Retell AI Inc., an artificial intelligence voice platform, today announced the addition of a quality assurance solution for voice AI named Retell Assure that monitors calls, surfaces issues and reduces the need for human spot checks.

The company positions itself as a third-generation voice AI platform designed to replace traditional call center operations by handling both inbound customer support and outbound sales calls. Despite launching in February 2024, Retell has already reached profitability and is experiencing what it says is rapid growth. The company raised $4.6 million in seed funding in August 2024, led by Alt Capital and Y Combinator.

Evie Wang, co-founder and chief marketing officer of Retell, told SiliconANGLE in an interview that as companies deploy AI at large scale, many are still relying on human reviewers to audit AI performance. This approach puts them on a path that cannot scale readily with call volume.

“One of their pain points is they still use humans to monitor the performance of AI voice agents,” Wang said. “They use Excel sheets and ask a group of 20 human agents to listen to AI calls and add columns to see if there are issues.”

Human oversight can capture nuance, but it cannot cover 100% of incoming calls or reliably surface operational difficulties across large volumes of AI-led conversations. As a result, enterprises struggle not only with identifying failures, but also with extracting actionable insights from thousands of interactions.

Retell Assure uses multiple models to analyze every call against customizable criteria, including latency, interruptions, hallucinations and even customer sentiment. The system automatically flags failures, assigns scores and specific reasons behind performance issues.

This provides a curated list of calls that humans can choose to listen into in order to understand the context of the interaction and determine how to correct AI behavior. Alongside scoring and indicating why the call “failed,” the system also provides recommendations on how to remediate the underlying problem.

According to Wang, most AI failures aren’t the result of model limitations, but poor configuration or incomplete knowledge guidance. Retell provides the QA system as a continuous training loop: identifying failures early, correcting them and systematically reducing error rates over time.

“Think of the voice AI agent as a human; humans make mistakes if they don’t have enough knowledge,” Wang said. “The best way to resolve the issue is to find ways to know that and teach it.”

Beyond individual call reviews, information technology teams and engineers can monitor aggregate performance through a dashboard that tracks trends such as average call duration, resolution rates and failure patterns over time. The reporting highlights common customer questions and recurring issues, while offering recommendations for improving system prompts, messaging and agent behavior.

“Voice AI agents are very powerful, but they indeed have some issues,” Wang said. “Most of the issues we see are around configuration and knowledge.”

Retell is already working with a mix of large enterprises and midmarket customers. One public case study highlights Switch Energy Inc., a Canadian electric vehicle charging company using voice AI for customer service.

Switch reduced overall costs by 50% across over 8,000 calls per month. Additionally, Retell’s platform brought in a system that could answer calls in about five seconds, in contrast with prior multi-minute hold times.

Retell offers its voice AI automation features through a usage-based pricing model, with the new QA platform available as an add-on. The company expects the feature set to roll out broadly on Jan. 1, 2026.

Images: SiliconANGLE/Microsoft Designer, Retell AI