OpenAI Improves Customer Support Process Through Every Interaction

OpenAI is moving beyond simple ticket-handling-focused support to an AI operating model in which every interaction improves the next support session. This aims for a structure in which the support system continuously learns and evolves, going beyond just introducing a chatbot.

This model consists of three core elements:

• Surfaces: The touchpoints where interactions occur, including chat, email, phone, and help embedded within the product.
• Knowledge: A living knowledge base that is continuously updated based on actual conversations and policies.
• Evals and classifiers: Tools through which software and people collaborate to define quality and measure feedback.

These elements form an interconnected loop. Patterns discovered in conversations with enterprise customers are reflected in developer FAQs, and Evals written for a specific case reinforce model performance across thousands of other cases. Because the same underlying Primitives are used across every touchpoint, improvements are automatically applied across all channels.

The role of support agents is also changing. Beyond simply responding, they are becoming Systems Thinkers who contribute to the system architecture — when they discover a new pattern, they propose Classifiers or prototype automation to fill workflow gaps.

This innovation is made possible by OpenAI's own technology stack:

• Agents SDK: Provides step-by-step tracing and observability.
• Responses API: Implements classifiers that judge tone, accuracy, and policy compliance.
• Realtime API: Enables voice support.
• Evals dashboard: Measures and visualizes quality.

OpenAI started with a simple Q&A answering tool and is now expanding, through the Agents SDK, to the stage of performing dynamic actions such as refunds, invoice processing, and outage lookups.