Berlin-based Parloa lets enterprises build AI customer service agents that actually work. After co-founder Stefan Ostwald spent a day listening to insurance agents repeat the same password resets and policy questions, he realized automation was the answer. The company now uses OpenAI models including GPT-5.4 to power its Agent Management Platform, handling millions of conversations across retail, travel, and insurance.

AMP lets business users define agent behavior in plain language instead of mapping rigid intent trees. Teams connect to internal systems, simulate customer conversations with models playing caller and agent, then evaluate performance before deployment. One global travel company reduced requests for human agents by 80% using the platform.

Parloa prioritizes real-world performance over abstract benchmarks. They test instruction-following and latency against realistic scenarios before any model reaches production. For voice interactions, every millisecond counts—the speech-to-text, reasoning, and text-to-speech pipeline must feel natural. The company works directly with OpenAI optimizing models for real-time conversations where pauses break the illusion.

Customer service is becoming multimodal, with conversations starting on phone and continuing in chat with interactive elements. AMP handles this as a single interaction rather than separate flows, using modular sub-agents for complex tasks like authentication while maintaining deterministic controls for critical operations.