> Markdown rendition of https://www.voiceflow.com/research ("AI Agent Research for Customer Experience | Voiceflow"). Canonical page: https://www.voiceflow.com/research · All pages: https://www.voiceflow.com/llms.txt # Applied research to advance agentic customer experiences Benchmarks, controlled experiments, and engineering insights for building reliable AI agents for customer experience. ## Research index 09 articles Topic All topics All topics Evaluation Retrieval & RAG Agent architecture No No categories found found. - ### [EvaluationOct 4, 2026Voiceflow vs Claude: where CX decisions go wrongIn 320 controlled CX trials, all four models made refund errors. Compare Voiceflow and Claude on policy decisions and exact amounts separately.Braden Ream](https://www.voiceflow.com/research/voiceflow-models-vs-claude-cx-policy-checks) - ### [Retrieval & RAGFeb 18, 2026Enhancing user experience: from NLU intents to RAG with embeddingsWe're excited to introduce intent recognition powered by Retrieval-Augmented Generation (RAG): enhancing user experience from NLU intents to RAG with embeddings.Sanasar Janjughazyan](https://www.voiceflow.com/stories/enhancing-user-experience-from-nlu-intents-to-rag-with-embeddings) - ### [EvaluationJul 9, 20245 tips to optimize your LLM intent classification promptsWe tested over 500 prompt variations across two datasets to see how to improve LLM intent classification.Denys Linkov](https://www.voiceflow.com/stories/5-tips-to-optimize-your-llm-intent-classification-prompts) - ### [EvaluationApr 21, 2024Benchmarking hybrid LLM classification systemsBenchmarking hybrid LLM classification systems: a deep dive into hallucinations, accuracy, costs, latencies, usability and context window analysis.Denys Linkov](https://www.voiceflow.com/stories/benchmarking-hybrid-llm-classification-systems) - ### [EvaluationMar 19, 2024How much do ChatGPT versions affect real-world performance?Do versions change your performance by 1% or 10%? Let’s dive into a production-impacting LLM task and how model versions impacts agent performance.Denys Linkov](https://www.voiceflow.com/stories/how-much-do-chatgpt-versions-affect-real-world-performance) - ### [Retrieval & RAGMar 19, 2024Improving performance of Hybrid Intent + RAG conversational AI agentsA case study on opinionated decisions when launching products, balancing speed to build, accuracy and speed to iterate.Denys Linkov](https://www.voiceflow.com/stories/improving-performance-of-hybrid-intent-rag-conversational-ai-agents) - ### [Agent architectureFeb 14, 2024Large Action Models change the way we build chatbots, againHow Large Action Models will change the way we build chatbots, again: applied research on Large Action Models and the way chatbots get built.Braden Ream](https://www.voiceflow.com/stories/large-action-models-change-the-way-we-build-chatbots-again) - ### [Agent architectureAug 30, 2023Creating the Voiceflow NLUWe started working on the Voiceflow NLU in February 2022, and this is a small slice of our story: the applied research behind creating the Voiceflow NLU.Denys Linkov](https://www.voiceflow.com/stories/creating-the-voiceflow-nlu) - ### [Agent architectureJun 13, 2022NLU design: How to train and use a natural language understanding modelLearn the basics of NLU models and how teams are training/using them for conversational AI experiences.Denys Linkov](https://www.voiceflow.com/stories/nlu-design-how-to-train-and-use-a-natural-language-understanding-model) No posts found Your search did not match any posts. Please try again. Clear search