Y Combinator Summer 2024 graduate Voker has launched an analytics platform designed to give companies visibility into how their AI agents perform in real conversations.

The platform automatically tracks three key metrics: user intents, corrections when agents fail to understand requests, and successful resolutions. Companies can integrate Voker through a Python SDK that instruments existing agent workflows.

"Do you really know what your agents are saying to your users?" the company asks on its launch page. The platform addresses what Voker calls the "flying blind" problem — where teams only discover agent failures through customer complaints or rising churn rates.

Voker's system classifies user goals from natural conversation without manual tagging. When a user says "Help me book my next vacation," the platform automatically categorizes this as a travel booking intent. It then tracks whether the agent successfully fulfills the request or requires user corrections.

The analytics dashboard shows metrics including total sessions, correction rates, resolution rates, and emerging intent categories. Teams can correlate conversational data with existing user metrics to measure business impact.

Self-service insights for non-technical teams

The platform targets companies with high interaction volumes — specifically those handling over 1,000 chat sessions monthly. Product managers, analysts, and business teams can access insights without engineering support.

Voker offers what it calls "AI-powered installation" — users paste a prompt into their preferred AI coding tool, which then scaffolds the SDK integration and configuration. The setup process takes approximately two minutes according to the company.

The platform aims to help teams prove AI agent ROI by connecting performance metrics to conversion, retention, and revenue outcomes. This addresses a common challenge where companies have usage statistics but struggle to quantify the business value of their AI investments.

Voker is targeting teams building "best-in-class agent products" with substantial user bases and complex conversational workflows that require systematic performance monitoring.