment in support operations.
Building the AI Agent
Vercel replaced its support form with a single open-ended prompt: "How can we help?" The AI agent, built with implementation partner Markprompt, draws on a multi-source knowledge base comprising website content, documentation, Next.js examples, and help articles. Powered by a large language model (LLM), the agent provides immediate answers, but it doesn't stop there—it can engage in follow-up questions and handle progressively complex queries. When the AI cannot fully resolve an issue, it pre-fills a support form before handing off to a human, eliminating repetitive data entry for the customer.
Implementation Principles
Vercel's rollout followed several deliberate decisions that shaped the final system.
Iterate Rather Than Rush
The first version focused on delivering objective, foundational answers with additional context available on request. This incremental start prioritized customer experience from the outset rather than aiming for a perfect system on day one.
Test Internally First
The AI assistant was exercised internally before customer exposure. This practice addressed compliance and security concerns—particularly around sensitive customer data—and ensured the tool met Vercel's own quality bar for customer interactions.
A/B Test Before Full Rollout
Vercel began with a 50% A/B test using Edge Middleware to validate the hypothesis in production. The first iteration delivered a 16% ticket reduction over two weeks. That result justified the move to a 100% rollout and evolution into a richer, conversational interface.
Keep Customers in Control
A persistent "Create Case" option allowed customers to abandon the AI experience and connect with a human support engineer at any step. This safeguard protected the brand, gave customers agency, and produced a feedback loop enabling rapid iteration.
Results Beyond Ticket Reduction
The AI agent successfully interpreted, answered, and resolved 31% of incoming tickets without human involvement. Customers also benefited from writing in any language, as the system is language-agnostic. Additional gains included:
- 42% reduction in first response time
- Higher job satisfaction reported by customer support engineers
- Improvements to documentation and product interfaces
The system created a data flywheel: analyzing resolved tickets revealed gaps in documentation, product interface, and the product itself, while every interaction improved the AI's future performance.
Getting Started with AI Support
Vercel's experience points to a pragmatic path for organizations exploring AI-driven support. Recommended starting points include:
- Explore AI applications: Build chatbots, writing tools, or virtual assistants with Next.js and the Vercel AI SDK.
- Prototype and collaborate: Use Vercel's DX Platform with feature flags and preview URLs for low-risk experimentation.
- Use quickstart templates: Deploy pre-built AI applications, including a chatbot template, in seconds.
- Launch at scale: Managed infrastructure handles scaling while maintaining performance.
Vercel's results align with the broader McKinsey findings, where organizations report both cost reductions and revenue increases from generative AI deployments. The company's experience suggests that customer support is a particularly promising starting point for AI adoption—provided the implementation keeps human oversight and customer choice front and center.
